April 28, 2026 • 61 mins
In this episode of Dynamics Corner, Kris and Brad are joined by Volodymyr Dvernytskyi — a BC developer from Ukraine, a recent Microsoft MVP, and the creator of many open-source tools — for a grounded, practical look at what AI-assisted AL development actually looks like in the real world. Volodymyr doesn't sugarcoat it: AI has made code cheaper to produce, but that speed comes with a trap that most developers aren't talking about yet. What happens when you generate thousands of lines of code you didn't write and don't fully understand? And who's responsible when that code breaks in production? The conversation takes an unexpected turn when Volodymyr reveals what's been happening to his open-source repositories — a flood of AI-generated pull requests from contributors who don't understand the problems they're trying to solve. But the episode isn't all cautionary tales. Volodymyr lays out a clear progression for anyone looking to get started: from IDE-based agents to CLI tools to MCPs, with a critical reminder that understanding Business Central's functionality matters now more than ever. Brad shares a testing experiment that produced eye-opening results on an old production project, and the group lands on a closing insight that reframes the entire "what do we do with the free time" debate. If you think AI development means less work, this episode will change your mind — it means better work, and the developers who understand that distinction are the ones who'll thrive.
#MSDyn365BC #BusinessCentral #BC #DynamicsCorner
Follow Kris and Brad for more content:
https://matalino.io/bio
https://bprendergast.bio.link/
Listen
Watch
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
SPEAKER_04 (00:00):
Task worker.
It's like a library fordevelopers where you can able to
uh easily to split some uh hugeslow processes into chunks and
run it in separate sessions.
Because I encounter this problemvery often in my experience.
SPEAKER_03 (00:16):
Welcome everyone to another episode of Dynamics
Corner.
What does Two Cats have anythingto do with AI, workflows,
framework, etc.?
We're about to find out.
I'm your co-host, Chris.
SPEAKER_01 (00:30):
And this is Brad.
This episode is recorded onApril 10th, 2026.
Chris, Chris, Chris.
Two cats in AL development.
That's an interesting concept.
Also, we learned about tips andtricks for starting your AI AL
development journey, as well aswhat you can do with your free
time now that we're using inliving in the world of AI.
(00:52):
With us today, we had theopportunity to speak with you
doing.
SPEAKER_03 (01:17):
It's morning for me.
It's eight o'clock in themorning right now.
SPEAKER_04 (01:20):
Oh, it's your also morning.
SPEAKER_03 (01:22):
Yeah.
unknown (01:22):
I see.
SPEAKER_04 (01:23):
It's uh it's 6 p.m.
for me, so it's for me, it'sevening.
SPEAKER_01 (01:27):
Oh dinner, dinner time.
Your day's over, it's good.
I like uh I like the the catsthat you have joining you as
well.
SPEAKER_04 (01:36):
Um it's not planet.
Immediately when I uh join uhmeeting during our session, they
just decide to be here.
I think they would like to.
SPEAKER_01 (01:48):
They wanted to join with you.
What are their names?
SPEAKER_04 (01:50):
So uh white cat is uh he is uh Hades.
SPEAKER_01 (01:55):
Say that again, please.
SPEAKER_04 (01:56):
Hades.
It's like uh yeah, Hades, Hades.
From the Greek uh Yeah, fromGreek mythology, it's actually
yeah.
And she is Nyx.
SPEAKER_02 (02:06):
What is that again?
SPEAKER_04 (02:07):
Nix Nix, it's also from Greece uh mythology as
well.
So that's like some interestinguh name choice, but uh I I love
them, they I like them, and uhthey're like very new for me.
Uh only two years ago.
SPEAKER_01 (02:24):
I I I I it's it's nice, it's it's good to have a
companion uh with you whenyou're you're sitting at home
working, and it's it's been awild, crazy couple of months
with uh with technology, withdevelopment, with AL.
(02:45):
Business Center released Wave 1for 2026, just a short period
ago.
Usually, and I have to figureout why, usually I go through
all of the features well beforethe release and experiment and
play with them.
But I think with the way thingsare going, I've been extremely
(03:07):
busy and and I was doing somereflection and figuring out why
am I so busy.
And I think part of it isbecause in the age of AI and
development, we can do so muchmore, so we're busier doing
more, whereas before we wouldtake the time to do things.
And I was looking at the list ofdevelopment editions for 2026
(03:35):
Wave One, and I'm curious toknow if you had the chance to
work with them because I becauseI just started working with them
recently.
Uh some of them I'm reallyexcited about.
SPEAKER_04 (03:49):
I actually take um short look on it.
Uh I also was not able to handleit uh fully, and uh it's like it
was a lot of uh new changesregarding agents, if you notice.
Uh like now you can uh buildsome agents in uh your sandboxes
(04:10):
environments, and uh after thatyou can export it and use it in
code.
Uh in theory, it's just uhperfect idea.
I like it.
Uh there's many differentchanges uh regarding um how uh
you can use unified interfacesto work with IA inside business
(04:30):
central, like um those specificum code units interfaces.
And I even use it, I I I try touse them.
But some of some of my part ofexperience was great, some of
it's not great, like there aresome limitations, like you
cannot uh pass um files and uhwork with them efficiently, like
(04:52):
it's it's still in uh in thestage where it's evolving, and
there are some challenges whichwe are uh right now trying to uh
to solve and Microsoft trying tosolve.
SPEAKER_01 (05:05):
So yeah, it's yes it it is, it's exciting.
So it's the trouble I haven'tplayed with it.
I I've looked at some chats, butthe enabled troubleshooting MCP
server for AL, I really want toplay with.
I I know that's due in May, soI'm I'm looking forward to
seeing that.
The AL tests from Visual StudioCode, I'm excited about that um
(05:30):
should be out now.
And also, as you had mentioned,BC Bench or uh agent preview or
agent playground to be able tocreate agents in the sandbox.
I've created many of those andhave been running the tasks
through the sandbox userinterface.
I really do hope that they makeit to where you can create the
(05:53):
agents and not have to createthe code for the agents with
what you create with thepreview.
SPEAKER_04 (05:57):
But it feels it feels so natural to just not to
quote it, but just to um likeconfigure it and use it.
I totally agree on that.
SPEAKER_01 (06:04):
Yes, I I've created many and you go back and the
task run, and I just feel like Ishould be able to click schedule
task or run task or dosomething, and uh that's my
hopes that one day we get therebecause as you had mentioned, it
feels so natural, and you can dosome uh great things.
I've had some demonstrationsthat I've done with it, and
that's what everybody's sayingis why do we have to code it
(06:26):
now?
It's already there.
SPEAKER_03 (06:28):
Uh maybe, maybe it won't take too long.
Maybe uh give it a few months.
SPEAKER_04 (06:33):
Yeah, it just it's just very fresh.
Or maybe we can contributesomehow.
Who knows?
Maybe, but still in future.
SPEAKER_03 (06:44):
That's what they want though.
That's what they want, right?
They want you to uh contributeand enhance it.
Here's the tool.
Make something right.
SPEAKER_01 (06:55):
So so those are some of the big ones that we're
excited about, but uh wedefinitely wanted to speak with
you about a lot of things frombusiness central uh with what
you're doing from the technicalpoint of view, because uh I know
you're doing quite a bit indevelopment and development
workflow.
Uh, but before we talk aboutthat, can you tell us a little
bit about yourself?
SPEAKER_04 (07:15):
Uh sure.
So um I'm Volodymyr, uh ABCdeveloper from Ukraine, a recent
Microsoft MVP, uh, and I havebeen in the ecosystem for nine
years.
Um I have worked with BC indifferent capacities.
I work at B Dynamics with Swisspartner, Microsoft Partner,
where I focus on enterpriselevel projects.
(07:35):
Additionally, I run companyVolodiner LC where I do
freelance work for smallmedium-sized clients and publish
my own app source extensions.
I spent also a good chunk of mytime contributing on open source
and writing a technical blog atuh vldbc.com.
Uh what I really focus on ismore developer tooling, building
(07:56):
things that make everyday lifeeasier for BC developers, uh
partners, customers.
Um my open source uh product,data unitor tool, for example,
uh, has over 160 stars on GitHuband almost 1000 installs on App
Source, and uncounted number ofinstalls on PT.
I don't know, I didn't count it.
(08:18):
And I enjoy Business Central.
I do all of this because Ibelieve in BC as a product and
its community.
I think open source helpseveryone grow, and I love
sharing what I learned, whetherthat's thoughts on my blog, at
events, or just in conversationslike this one.
SPEAKER_01 (08:34):
That's great.
So you you have the open sourceproject that, as you mentioned,
has a number of stars and a lotof installs.
And I'd like to talk about opensource projects, but some of
these tools that you've createdfor development, how are they
impacting the developmentworkflow and the benefits to
(08:55):
users of the tool when theirdevelopment process?
SPEAKER_04 (09:00):
Great question.
Uh I think uh the idea behind itis to uh fill all these gaps
what Microsoft are uh havebecause they are have many many
focuses, like IA, uh all ofthese important uh tools as
well.
But there are some gaps which Ican see uh that I can add some
(09:24):
value and add some value tocommunity to to create something
that can be useful specificallyuh to solve some specific
problem.
For example, data you tooltrying to solve problems, uh how
you can manage, edit, view, datamostly for testing uh in
environments, as you remember inold days uh you were able to do
(09:45):
it in Object Designer, uh inNavision Classic or RTC, doesn't
matter, or even a SQL.
Uh but after BC14, if I remembercorrectly, they change it and
you cannot manage data.
Uh same story.
Uh recently I released umparallel task worker.
(10:05):
It's like a library fordevelopers uh where you can able
to uh easily to split some uhhuge slow processes into chunks
and run it in separate sessions.
Because I encounter this problemvery often in my experience that
some huge processes which youcannot make it faster, it's like
impossible.
There is no uh room for that.
(10:26):
So I think uh idea behind it isto find some problem and trying
to solve it.
And if you don't see anysolutions in um in the internet
and community, then it's evenbest uh or greater uh idea to
put some effort on it and tomake life uh for developers and
partners easier.
(10:48):
That's what I think about it.
SPEAKER_01 (10:50):
No, no, it's important to see those tools.
And you had mentioned, I knowwith Business Central, you can
now, and you have been for manyyears, able to make
contributions to the applicationto enhance the features and the
functionality of theapplication.
And then also, as you hadmentioned, many individuals are
making open source contributionsto help with uh PTEs uh or for
(11:15):
the application or even somedevelopment tools.
I know a lot of individuals areadding development tools um uh
uh options available.
The cats are going a littlecrazy there.
It's kind of why I had a briefpause.
It's fighting, fighting.
Yeah, I thought there was goingto be a battle, a battle soon
with uh with the cats.
SPEAKER_04 (11:34):
They started when you mentioned open source uh
contributions.
Interesting.
Why?
SPEAKER_01 (11:40):
They're excited about the open source
contributions and well withthat.
Uh we appreciate all that you'readding, and you have a lot of
great tools available on your uhGitHub repository.
But another thing that we weretalking about previously is the
impact of AI in development ingeneral, and in general also
(12:05):
with AL.
So you've been working withdevelopment for nine years, so
you've come into the tail end, Ibelieve, of um the old way, as
everybody talks about with Cal,transitioned into AL, working
with AL.
And now, as we talked about,Microsoft's even adding
development tools, as we talkedabout the introduction to this
(12:26):
session.
How has AI impacted your ALdevelopment, or how do you see
it impacting AL developers?
SPEAKER_04 (12:36):
That's a very uh great question.
Let me think about it more.
Um first of all, I would like toadmit you mentioned very
correctly there is not only opensource projects which you
maintain by your own, there'salso contributions to uh direct
contributions to businesscentral uh itself.
You can uh contribute uh as opensource contribution uh to system
(12:58):
application, to uh baseapplication also as well.
And uh I would like to like tohighlight it to all developers
uh who listen it, or maybe willlisten it, don't afraid to
contribute to business centralitself.
It's actually add a lot ofvalue, it's uh you solve your
problem, your community problem,and product itself will gain
(13:20):
some value from it.
So don't afraid don't afraid todo it.
Start with something simple.
And uh back to your question uhabout the effect of uh how
artificial intelligence canaffect uh development process
specifically for AI and ingeneral.
Um I will start with uh somevaluable uh insights.
(13:44):
What I found out when I uh startto um realize that I is closer
and closer uh in developmentprocess.
I start to use uh IA firstly, uhI think when uh GPT 3.5 was
released.
Uh it is 2020, 21.
SPEAKER_01 (14:02):
Not I don't remember exactly the year, but it goes so
fast I can't keep up with iteither.
It seems like yesterday a lot oftimes that something was
released and someone will tellme it was months ago or years
ago.
SPEAKER_04 (14:17):
I totally agree, it's evolving very fast every
month, every six months, everyuh place or situation is
changing dramatically.
That's what's happening.
Uh and today, if you speak abouttoday, uh what I actually see
that uh IA is to make code ischeaper, you can generate and
(14:39):
create many uh many products uhwithout spending much amount of
time uh for writing all specificline of code.
You can spend more amount oftime uh to focus on architecture
of product, how you can buildthings, you can do actually real
(15:02):
software engineering, because ifyou think about software
engineering, it's not just aboutuh uh write some lines in text
in on in in in notebook, it'sactually to design, to think
about architecture, to thinkabout limitations, to think how
it will be maintained in thefuture.
And I think II give us a huge uhtool to avoid some um boring
(15:30):
work which are not superactually software engineering if
you think it about more.
So I think it gives you anopportunity to generate a lot of
code, but at the same time, Ithink uh it's also a dangerous
um because when you willgenerate too much line of code,
(15:50):
how you will understand themvery well.
Someone will tell you don't needto understand it because you can
just uh delegate all work toagent and it will do everything
for you.
Maybe, but we are not on thisstage where we can avoid reading
code, we still need to readcode, uh, and there is no real
huge product uh in the marketwhich was fully wipe coded
(16:14):
without reading and fixing thereal code.
It's still a case where we needto read code.
And when we generate too manylines of code, we also uh lose
some um how to say, you are youare not fully understand what it
was, uh how it was generated,how it actually works, because
you are not who writes theselines, you are not aware of
(16:38):
them, and you have no psychologybound to project.
You are not understanding it uhso good as you can do it when
you write it.
So I think agents can alsocreate problems here.
So there is some performancespeed up, and there are some
issues with quality, I think.
It's my from my my perspective,uh it can be some issues with
(17:00):
quality, but still many of theseissues can be fixed with proper
usage of it.
Like you can uh generate uh youcan use some strict uh uh
limiters uh to avoid somestandard uh problems, you can
improve uh what uh tools you useinside your agent, for example,
uh you can improve uh system,how you work with uh cloud or
(17:23):
agent MD file, how you try uhlike everything how you can
handle and increase quality ofcode, it actually gives you a
lot of truth.
And MCPs as business central MCPis a good example, as you
mentioned before, businesscentral MCP, which is a great
tool where you would like tomanipulate data inside business
(17:46):
central uh through your agent,as example, and because of it,
you can simplify processes.
There's many points, uh testing,testing, you can generate tests
uh for A and before, and I'm notI'm not sure if you're aware, I
think you're aware, but beforeuh if a customer asks you to
(18:08):
write some or to build someproject, and if you will try to
try to mention tests and youwill say some numbers, they will
mostly say no because I don'tcustomers don't see value in
tests, direct value.
There's value, of course, butthey don't see it as direct
value.
But now with IA you can generatetests very easily, and and you
(18:29):
know and not just you can, but Ithink you have to generate
tests.
I think test is now uh standardfor uh IL.
We all need to generate tests,we all need to understand a
little bit how it works, butmostly we can delegate this.
And uh I like the idea that Ilistened or heard on uh LinkedIn
(18:52):
from Tiny or from uh Vecco thatTDD is a great choice for L.
Like we can start from tests tobe sure tests are correct, and
after only after that go todevelop something.
SPEAKER_01 (19:08):
You you hit on a number of key points in that
whole conversation.
SPEAKER_04 (19:12):
It's yeah, it's uh just in general, but we can
focus on something specifically.
SPEAKER_01 (19:17):
We'll get into a lot of those because uh as with many
uh individuals that we speakwith, I hear some of the things
that you're saying, and I justget excited, and my mind goes
deeper into it because I spend alot of time now doing this.
I uh I have some questions foryou, but I want to make some
comments on.
We can start with the tail goingforward here or backwards.
(19:41):
Testing, I agree with you.
Customers used to not see thevalue of tests, but I don't know
if it was necessary to tell themthat we were doing the test or
if they should have just beenincluded in the project.
Thankfully, now with AI and therate of development that we can
(20:01):
make, those tests can now be apart of it.
Uh, it's interesting abouttests.
I did some experiments the otherday with test TDD, I've been a
fan of for a long time because Ifound it also helps architect
the code better because now itforced you to modulize the code
to have small tests in theprocess.
(20:24):
And by doing it now, AI can helpyou do the tests and also plan
the development for it from thetest.
But an interesting thing, uh uhjust had a conversation with
someone this morning aboutcreating tests, and now the same
thing.
Everybody's starting to saytests are required, and not only
having tests, but proper tests,right?
(20:44):
The tests need to fail properly,the tests need to pass properly,
not just pass fail.
You need to make sure it'sreally feeling failing for the
right reason.
But I went back to an oldproject that had few tests
because sometimes it's notalways easy to come up with all
the scenarios that you want totest for, to be honest with you,
because users will do thingsthat you may not think of.
(21:06):
So I asked a you know, the AImodel or mentioned to the AI
model or the framework, excuseme, are there any other tests
that I should test for?
I'm simplifying it.
And it went through and itcreated or suggested, I was
planning at this point, so manytests that I didn't even think
(21:27):
of, but were valid tests,meaning the code should pass
these tests.
I then had it again, I hadexisting tests.
SPEAKER_03 (21:38):
So this isn't just is this against an old project
that you did this against?
Okay.
SPEAKER_01 (21:43):
Against an old project that had some tests in
there already.
So it had my I have aninstruction file that we're
using now for uh modeling how wewant the tests to be created.
But I said, go back and can yousuggest some tests?
It created a whole List oftests, and then the question is
here's the plan, this is whatI'm going to do, and we all see
(22:05):
it.
Now that I have the plan, do youwant me to go ahead and do this?
Right?
I said yes.
It went through and created allof those tests, and it made one
mistake for the set of probably10 new tests that it made.
SPEAKER_04 (22:19):
It's a suppressive, right?
SPEAKER_01 (22:21):
Followed the structure because I do this in I
follow the Garrickin scenario,the you know, the g the the
scenario, the give and the win,the then.
And it created, I had anexisting library code unit with
certain functions, and then Ihad tests broken down by area.
It did it.
SPEAKER_03 (22:39):
And I was is this a project that's already live that
you went back and ran the test?
SPEAKER_01 (22:46):
Yes.
It's an extension that'screative.
So we had some tests and it wentthrough and it created the tests
for the code.
And it just goes to show, as youhad mentioned, there is no
(23:08):
reason now to not have thetests.
And now also um, what I foundwith testings from a previous
scenario is code was introducedto a project that if we had not
had tests for it, we would nothave identified a problem that
(23:30):
it created.
Right?
So now you have extensions, youhave code that works.
If you don't have tests for it,if somebody wants to add
functionality to it, you mayinadvertently, because you're
not aware of the flow, and thisis where I think you would get
into it becomes more importantbecause now AI is just creating
so much code that it's very easyto break something else that's
(23:56):
auxiliary to uh your project.
SPEAKER_03 (23:59):
So I think it goes back to the value component
because I see that you went backto test an old project that's
already in production.
A end user or client would notsee that as like a direct value.
Obviously, it's a value to youas a developer because now
you're like saving this clientfrom future headaches if someone
(24:21):
was to extend it.
But I think that's the missingcomponents.
Like uh back in the day, whetheryou test it or not, you create
test scenarios, you you uh theclients imply that you're
already doing that.
Like you've it's like you shouldbe doing that, right?
But there's different types oftesting.
SPEAKER_01 (24:39):
And I think it's it's too much to when I I think
I think testing's important'salways been an important, I
think it's even more importantnow.
Not again, I'm I'm nottrivializing how important it
was before.
I I've had years of sessions atconferences on creating tests.
But I think telling some, I Ithink saying I'm developing an
(25:03):
app and part of this is creatingtests, part of it's creating the
documentation, part of it'screating this.
To a customer, it's hard toassess the value because they
don't know the value.
They the tests are going to savethem in the future from
mistakes.
But to your point, Chris, Iwould assume that if you're
going to deliver something tome, it should work.
(25:24):
Why why why should I have to saytest?
So it's more of here's whatwe're doing, this is what the
cost will be, is it of value toyou?
And I'm going to give youdocumentation which AI can help
create.
The testing shouldn't even beit's almost like saying I'm
coming over to your house,Chris, but I'm going to walk
(25:46):
over there.
Like, does it is it really evenmatter?
It just matters that I'm goingto get there.
SPEAKER_04 (25:50):
From my experience, customers some sometimes try to
uh have more detailed numbers,like they try to understand uh
why you have numbers, why youhave this amount of foul
spending.
And and if you will say I spentsome amount of files to test in
the past without IA, they willbe not so happy, you know.
SPEAKER_01 (26:12):
No, I understand.
I I've lived it and I understandit, but it's almost a matter of
somebody needs to test itsomewhere, whether it's
automated testing or developertesting or user testing.
No one should implementsomething without testing.
SPEAKER_04 (26:30):
Yeah, yes.
My favorite type of testing isproduction testing, right?
SPEAKER_01 (26:36):
There are many implementations where that was
their model.
They they assumed that thedeveloper uh developed the
extension properly and they justthrew it in, and then they get
mad when nobody tested it.
And it's it's uh interesting.
So I mean, so the testing isextremely helpful, and I think
(26:58):
now it's even more important,but there is an offset because
of the speed at which we cancreate code, we now can include
more such as testing, extensiondocumentation, and those
infamous guides that nobody usedto write.
SPEAKER_04 (27:14):
Yeah, and I think together with it, it actually
means you are not losing qualityor almost not losing quality,
because on top of uh fastgeneration manual line of code,
you have tests, you havedocumentations which you don't
have before in the past withwithout artificial intelligence.
So actually, I think uh qualitymay be even better a little bit
(27:38):
if you have processes in correctway configured, you have CI CD,
you follow all best practicewith IA, with working with IA,
it actually can even uh increasequality.
SPEAKER_01 (27:50):
Yes, absolutely.
So another thing you went backto talking about was the amount
of code that we can create umwith AI.
And testing, just to come backand loop up a testing, testing
is not just extensions thatpartners may deliver or
contractors may deliver.
If you're a developer for acustomer, some some more
(28:13):
implementations have their owninternal team that can create
extensions, they should alsotest as well.
The testing is not specific todeliverables from partners or
outside agencies.
Even if you're workinginternally, you should take the
time to create those tests foryourselves and you get better
quality.
And as I had mentioned before,you can make sure that any
(28:37):
changes or additions that youmake won't impact your existing
workflow or your existing code.
Uh even to the point where ifyou have multiple extensions, I
always say run your test for allof the extensions as well.
Because I have found, and thisis uh where we jump into AI
(28:58):
assisted development, where haveyou found AI to be helpful or
not so helpful with development?
Have there been any situationswhere you can say that AI for
sure saved you a lot of time?
And other cases where you saidAI probably created more work
(29:23):
than it saved?
SPEAKER_04 (29:25):
Good question.
Actually, I have severalexamples where I have many
examples where it's helped me tosave my time.
For example, when you writetests, as you mentioned, it's
actually a good uh point to showwhere tests are save you time
time dramatically.
Um in my experience, but it'smore interesting to to to speak
(29:48):
about where it's actually uhcreate some problems for you or
it can take your time.
Uh and first of all, uh it cancreate you some uh difficulties
when you uh collaborate withother people who use IA, but
they don't uh understandproperly how to properly use it.
(30:09):
They are not using um uhspecific uh uh best practices
with agents, they're not uhunderstanding how it works, they
think it is like a magic and uhthey don't provide enough
context to agents, and uh thequality of this work can be uh
less than you expect.
And as open source maintainer,uh I realize that sometimes
(30:34):
people come to uh repository, tomy repository of my open source
project, and they try to uhcreate pull requests, uh
contributions to uh to my uhthis like for example that
editor tool, and they generateit via agent without proper
specified context of problem,without proper specifying uh
(30:56):
what is a problem about, what isit we you try to solve, and
agents generate something whichcan be like matched.
And I heard that many uh opensource uh maintainers encounter
issues that they are overwhelmedwith number of pull requests,
with number of contributionswhich people try to bring to uh
projects uh because of IA,because I just generated, and
(31:20):
they are not understand how AIworks, and it creates some
overload without um necessary uhpurpose.
It's one example where it can bea problem.
SPEAKER_01 (31:32):
Yes, yes.
Uh I I've heard that as wellwith open source projects that a
lot of AI pull requests arebeing made, and it's difficult
for the maintainers to keep upwith those changes to review
them properly.
It's interest, it's it's it's aninteresting time that we live
in.
The stepping back into ALdevelopment with AI and some
(31:58):
challenges that you may havehad, or any suggestions for
someone that may be working withAI and AL development uh when
they're creating extensions.
What are some things that youhave found, whether it's working
with ISVs and how AI could uhhelp with that, or any other
(32:22):
places where you notice somechallenges using AI?
SPEAKER_04 (32:26):
Yeah.
I see some I got some patternsfrom my experience when I work
with AI, how I use it, how I tryto get most possible value from
it.
Uh and first of all, I think youalways have to start with
planning mode.
Like if you um have some problemor you would like to build
something new, uh like totallyentirely new application, do not
(32:50):
start quote or ask uh uh agentto quote for you.
Focus on plan.
Ask agent to plan.
Like give all possible contextwhich you can give, like what
you can know about your problemor about idea of what you would
like to implement.
Give proper context to model.
(33:10):
Try to uh highlight whatspecific if it's some
modification of some existapplication in uh ISV of a PGA
extension, highlight specificobjects which you think are um
related or can be related tothis problem, highlight why you
think about it.
Try to communicate with agent asyou are communicate with uh
(33:31):
person, like try to understandthat it can magically understand
uh get all information if youare not specified.
It can, but it will lose contextwindow, and quality of uh
generation will be lower andlower because context management
is very important when you workwith agents.
So you try to provide as lessbut as much as possible uh
(33:53):
context because it's actuallylike you're always fighting with
uh two different uh problems.
Like one problem you provide notenough context, so agent cannot
generate something for you.
Another one you provide too muchuh context and it's like
overwhelmed and it's unfoundedon uh your specific idea and
task.
SPEAKER_01 (34:11):
So that's a key point I've I've heard from
people.
It's less is more.
Don't get into please thank youand keep adding additional words
or even try to tell it what todo, in a sense.
If you the more boundaries thatyou give it, it seems to have
(34:32):
more of a challenge with theoutput.
SPEAKER_04 (34:36):
I agree.
And when you set someboundaries, sometimes if you're
very sure about it, it can beuseful.
But many times I realize whenyou set some boundaries, like
you try to lead agent in somespecific way, but you can be
sure that your way is the bestway to do things.
Like you have to learn from itas well, and try to try to look
(34:56):
what it suggests, and then youwill see all possible uh
branches of solutions how youcan solve the problem.
Uh you can like choose the bestone.
Um so plan mode, contextmanagement, uh session
management.
Of course, you don't say I I I Isaw this mistake very often in
(35:17):
the past.
People uh speak with agents inone session, they're not pay
attention on context window, anduh agent just overwhelmed with
all different specific taskswhich uh people try to manage in
one session and screen problem.
Start new sessions, try tomanage it.
Um what else?
Also, it's important to chooseright tool because there's many
(35:41):
benchmarks in the internet whichthey specify, like we are best
agent or we are best model forspecific agent.
Uh it's hard to uh predict whatis better for AL development on
those benches because they aremostly based on C sharp, Python,
on other languages.
But uh recently Microsoftreleased a BC uh bench uh for
(36:06):
agents and they uh it was as Iremember they only count two
agents, quote quote, and maybeuh codex, not sure about codex
and it's one point when you canlook on it.
Try to find IL benches tounderstand it, but actually what
I will uh give one small adviceto everyone who is interested in
(36:27):
it is try to yours try to use itby yourself, try to use
different tools.
Uh because if if it's somethinggood for one person, it can be
not good for you.
You have to choose, you have totry, and you have to spend time
on it, and and it will give alot of value uh later when you
will realize what tool isbetter.
(36:48):
Because there's like cursor, uhcodex, cloud code, uh uh
copilot, all different toolswhat you have.
But my honest answer, try it,and you will see the difference.
Try to compare it by yourself,by your internal tickets.
I even have some small internalum tasks, and I try to uh
(37:12):
evaluate those tasks every threemonths or something like that.
Six three months uh on all uhpopular agents or clients which
I can use, and I can see thesituation is changing.
Like six months ago, codex wasbetter, six months ago, compilot
was better, and so on and so on.
It's constantly changingsituation.
(37:32):
I prefer Clot Code, I even uhwrote about it uh several times.
Um but I'm not sure if it willbe forever.
Like maybe tomorrow it will bebetter, something from Google,
who knows?
SPEAKER_01 (37:46):
Like so, one of the things as you're going through
your development workflow withAL development is not all models
or frameworks work the same, andwhat may be good for one person
for one task may yield differentresults for somebody with a
(38:06):
different task.
SPEAKER_04 (38:08):
Yeah, and also it uh depends on how you communicate
with model.
Every parse person has a uniquestyle of communication, right?
And it also affects uh how uhhow what output you will get
from agent from models.
SPEAKER_01 (38:22):
It's a challenge to keep up with.
SPEAKER_03 (38:24):
For the both of you, so how does that work when you
have uh when you're working on aproject together and you have
maybe three of you working on asolution, and then they may have
someone that's a third party,outsource whatever, uh that a
client hires.
Do you all use different uhframework?
(38:47):
And how does how do you maintainthat?
Like how does that work?
What's the workflow for that?
SPEAKER_04 (38:52):
Uh from my perspective, I have such
experience um very often thatpeople use different tools and
agents.
They uh one developer usescodex, one developer uses
copilot, uh, one developer useuh copilot inside Visual Studio
Code, one is using copilot uh asuh command line, right?
(39:15):
Technically, it's not soimportant that people use
different tools because the mainthings to get the best output
from it.
Uh I think like even if you areusing two uh different tools,
but it gives you good output, itdoesn't matter what tool
(39:36):
created.
But how you uh communicate withdifferent agents uh and your
experience, it can affectquality of work what is uh will
be as output from agent.
But at the end, it doesn'tmatter what was by what tool it
was generated.
(39:57):
That's what I see.
SPEAKER_01 (39:59):
Yeah, I agree with you.
I I uh it's a it's a goodquestion, Chris, because it's
not to confuse the tool thatsomebody's using to create
something, it's what they createis the important thing.
As Old Me had mentioned, it'sthe output of the quality.
What becomes important in thosesituations is not the focus of
(40:23):
which AI tool they used or whichedit IDE they used, it's what
they did and how you manage it.
How you manage your developmentworkflow with source control
management, with branching, withmerging, with code review, those
are the things that becomeimportant.
So you can have an internal teamthat has multiple developers,
(40:45):
and then you may need to haveanother person also contribute,
they still can fall into yourworkflow.
What they use to create it's notimportant.
So even if you are a customer ofor a user of Business Central,
excuse me, and you're managingyour own source code and your
(41:06):
own uh repository and workflow,you can bring in others and have
them fall into your workflow.
You just need to have thediscipline of how to manage the
different branches.
SPEAKER_03 (41:17):
Yeah, I think I think that's a good point
because I think down the roadthere's going to be, I feel as
though there's going to be moreof that as a lot of clients uh
understand that you can havemultiple partners working on a
solution and as people vibecode, you know, someone has to
manage that.
And then at some point, I mean,you're right, it doesn't matter
(41:39):
what tool they use as long asthe output is the the the right
expectation or the correctexpectation of the output.
Um but at the same time you mayhave some balance of like if
you're using a specificframework, then your testing
scenarios may produce differentthings or different testing
scenarios versus someone thatjust you know uh committed their
(42:00):
their code into.
So it's it's an interestingbalance of what that looks like
in the near future, uh,especially when your clients may
want to be more involved inthat.
SPEAKER_04 (42:10):
Uh because you know, actually, yeah, it's uh great
points.
And uh CICD is what can help usto manage all this uh mass
workflow.
Uh we just can um try toautomate as much as many process
as we can.
Automate uh test running,automate uh linters, uh automate
(42:30):
uh compilation of projects,automate everything of it, and
it creates and it starts to beeven more it was important
before, before Artisan Terenceuh was starts to be more
popular, but uh it's importantnow even more, same as testing
all the CICDE processes, and uhthat's why we all need to look
on LGO, how it works, how we canimplement it in our projects,
(42:55):
and it's actually for free.
You can implement it for your uhopen source project and for PG
extension.
Go ahead, use it.
And I actually very um I almostall my projects that I work with
customers.
I always try to say we need toimplement CICD, we need to
implement LGO in uh thisspecific project, and I
(43:16):
explained why, like uh like whatvalue it brings to customers,
and mostly most of the time theyare actually supported because
they see it's actually very easyto show them uh value of
workflows of how automationshelp.
Uh for example, from to save uhproject from regression bugs or
(43:39):
problems, a lot of things can beum automated, and but I think
it's main tool also what we canfocus as developers of L
developers and to using ALGO, tousing uh CI CD processes.
SPEAKER_01 (43:55):
Yeah you brought up a good point.
ALGO is It is a free tool thatruns in GitHub.
There could be a cost to therunners that you use, depending
on how you set that up for yourprojects, just to put that in
there.
There is a tool for it.
But I I I literally had aconversation with someone this
morning about that inimplementing these on projects,
(44:18):
that they were having problemswith developers building the
workflow is to have a branch,review the branch, bring it into
the merge it into the mainbranch, as we'll call it, and
then that gets deployed.
Whereas if you have a developerthat didn't update their main
(44:39):
branch with other code changesfrom other branches, and they
just did it with their branchmerged in and then uh built the
application and deployed it, youin essence can remove changes
that were made.
So by using tools such as ALGOand GitHub, the extreme benefit
is you have the security ofknowing whatever was built, the
(45:03):
artifact that gets built as partof your CI CD is what should be
deployed.
You don't have to auto-deployit, but that's where whomever is
going to load it.
And I'll be the first to say isI don't think developers should
be the ones responsible forloading the changes that they
make.
You should have someone whoknows how to pull down the
artifact, and then that's whatgets loaded.
But that just verifies that thetests were run because the AL Go
(45:25):
will run the tests that arewithin the extension.
And it also validates thatanything that's in the main
branch is what gets built, used,and deployed, and you're not
relying on somebody refreshingtheir branches or refreshing the
changes of the code that theyhave because using Git you have
a local copy of it.
Yes, it does sync, you know, youdo pull and push from the
(45:48):
central location for it, but youstill have to go through that
exercise of doing that.
Yes, it can be automated, yes,you can do it, but some people
still like to do it manually.
So the point that you had madeis using tools like that in this
world of AI development to meand team development becomes
even more important.
Not that it was less importantyesterday, it's more important
(46:09):
today.
So you these are some goodthings.
I know we're talking about AIand AL development and open
source, but um so it with yourAL development and AI, what
tools do you find useful to helpyou with it?
Are there any other tools thatyou have found can assist you in
(46:30):
this journey, or even uh ifsomebody wants to start using
it?
You had mentioned a little bitago, just start using the tools,
right?
Just start using AI for yourprojects.
What are some things that youfound were helpful to get you to
where you are with yourdevelopment journey?
SPEAKER_04 (46:48):
Um, yeah, there's several tools which I uh think
is important uh to um to startwith or to learn or to look on
it uh a little bit.
Uh first of all, uh I think it'sa good point to get some basics.
Uh because if you look on uhdifferent tools, as example,
like uh agents in common line,uh they all have some different
(47:13):
same uh similar uh technologiesbehind like MCP.
What is MCP like if you try tounderstand what is MCP it can
help you to uh get moreefficient with any agent?
Like you are not focused onspecific tool.
Uh so I will start.
I think it's better to startwith some basics, like to
(47:34):
understand this MCP.
Like, okay, MCP is some protocolwhich will uh allow my uh IA to
uh communicate and to changesomething on external services.
It can be business central, itcan be your browser MCP, it can
run uh for you some uh businesscentral um page and even try to
click some uh buttons directlyin uh your um browser if you
(47:59):
need it.
And all these are done by viaMCPs.
So start from basics tounderstand the context windows,
MCPs, but speaking aboutspecific tools, yeah, I have
several in my mind.
So I started uh originally withuh um cursor, if you heard about
it.
Uh it's like uh form of VisualStudio Code, but with
(48:23):
implemented um models of AI.
But it was like in the beginningof the uh my journey, and um I
realized that now you are notneeded uh cursor because
everything works inside VisualStudio Code.
Uh naturally you can have uhseveral different models.
(48:45):
So now I think it's good tostart from Visual Studio Code.
It's our daily uh EDE where weas L developers work every day.
So it may be uh good startingpoint to use uh copilot inside
Visual Studio Code to selectsome models.
You can select different models.
You are um you can select uhanthropic models, you can try
(49:06):
open eye models.
Try to mix it, try to to startwith EDE.
I think ED is a good startingpoint.
Uh next, what I uh think isworth to try after you focus on
some Visual Studio Code.
Um I think it's next this willbe useful is uh common light
(49:27):
agents.
They are independent from ED,you can run them inside your
terminal.
And there is some value behindit.
First of all, you can run iteverything there.
Next, they are working this filesystem differently than ED
(49:49):
works.
It works uh per my experiencemore efficiently with um all
these uh commands which you canexecute inside your uh terminal,
like bash command, uh PowerShellcommand.
It's very easy to automate, andyou as a developer can easily
handle it.
Um so try some command-lineagents, and there are several
(50:11):
popular agents on the market orin general.
There's like uh compiler agentas well, there's uh codex from
OpenAA, and there's uh CloudCode uh from Anthropic.
Uh I try to use them and try tomix a little bit to understand
what tool is better now to whatis better than now because it's
(50:34):
changing dramatically, as Imentioned.
So try with Visual Studio codeED agent inside it, try to um
read the basics.
What is MCP, what is contextwindow, what is uh slash
comments, or uh uh what is um MDfile skills, it's it's actually
(50:55):
called skills now.
Before it was slash comments,now it's skills.
Uh how skills can affect your umgeneration, agent generation.
Try to next do next steps to usesome MCPs actually to on
practice.
Next steps try to use commentline agents, try to compare
different agents to see what uhvalue uh you can get from it.
(51:19):
I think uh also try businesscentral MCP, it's also a good
point to check uh how it works,how data is flowing, how you can
just tell to agent, please showme or delete customer numbers.
Of course, it's not forproduction, you just do your
daily work on some sandboxes oron some uh uh Docker
(51:40):
environment.
So I think those like mainentrance points that you can go.
SPEAKER_01 (51:50):
It does sound like there is a lot, but as with
anything else, it sounds worsethan it is.
And I think once you startworking with it, it all comes
into play.
And I think the important partis to start using it and not try
to know everything, just knowwhat you need to know to get
(52:11):
kind of work your way throughthe task and see what you need
to do for that task to to be uhuh be successful with it.
SPEAKER_04 (52:21):
Yeah, and I I can tell you practice and practice.
Like if you just will readtheory, it's will not be so
useful.
Try to do it in practice.
I realize by myself that when Ido practice, I um start to
understand things better.
But when you just read theory,yeah, you know MCP, but try to
it in practice, and you willrealize uh actually how it can
(52:44):
be useful for you and how it canuh give you some output,
valuable output.
SPEAKER_01 (52:53):
It's it's um yeah, it's a strange world now with
all of this output and uhbusiness central developments
coming along, and the businesscentral tools that they keep
adding are also nice to have,and it's nice to see that
they're adding all these toolsto assist with the development
point of view.
(53:13):
And in my opinion, it's alsoreinforcing that the days of
just development are gone.
I think that developers need tohave a good understanding of the
application and the function ofthe application so that they can
work with AI to develop thesesolutions.
(53:34):
And also, as you had mentionedat the beginning, the importance
of still understanding and theability to know the architecture
and the code.
So it's not it's not one ofthose things that, oh, AI, I can
create this extension, it'sgoing to do this thing.
I think you do need to have asense of understanding and
(53:56):
review how to review what getscreated.
Uh, I think I keep sayingbusiness central to me is a
little bit different becauseit's in a box.
It's a predefined applicationthat you need to work with
within the constraints of howthe process and the application
works.
We can't just change a completeapplication, every build is a
(54:18):
standalone application, it hasto work within a framework.
So it's important to be able tounderstand that framework so you
can review what gets generatedand then deliver something of
quality.
SPEAKER_04 (54:31):
It's is that I actually agree, and you bring a
great point.
Like as business-centereddevelopers, we are not just
developers like um Pythondevelopers who just know
language and know all these umkeywords and how to do and write
some algorithms, but withbusiness centered as
business-centered developers, weare also should and we actually
(54:54):
should really and must know thebasic functionality of business
center, how it actually worksright now, how it actually works
from a functionalityperspective, not just from code
how it's coded.
It's one part, and butunderstanding how it works for
different types of businesses,how you can configure it, it's
(55:14):
actually super important in uhany era, even before I it's also
was super important.
Understanding of how it works,it's actually because if you
without this understanding, youcan build efficient tools, you
can build efficient uh uhsolutions for your customers
because you will not aware howto handle it in the best way.
(55:36):
That uh you don't need to quoteanything what you hear what the
customer asks you.
You need to think, okay, is thisalready a functionality existing
business central?
Can you just implement just useit without coding anything else?
Because I saw in my experience,many poor developers try to
build something immediately whensomeone asks them, like, please
(55:58):
create me this and this.
No, try to think if this isactually required to build, if
it's actually maybe already apart of standard business
central library, maybe you'renot aware of it.
Try to do some research beforeyou say something.
Try to think.
And yeah, so I thinkunderstanding of functionality
is important, and now with AI,it's even more important because
now we have more time forarchitecture and also more time
(56:23):
for understanding functionalityof business central as L
developers.
Like now you cannot say, Oh, I'mjust code.
Noah, it's not on, man.
SPEAKER_01 (56:31):
You hit that well, and you answered the question,
and I think it's a great way, Ihate to say it, it's a great way
to wrap it up.
What do you do with all thatfree time?
That's the question thateverybody's saying is okay, with
all this free time, I don't havefree time.
Don't look at it as you havefree time to go uh you know, not
(56:53):
work as much.
Now you have free time to, asyou had mentioned, learn the
application.
You have more time to understandthe architect, the application,
more time to architect and plana solution.
So it's not necessarily that wehave all this free time.
I can get work done in twominutes and now I'm done for the
(57:14):
day.
It's I can deliver betterquality work.
And this is where that questiongets asked.
What do you do with the time?
You take the extra time to addthat extra quality to it.
And again, not to say thatquality wasn't important before.
I'm not trying to discount it,but now you can deliver a better
(57:34):
quality product in a shorterperiod of time.
That to me is the answer to allof this.
It's better quality in a shorterperiod of time.
SPEAKER_04 (57:46):
I fully agree.
SPEAKER_03 (57:49):
That's a good one.
That was a good, good with that.
SPEAKER_01 (57:53):
I think uh would like to thank you for taking the
time to speak with us today.
I could talk about this stuff,and I'll probably now I'll just
have to put you on my list ofcalls to have that I could talk
about this stuff all day, everyday.
SPEAKER_04 (58:03):
Because we can spend days and hours to speak about
it.
SPEAKER_01 (58:06):
It's actually board topics like oh yes, and it's uh
you're in some of the same chatgroups we're in, and we get to
talk about this a lot, and uh,you know, it's the wonderful
chat groups because I'm in theUnited States, most of you are
in Europe.
I can wake up in the morning andthere's like 65, 70 messages I
have to read through and screenthrough.
(58:28):
And it's just like you know,really what's going on.
But it's it's it's a it's anexciting time for everybody.
And again, thank you very muchfor taking the time to speak
with us today and sharing yourvaluable insights.
I think anybody who's in the ALjourney, uh, we're all in the
beginning of this journey.
Anybody who says they're anexpert on it, I mean, how can
(58:49):
you be an expert fortechnologies that's so new and
it's changing?
So anybody that's uh currentlyusing AI, hopefully that uh
thank you for sharing so manyinsights.
Somebody that's new to it, youalso shared some insights for
that as well.
And if anybody would like totalk to you more to learn about
your open source projects ormaybe any other AI for AL
(59:10):
development tips that you canshare with them, what's the best
way to contact you?
SPEAKER_04 (59:15):
Sure, um there are several ways.
Um you can have just uh commentand put a comment in uh my blog.
I have a blog uh where I uh tryto share my um knowledge, what I
learned from uh my experienceabout development specifically
for business center development,also some some IA insights and
tools.
Uh you can comment it.
(59:36):
You can contact on LinkedIn ifyou want to ask anything.
I'm very open-minded.
If you just just ask me onLinkedIn, I will immediately
answer you, or when I will havetime, of course.
Um, so LinkedIn, my blog, I havealso um uh Blue Sky.
Anything of it will be okay tocontact me.
SPEAKER_01 (59:59):
Uh great, thank you.
And we'll have your contactinformation with the show as
well.
Uh again, thank you for takingthe time to speak with us.
Look forward to speaking withyou again soon.
Uh enjoy the rest of your dayand have a great weekend.
Talk to you soon.
Chao ciao.
SPEAKER_04 (01:00:12):
Take care.
SPEAKER_01 (01:00:13):
Bye.
SPEAKER_00 (01:00:15):
Thank you, Chris, for your time for another
episode of In the DynamicsCorner Chair, and thank you to
our guests for participating.
SPEAKER_03 (01:00:23):
Thank you, Brad, for your time.
It is a wonderful episode ofDynamics Corner Chair.
I would also like to thank ourguests for joining us.
Thank you for all of ourlisteners tuning in as well.
You can find Brad atdeveloperlife.com.
That is D V L P R L I F E dotcom.
(01:00:43):
And you can interact with themvia Twitter, D V L P R L I F E.
You can also find me atmattalino.io, m-a-t-a-l-in-o.io,
and my Twitter handle ismattalino16.
And see you can see those linksdown below in the show notes.
(01:01:06):
Again, thank you everyone.
Thank you, and take care.
Task worker.
It's like a library fordevelopers where you can able to
uh easily to split some uh hugeslow processes into chunks and
run it in separate sessions.
Because I encounter this problemvery often in my experience.
SPEAKER_03 (00:16):
Welcome everyone to another episode of Dynamics
Corner.
What does Two Cats have anythingto do with AI, workflows,
framework, etc.?
We're about to find out.
I'm your co-host, Chris.
SPEAKER_01 (00:30):
And this is Brad.
This episode is recorded onApril 10th, 2026.
Chris, Chris, Chris.
Two cats in AL development.
That's an interesting concept.
Also, we learned about tips andtricks for starting your AI AL
development journey, as well aswhat you can do with your free
time now that we're using inliving in the world of AI.
(00:52):
With us today, we had theopportunity to speak with you
doing.
SPEAKER_03 (01:17):
It's morning for me.
It's eight o'clock in themorning right now.
SPEAKER_04 (01:20):
Oh, it's your also morning.
SPEAKER_03 (01:22):
Yeah.
unknown (01:22):
I see.
SPEAKER_04 (01:23):
It's uh it's 6 p.m.
for me, so it's for me, it'sevening.
SPEAKER_01 (01:27):
Oh dinner, dinner time.
Your day's over, it's good.
I like uh I like the the catsthat you have joining you as
well.
SPEAKER_04 (01:36):
Um it's not planet.
Immediately when I uh join uhmeeting during our session, they
just decide to be here.
I think they would like to.
SPEAKER_01 (01:48):
They wanted to join with you.
What are their names?
SPEAKER_04 (01:50):
So uh white cat is uh he is uh Hades.
SPEAKER_01 (01:55):
Say that again, please.
SPEAKER_04 (01:56):
Hades.
It's like uh yeah, Hades, Hades.
From the Greek uh Yeah, fromGreek mythology, it's actually
yeah.
And she is Nyx.
SPEAKER_02 (02:06):
What is that again?
SPEAKER_04 (02:07):
Nix Nix, it's also from Greece uh mythology as
well.
So that's like some interestinguh name choice, but uh I I love
them, they I like them, and uhthey're like very new for me.
Uh only two years ago.
SPEAKER_01 (02:24):
I I I I it's it's nice, it's it's good to have a
companion uh with you whenyou're you're sitting at home
working, and it's it's been awild, crazy couple of months
with uh with technology, withdevelopment, with AL.
(02:45):
Business Center released Wave 1for 2026, just a short period
ago.
Usually, and I have to figureout why, usually I go through
all of the features well beforethe release and experiment and
play with them.
But I think with the way thingsare going, I've been extremely
(03:07):
busy and and I was doing somereflection and figuring out why
am I so busy.
And I think part of it isbecause in the age of AI and
development, we can do so muchmore, so we're busier doing
more, whereas before we wouldtake the time to do things.
And I was looking at the list ofdevelopment editions for 2026
(03:35):
Wave One, and I'm curious toknow if you had the chance to
work with them because I becauseI just started working with them
recently.
Uh some of them I'm reallyexcited about.
SPEAKER_04 (03:49):
I actually take um short look on it.
Uh I also was not able to handleit uh fully, and uh it's like it
was a lot of uh new changesregarding agents, if you notice.
Uh like now you can uh buildsome agents in uh your sandboxes
(04:10):
environments, and uh after thatyou can export it and use it in
code.
Uh in theory, it's just uhperfect idea.
I like it.
Uh there's many differentchanges uh regarding um how uh
you can use unified interfacesto work with IA inside business
(04:30):
central, like um those specificum code units interfaces.
And I even use it, I I I try touse them.
But some of some of my part ofexperience was great, some of
it's not great, like there aresome limitations, like you
cannot uh pass um files and uhwork with them efficiently, like
(04:52):
it's it's still in uh in thestage where it's evolving, and
there are some challenges whichwe are uh right now trying to uh
to solve and Microsoft trying tosolve.
SPEAKER_01 (05:05):
So yeah, it's yes it it is, it's exciting.
So it's the trouble I haven'tplayed with it.
I I've looked at some chats, butthe enabled troubleshooting MCP
server for AL, I really want toplay with.
I I know that's due in May, soI'm I'm looking forward to
seeing that.
The AL tests from Visual StudioCode, I'm excited about that um
(05:30):
should be out now.
And also, as you had mentioned,BC Bench or uh agent preview or
agent playground to be able tocreate agents in the sandbox.
I've created many of those andhave been running the tasks
through the sandbox userinterface.
I really do hope that they makeit to where you can create the
(05:53):
agents and not have to createthe code for the agents with
what you create with thepreview.
SPEAKER_04 (05:57):
But it feels it feels so natural to just not to
quote it, but just to um likeconfigure it and use it.
I totally agree on that.
SPEAKER_01 (06:04):
Yes, I I've created many and you go back and the
task run, and I just feel like Ishould be able to click schedule
task or run task or dosomething, and uh that's my
hopes that one day we get therebecause as you had mentioned, it
feels so natural, and you can dosome uh great things.
I've had some demonstrationsthat I've done with it, and
that's what everybody's sayingis why do we have to code it
(06:26):
now?
It's already there.
SPEAKER_03 (06:28):
Uh maybe, maybe it won't take too long.
Maybe uh give it a few months.
SPEAKER_04 (06:33):
Yeah, it just it's just very fresh.
Or maybe we can contributesomehow.
Who knows?
Maybe, but still in future.
SPEAKER_03 (06:44):
That's what they want though.
That's what they want, right?
They want you to uh contributeand enhance it.
Here's the tool.
Make something right.
SPEAKER_01 (06:55):
So so those are some of the big ones that we're
excited about, but uh wedefinitely wanted to speak with
you about a lot of things frombusiness central uh with what
you're doing from the technicalpoint of view, because uh I know
you're doing quite a bit indevelopment and development
workflow.
Uh, but before we talk aboutthat, can you tell us a little
bit about yourself?
SPEAKER_04 (07:15):
Uh sure.
So um I'm Volodymyr, uh ABCdeveloper from Ukraine, a recent
Microsoft MVP, uh, and I havebeen in the ecosystem for nine
years.
Um I have worked with BC indifferent capacities.
I work at B Dynamics with Swisspartner, Microsoft Partner,
where I focus on enterpriselevel projects.
(07:35):
Additionally, I run companyVolodiner LC where I do
freelance work for smallmedium-sized clients and publish
my own app source extensions.
I spent also a good chunk of mytime contributing on open source
and writing a technical blog atuh vldbc.com.
Uh what I really focus on ismore developer tooling, building
(07:56):
things that make everyday lifeeasier for BC developers, uh
partners, customers.
Um my open source uh product,data unitor tool, for example,
uh, has over 160 stars on GitHuband almost 1000 installs on App
Source, and uncounted number ofinstalls on PT.
I don't know, I didn't count it.
(08:18):
And I enjoy Business Central.
I do all of this because Ibelieve in BC as a product and
its community.
I think open source helpseveryone grow, and I love
sharing what I learned, whetherthat's thoughts on my blog, at
events, or just in conversationslike this one.
SPEAKER_01 (08:34):
That's great.
So you you have the open sourceproject that, as you mentioned,
has a number of stars and a lotof installs.
And I'd like to talk about opensource projects, but some of
these tools that you've createdfor development, how are they
impacting the developmentworkflow and the benefits to
(08:55):
users of the tool when theirdevelopment process?
SPEAKER_04 (09:00):
Great question.
Uh I think uh the idea behind itis to uh fill all these gaps
what Microsoft are uh havebecause they are have many many
focuses, like IA, uh all ofthese important uh tools as
well.
But there are some gaps which Ican see uh that I can add some
(09:24):
value and add some value tocommunity to to create something
that can be useful specificallyuh to solve some specific
problem.
For example, data you tooltrying to solve problems, uh how
you can manage, edit, view, datamostly for testing uh in
environments, as you remember inold days uh you were able to do
(09:45):
it in Object Designer, uh inNavision Classic or RTC, doesn't
matter, or even a SQL.
Uh but after BC14, if I remembercorrectly, they change it and
you cannot manage data.
Uh same story.
Uh recently I released umparallel task worker.
(10:05):
It's like a library fordevelopers uh where you can able
to uh easily to split some uhhuge slow processes into chunks
and run it in separate sessions.
Because I encounter this problemvery often in my experience that
some huge processes which youcannot make it faster, it's like
impossible.
There is no uh room for that.
(10:26):
So I think uh idea behind it isto find some problem and trying
to solve it.
And if you don't see anysolutions in um in the internet
and community, then it's evenbest uh or greater uh idea to
put some effort on it and tomake life uh for developers and
partners easier.
(10:48):
That's what I think about it.
SPEAKER_01 (10:50):
No, no, it's important to see those tools.
And you had mentioned, I knowwith Business Central, you can
now, and you have been for manyyears, able to make
contributions to the applicationto enhance the features and the
functionality of theapplication.
And then also, as you hadmentioned, many individuals are
making open source contributionsto help with uh PTEs uh or for
(11:15):
the application or even somedevelopment tools.
I know a lot of individuals areadding development tools um uh
uh options available.
The cats are going a littlecrazy there.
It's kind of why I had a briefpause.
It's fighting, fighting.
Yeah, I thought there was goingto be a battle, a battle soon
with uh with the cats.
SPEAKER_04 (11:34):
They started when you mentioned open source uh
contributions.
Interesting.
Why?
SPEAKER_01 (11:40):
They're excited about the open source
contributions and well withthat.
Uh we appreciate all that you'readding, and you have a lot of
great tools available on your uhGitHub repository.
But another thing that we weretalking about previously is the
impact of AI in development ingeneral, and in general also
(12:05):
with AL.
So you've been working withdevelopment for nine years, so
you've come into the tail end, Ibelieve, of um the old way, as
everybody talks about with Cal,transitioned into AL, working
with AL.
And now, as we talked about,Microsoft's even adding
development tools, as we talkedabout the introduction to this
(12:26):
session.
How has AI impacted your ALdevelopment, or how do you see
it impacting AL developers?
SPEAKER_04 (12:36):
That's a very uh great question.
Let me think about it more.
Um first of all, I would like toadmit you mentioned very
correctly there is not only opensource projects which you
maintain by your own, there'salso contributions to uh direct
contributions to businesscentral uh itself.
You can uh contribute uh as opensource contribution uh to system
(12:58):
application, to uh baseapplication also as well.
And uh I would like to like tohighlight it to all developers
uh who listen it, or maybe willlisten it, don't afraid to
contribute to business centralitself.
It's actually add a lot ofvalue, it's uh you solve your
problem, your community problem,and product itself will gain
(13:20):
some value from it.
So don't afraid don't afraid todo it.
Start with something simple.
And uh back to your question uhabout the effect of uh how
artificial intelligence canaffect uh development process
specifically for AI and ingeneral.
Um I will start with uh somevaluable uh insights.
(13:44):
What I found out when I uh startto um realize that I is closer
and closer uh in developmentprocess.
I start to use uh IA firstly, uhI think when uh GPT 3.5 was
released.
Uh it is 2020, 21.
SPEAKER_01 (14:02):
Not I don't remember exactly the year, but it goes so
fast I can't keep up with iteither.
It seems like yesterday a lot oftimes that something was
released and someone will tellme it was months ago or years
ago.
SPEAKER_04 (14:17):
I totally agree, it's evolving very fast every
month, every six months, everyuh place or situation is
changing dramatically.
That's what's happening.
Uh and today, if you speak abouttoday, uh what I actually see
that uh IA is to make code ischeaper, you can generate and
(14:39):
create many uh many products uhwithout spending much amount of
time uh for writing all specificline of code.
You can spend more amount oftime uh to focus on architecture
of product, how you can buildthings, you can do actually real
(15:02):
software engineering, because ifyou think about software
engineering, it's not just aboutuh uh write some lines in text
in on in in in notebook, it'sactually to design, to think
about architecture, to thinkabout limitations, to think how
it will be maintained in thefuture.
And I think II give us a huge uhtool to avoid some um boring
(15:30):
work which are not superactually software engineering if
you think it about more.
So I think it gives you anopportunity to generate a lot of
code, but at the same time, Ithink uh it's also a dangerous
um because when you willgenerate too much line of code,
(15:50):
how you will understand themvery well.
Someone will tell you don't needto understand it because you can
just uh delegate all work toagent and it will do everything
for you.
Maybe, but we are not on thisstage where we can avoid reading
code, we still need to readcode, uh, and there is no real
huge product uh in the marketwhich was fully wipe coded
(16:14):
without reading and fixing thereal code.
It's still a case where we needto read code.
And when we generate too manylines of code, we also uh lose
some um how to say, you are youare not fully understand what it
was, uh how it was generated,how it actually works, because
you are not who writes theselines, you are not aware of
(16:38):
them, and you have no psychologybound to project.
You are not understanding it uhso good as you can do it when
you write it.
So I think agents can alsocreate problems here.
So there is some performancespeed up, and there are some
issues with quality, I think.
It's my from my my perspective,uh it can be some issues with
(17:00):
quality, but still many of theseissues can be fixed with proper
usage of it.
Like you can uh generate uh youcan use some strict uh uh
limiters uh to avoid somestandard uh problems, you can
improve uh what uh tools you useinside your agent, for example,
uh you can improve uh system,how you work with uh cloud or
(17:23):
agent MD file, how you try uhlike everything how you can
handle and increase quality ofcode, it actually gives you a
lot of truth.
And MCPs as business central MCPis a good example, as you
mentioned before, businesscentral MCP, which is a great
tool where you would like tomanipulate data inside business
(17:46):
central uh through your agent,as example, and because of it,
you can simplify processes.
There's many points, uh testing,testing, you can generate tests
uh for A and before, and I'm notI'm not sure if you're aware, I
think you're aware, but beforeuh if a customer asks you to
(18:08):
write some or to build someproject, and if you will try to
try to mention tests and youwill say some numbers, they will
mostly say no because I don'tcustomers don't see value in
tests, direct value.
There's value, of course, butthey don't see it as direct
value.
But now with IA you can generatetests very easily, and and you
(18:29):
know and not just you can, but Ithink you have to generate
tests.
I think test is now uh standardfor uh IL.
We all need to generate tests,we all need to understand a
little bit how it works, butmostly we can delegate this.
And uh I like the idea that Ilistened or heard on uh LinkedIn
(18:52):
from Tiny or from uh Vecco thatTDD is a great choice for L.
Like we can start from tests tobe sure tests are correct, and
after only after that go todevelop something.
SPEAKER_01 (19:08):
You you hit on a number of key points in that
whole conversation.
SPEAKER_04 (19:12):
It's yeah, it's uh just in general, but we can
focus on something specifically.
SPEAKER_01 (19:17):
We'll get into a lot of those because uh as with many
uh individuals that we speakwith, I hear some of the things
that you're saying, and I justget excited, and my mind goes
deeper into it because I spend alot of time now doing this.
I uh I have some questions foryou, but I want to make some
comments on.
We can start with the tail goingforward here or backwards.
(19:41):
Testing, I agree with you.
Customers used to not see thevalue of tests, but I don't know
if it was necessary to tell themthat we were doing the test or
if they should have just beenincluded in the project.
Thankfully, now with AI and therate of development that we can
(20:01):
make, those tests can now be apart of it.
Uh, it's interesting abouttests.
I did some experiments the otherday with test TDD, I've been a
fan of for a long time because Ifound it also helps architect
the code better because now itforced you to modulize the code
to have small tests in theprocess.
(20:24):
And by doing it now, AI can helpyou do the tests and also plan
the development for it from thetest.
But an interesting thing, uh uhjust had a conversation with
someone this morning aboutcreating tests, and now the same
thing.
Everybody's starting to saytests are required, and not only
having tests, but proper tests,right?
(20:44):
The tests need to fail properly,the tests need to pass properly,
not just pass fail.
You need to make sure it'sreally feeling failing for the
right reason.
But I went back to an oldproject that had few tests
because sometimes it's notalways easy to come up with all
the scenarios that you want totest for, to be honest with you,
because users will do thingsthat you may not think of.
(21:06):
So I asked a you know, the AImodel or mentioned to the AI
model or the framework, excuseme, are there any other tests
that I should test for?
I'm simplifying it.
And it went through and itcreated or suggested, I was
planning at this point, so manytests that I didn't even think
(21:27):
of, but were valid tests,meaning the code should pass
these tests.
I then had it again, I hadexisting tests.
SPEAKER_03 (21:38):
So this isn't just is this against an old project
that you did this against?
Okay.
SPEAKER_01 (21:43):
Against an old project that had some tests in
there already.
So it had my I have aninstruction file that we're
using now for uh modeling how wewant the tests to be created.
But I said, go back and can yousuggest some tests?
It created a whole List oftests, and then the question is
here's the plan, this is whatI'm going to do, and we all see
(22:05):
it.
Now that I have the plan, do youwant me to go ahead and do this?
Right?
I said yes.
It went through and created allof those tests, and it made one
mistake for the set of probably10 new tests that it made.
SPEAKER_04 (22:19):
It's a suppressive, right?
SPEAKER_01 (22:21):
Followed the structure because I do this in I
follow the Garrickin scenario,the you know, the g the the
scenario, the give and the win,the then.
And it created, I had anexisting library code unit with
certain functions, and then Ihad tests broken down by area.
It did it.
SPEAKER_03 (22:39):
And I was is this a project that's already live that
you went back and ran the test?
SPEAKER_01 (22:46):
Yes.
It's an extension that'screative.
So we had some tests and it wentthrough and it created the tests
for the code.
And it just goes to show, as youhad mentioned, there is no
(23:08):
reason now to not have thetests.
And now also um, what I foundwith testings from a previous
scenario is code was introducedto a project that if we had not
had tests for it, we would nothave identified a problem that
(23:30):
it created.
Right?
So now you have extensions, youhave code that works.
If you don't have tests for it,if somebody wants to add
functionality to it, you mayinadvertently, because you're
not aware of the flow, and thisis where I think you would get
into it becomes more importantbecause now AI is just creating
so much code that it's very easyto break something else that's
(23:56):
auxiliary to uh your project.
SPEAKER_03 (23:59):
So I think it goes back to the value component
because I see that you went backto test an old project that's
already in production.
A end user or client would notsee that as like a direct value.
Obviously, it's a value to youas a developer because now
you're like saving this clientfrom future headaches if someone
(24:21):
was to extend it.
But I think that's the missingcomponents.
Like uh back in the day, whetheryou test it or not, you create
test scenarios, you you uh theclients imply that you're
already doing that.
Like you've it's like you shouldbe doing that, right?
But there's different types oftesting.
SPEAKER_01 (24:39):
And I think it's it's too much to when I I think
I think testing's important'salways been an important, I
think it's even more importantnow.
Not again, I'm I'm nottrivializing how important it
was before.
I I've had years of sessions atconferences on creating tests.
But I think telling some, I Ithink saying I'm developing an
(25:03):
app and part of this is creatingtests, part of it's creating the
documentation, part of it'screating this.
To a customer, it's hard toassess the value because they
don't know the value.
They the tests are going to savethem in the future from
mistakes.
But to your point, Chris, Iwould assume that if you're
going to deliver something tome, it should work.
(25:24):
Why why why should I have to saytest?
So it's more of here's whatwe're doing, this is what the
cost will be, is it of value toyou?
And I'm going to give youdocumentation which AI can help
create.
The testing shouldn't even beit's almost like saying I'm
coming over to your house,Chris, but I'm going to walk
(25:46):
over there.
Like, does it is it really evenmatter?
It just matters that I'm goingto get there.
SPEAKER_04 (25:50):
From my experience, customers some sometimes try to
uh have more detailed numbers,like they try to understand uh
why you have numbers, why youhave this amount of foul
spending.
And and if you will say I spentsome amount of files to test in
the past without IA, they willbe not so happy, you know.
SPEAKER_01 (26:12):
No, I understand.
I I've lived it and I understandit, but it's almost a matter of
somebody needs to test itsomewhere, whether it's
automated testing or developertesting or user testing.
No one should implementsomething without testing.
SPEAKER_04 (26:30):
Yeah, yes.
My favorite type of testing isproduction testing, right?
SPEAKER_01 (26:36):
There are many implementations where that was
their model.
They they assumed that thedeveloper uh developed the
extension properly and they justthrew it in, and then they get
mad when nobody tested it.
And it's it's uh interesting.
So I mean, so the testing isextremely helpful, and I think
(26:58):
now it's even more important,but there is an offset because
of the speed at which we cancreate code, we now can include
more such as testing, extensiondocumentation, and those
infamous guides that nobody usedto write.
SPEAKER_04 (27:14):
Yeah, and I think together with it, it actually
means you are not losing qualityor almost not losing quality,
because on top of uh fastgeneration manual line of code,
you have tests, you havedocumentations which you don't
have before in the past withwithout artificial intelligence.
So actually, I think uh qualitymay be even better a little bit
(27:38):
if you have processes in correctway configured, you have CI CD,
you follow all best practicewith IA, with working with IA,
it actually can even uh increasequality.
SPEAKER_01 (27:50):
Yes, absolutely.
So another thing you went backto talking about was the amount
of code that we can create umwith AI.
And testing, just to come backand loop up a testing, testing
is not just extensions thatpartners may deliver or
contractors may deliver.
If you're a developer for acustomer, some some more
(28:13):
implementations have their owninternal team that can create
extensions, they should alsotest as well.
The testing is not specific todeliverables from partners or
outside agencies.
Even if you're workinginternally, you should take the
time to create those tests foryourselves and you get better
quality.
And as I had mentioned before,you can make sure that any
(28:37):
changes or additions that youmake won't impact your existing
workflow or your existing code.
Uh even to the point where ifyou have multiple extensions, I
always say run your test for allof the extensions as well.
Because I have found, and thisis uh where we jump into AI
(28:58):
assisted development, where haveyou found AI to be helpful or
not so helpful with development?
Have there been any situationswhere you can say that AI for
sure saved you a lot of time?
And other cases where you saidAI probably created more work
(29:23):
than it saved?
SPEAKER_04 (29:25):
Good question.
Actually, I have severalexamples where I have many
examples where it's helped me tosave my time.
For example, when you writetests, as you mentioned, it's
actually a good uh point to showwhere tests are save you time
time dramatically.
Um in my experience, but it'smore interesting to to to speak
(29:48):
about where it's actually uhcreate some problems for you or
it can take your time.
Uh and first of all, uh it cancreate you some uh difficulties
when you uh collaborate withother people who use IA, but
they don't uh understandproperly how to properly use it.
(30:09):
They are not using um uhspecific uh uh best practices
with agents, they're not uhunderstanding how it works, they
think it is like a magic and uhthey don't provide enough
context to agents, and uh thequality of this work can be uh
less than you expect.
And as open source maintainer,uh I realize that sometimes
(30:34):
people come to uh repository, tomy repository of my open source
project, and they try to uhcreate pull requests, uh
contributions to uh to my uhthis like for example that
editor tool, and they generateit via agent without proper
specified context of problem,without proper specifying uh
(30:56):
what is a problem about, what isit we you try to solve, and
agents generate something whichcan be like matched.
And I heard that many uh opensource uh maintainers encounter
issues that they are overwhelmedwith number of pull requests,
with number of contributionswhich people try to bring to uh
projects uh because of IA,because I just generated, and
(31:20):
they are not understand how AIworks, and it creates some
overload without um necessary uhpurpose.
It's one example where it can bea problem.
SPEAKER_01 (31:32):
Yes, yes.
Uh I I've heard that as wellwith open source projects that a
lot of AI pull requests arebeing made, and it's difficult
for the maintainers to keep upwith those changes to review
them properly.
It's interest, it's it's it's aninteresting time that we live
in.
The stepping back into ALdevelopment with AI and some
(31:58):
challenges that you may havehad, or any suggestions for
someone that may be working withAI and AL development uh when
they're creating extensions.
What are some things that youhave found, whether it's working
with ISVs and how AI could uhhelp with that, or any other
(32:22):
places where you notice somechallenges using AI?
SPEAKER_04 (32:26):
Yeah.
I see some I got some patternsfrom my experience when I work
with AI, how I use it, how I tryto get most possible value from
it.
Uh and first of all, I think youalways have to start with
planning mode.
Like if you um have some problemor you would like to build
something new, uh like totallyentirely new application, do not
(32:50):
start quote or ask uh uh agentto quote for you.
Focus on plan.
Ask agent to plan.
Like give all possible contextwhich you can give, like what
you can know about your problemor about idea of what you would
like to implement.
Give proper context to model.
(33:10):
Try to uh highlight whatspecific if it's some
modification of some existapplication in uh ISV of a PGA
extension, highlight specificobjects which you think are um
related or can be related tothis problem, highlight why you
think about it.
Try to communicate with agent asyou are communicate with uh
(33:31):
person, like try to understandthat it can magically understand
uh get all information if youare not specified.
It can, but it will lose contextwindow, and quality of uh
generation will be lower andlower because context management
is very important when you workwith agents.
So you try to provide as lessbut as much as possible uh
(33:53):
context because it's actuallylike you're always fighting with
uh two different uh problems.
Like one problem you provide notenough context, so agent cannot
generate something for you.
Another one you provide too muchuh context and it's like
overwhelmed and it's unfoundedon uh your specific idea and
task.
SPEAKER_01 (34:11):
So that's a key point I've I've heard from
people.
It's less is more.
Don't get into please thank youand keep adding additional words
or even try to tell it what todo, in a sense.
If you the more boundaries thatyou give it, it seems to have
(34:32):
more of a challenge with theoutput.
SPEAKER_04 (34:36):
I agree.
And when you set someboundaries, sometimes if you're
very sure about it, it can beuseful.
But many times I realize whenyou set some boundaries, like
you try to lead agent in somespecific way, but you can be
sure that your way is the bestway to do things.
Like you have to learn from itas well, and try to try to look
(34:56):
what it suggests, and then youwill see all possible uh
branches of solutions how youcan solve the problem.
Uh you can like choose the bestone.
Um so plan mode, contextmanagement, uh session
management.
Of course, you don't say I I I Isaw this mistake very often in
(35:17):
the past.
People uh speak with agents inone session, they're not pay
attention on context window, anduh agent just overwhelmed with
all different specific taskswhich uh people try to manage in
one session and screen problem.
Start new sessions, try tomanage it.
Um what else?
Also, it's important to chooseright tool because there's many
(35:41):
benchmarks in the internet whichthey specify, like we are best
agent or we are best model forspecific agent.
Uh it's hard to uh predict whatis better for AL development on
those benches because they aremostly based on C sharp, Python,
on other languages.
But uh recently Microsoftreleased a BC uh bench uh for
(36:06):
agents and they uh it was as Iremember they only count two
agents, quote quote, and maybeuh codex, not sure about codex
and it's one point when you canlook on it.
Try to find IL benches tounderstand it, but actually what
I will uh give one small adviceto everyone who is interested in
(36:27):
it is try to yours try to use itby yourself, try to use
different tools.
Uh because if if it's somethinggood for one person, it can be
not good for you.
You have to choose, you have totry, and you have to spend time
on it, and and it will give alot of value uh later when you
will realize what tool isbetter.
(36:48):
Because there's like cursor, uhcodex, cloud code, uh uh
copilot, all different toolswhat you have.
But my honest answer, try it,and you will see the difference.
Try to compare it by yourself,by your internal tickets.
I even have some small internalum tasks, and I try to uh
(37:12):
evaluate those tasks every threemonths or something like that.
Six three months uh on all uhpopular agents or clients which
I can use, and I can see thesituation is changing.
Like six months ago, codex wasbetter, six months ago, compilot
was better, and so on and so on.
It's constantly changingsituation.
(37:32):
I prefer Clot Code, I even uhwrote about it uh several times.
Um but I'm not sure if it willbe forever.
Like maybe tomorrow it will bebetter, something from Google,
who knows?
SPEAKER_01 (37:46):
Like so, one of the things as you're going through
your development workflow withAL development is not all models
or frameworks work the same, andwhat may be good for one person
for one task may yield differentresults for somebody with a
(38:06):
different task.
SPEAKER_04 (38:08):
Yeah, and also it uh depends on how you communicate
with model.
Every parse person has a uniquestyle of communication, right?
And it also affects uh how uhhow what output you will get
from agent from models.
SPEAKER_01 (38:22):
It's a challenge to keep up with.
SPEAKER_03 (38:24):
For the both of you, so how does that work when you
have uh when you're working on aproject together and you have
maybe three of you working on asolution, and then they may have
someone that's a third party,outsource whatever, uh that a
client hires.
Do you all use different uhframework?
(38:47):
And how does how do you maintainthat?
Like how does that work?
What's the workflow for that?
SPEAKER_04 (38:52):
Uh from my perspective, I have such
experience um very often thatpeople use different tools and
agents.
They uh one developer usescodex, one developer uses
copilot, uh, one developer useuh copilot inside Visual Studio
Code, one is using copilot uh asuh command line, right?
(39:15):
Technically, it's not soimportant that people use
different tools because the mainthings to get the best output
from it.
Uh I think like even if you areusing two uh different tools,
but it gives you good output, itdoesn't matter what tool
(39:36):
created.
But how you uh communicate withdifferent agents uh and your
experience, it can affectquality of work what is uh will
be as output from agent.
But at the end, it doesn'tmatter what was by what tool it
was generated.
(39:57):
That's what I see.
SPEAKER_01 (39:59):
Yeah, I agree with you.
I I uh it's a it's a goodquestion, Chris, because it's
not to confuse the tool thatsomebody's using to create
something, it's what they createis the important thing.
As Old Me had mentioned, it'sthe output of the quality.
What becomes important in thosesituations is not the focus of
(40:23):
which AI tool they used or whichedit IDE they used, it's what
they did and how you manage it.
How you manage your developmentworkflow with source control
management, with branching, withmerging, with code review, those
are the things that becomeimportant.
So you can have an internal teamthat has multiple developers,
(40:45):
and then you may need to haveanother person also contribute,
they still can fall into yourworkflow.
What they use to create it's notimportant.
So even if you are a customer ofor a user of Business Central,
excuse me, and you're managingyour own source code and your
(41:06):
own uh repository and workflow,you can bring in others and have
them fall into your workflow.
You just need to have thediscipline of how to manage the
different branches.
SPEAKER_03 (41:17):
Yeah, I think I think that's a good point
because I think down the roadthere's going to be, I feel as
though there's going to be moreof that as a lot of clients uh
understand that you can havemultiple partners working on a
solution and as people vibecode, you know, someone has to
manage that.
And then at some point, I mean,you're right, it doesn't matter
(41:39):
what tool they use as long asthe output is the the the right
expectation or the correctexpectation of the output.
Um but at the same time you mayhave some balance of like if
you're using a specificframework, then your testing
scenarios may produce differentthings or different testing
scenarios versus someone thatjust you know uh committed their
(42:00):
their code into.
So it's it's an interestingbalance of what that looks like
in the near future, uh,especially when your clients may
want to be more involved inthat.
SPEAKER_04 (42:10):
Uh because you know, actually, yeah, it's uh great
points.
And uh CICD is what can help usto manage all this uh mass
workflow.
Uh we just can um try toautomate as much as many process
as we can.
Automate uh test running,automate uh linters, uh automate
(42:30):
uh compilation of projects,automate everything of it, and
it creates and it starts to beeven more it was important
before, before Artisan Terenceuh was starts to be more
popular, but uh it's importantnow even more, same as testing
all the CICDE processes, and uhthat's why we all need to look
on LGO, how it works, how we canimplement it in our projects,
(42:55):
and it's actually for free.
You can implement it for your uhopen source project and for PG
extension.
Go ahead, use it.
And I actually very um I almostall my projects that I work with
customers.
I always try to say we need toimplement CICD, we need to
implement LGO in uh thisspecific project, and I
(43:16):
explained why, like uh like whatvalue it brings to customers,
and mostly most of the time theyare actually supported because
they see it's actually very easyto show them uh value of
workflows of how automationshelp.
Uh for example, from to save uhproject from regression bugs or
(43:39):
problems, a lot of things can beum automated, and but I think
it's main tool also what we canfocus as developers of L
developers and to using ALGO, tousing uh CI CD processes.
SPEAKER_01 (43:55):
Yeah you brought up a good point.
ALGO is It is a free tool thatruns in GitHub.
There could be a cost to therunners that you use, depending
on how you set that up for yourprojects, just to put that in
there.
There is a tool for it.
But I I I literally had aconversation with someone this
morning about that inimplementing these on projects,
(44:18):
that they were having problemswith developers building the
workflow is to have a branch,review the branch, bring it into
the merge it into the mainbranch, as we'll call it, and
then that gets deployed.
Whereas if you have a developerthat didn't update their main
(44:39):
branch with other code changesfrom other branches, and they
just did it with their branchmerged in and then uh built the
application and deployed it, youin essence can remove changes
that were made.
So by using tools such as ALGOand GitHub, the extreme benefit
is you have the security ofknowing whatever was built, the
(45:03):
artifact that gets built as partof your CI CD is what should be
deployed.
You don't have to auto-deployit, but that's where whomever is
going to load it.
And I'll be the first to say isI don't think developers should
be the ones responsible forloading the changes that they
make.
You should have someone whoknows how to pull down the
artifact, and then that's whatgets loaded.
But that just verifies that thetests were run because the AL Go
(45:25):
will run the tests that arewithin the extension.
And it also validates thatanything that's in the main
branch is what gets built, used,and deployed, and you're not
relying on somebody refreshingtheir branches or refreshing the
changes of the code that theyhave because using Git you have
a local copy of it.
Yes, it does sync, you know, youdo pull and push from the
(45:48):
central location for it, but youstill have to go through that
exercise of doing that.
Yes, it can be automated, yes,you can do it, but some people
still like to do it manually.
So the point that you had madeis using tools like that in this
world of AI development to meand team development becomes
even more important.
Not that it was less importantyesterday, it's more important
(46:09):
today.
So you these are some goodthings.
I know we're talking about AIand AL development and open
source, but um so it with yourAL development and AI, what
tools do you find useful to helpyou with it?
Are there any other tools thatyou have found can assist you in
(46:30):
this journey, or even uh ifsomebody wants to start using
it?
You had mentioned a little bitago, just start using the tools,
right?
Just start using AI for yourprojects.
What are some things that youfound were helpful to get you to
where you are with yourdevelopment journey?
SPEAKER_04 (46:48):
Um, yeah, there's several tools which I uh think
is important uh to um to startwith or to learn or to look on
it uh a little bit.
Uh first of all, uh I think it'sa good point to get some basics.
Uh because if you look on uhdifferent tools, as example,
like uh agents in common line,uh they all have some different
(47:13):
same uh similar uh technologiesbehind like MCP.
What is MCP like if you try tounderstand what is MCP it can
help you to uh get moreefficient with any agent?
Like you are not focused onspecific tool.
Uh so I will start.
I think it's better to startwith some basics, like to
(47:34):
understand this MCP.
Like, okay, MCP is some protocolwhich will uh allow my uh IA to
uh communicate and to changesomething on external services.
It can be business central, itcan be your browser MCP, it can
run uh for you some uh businesscentral um page and even try to
click some uh buttons directlyin uh your um browser if you
(47:59):
need it.
And all these are done by viaMCPs.
So start from basics tounderstand the context windows,
MCPs, but speaking aboutspecific tools, yeah, I have
several in my mind.
So I started uh originally withuh um cursor, if you heard about
it.
Uh it's like uh form of VisualStudio Code, but with
(48:23):
implemented um models of AI.
But it was like in the beginningof the uh my journey, and um I
realized that now you are notneeded uh cursor because
everything works inside VisualStudio Code.
Uh naturally you can have uhseveral different models.
(48:45):
So now I think it's good tostart from Visual Studio Code.
It's our daily uh EDE where weas L developers work every day.
So it may be uh good startingpoint to use uh copilot inside
Visual Studio Code to selectsome models.
You can select different models.
You are um you can select uhanthropic models, you can try
(49:06):
open eye models.
Try to mix it, try to to startwith EDE.
I think ED is a good startingpoint.
Uh next, what I uh think isworth to try after you focus on
some Visual Studio Code.
Um I think it's next this willbe useful is uh common light
(49:27):
agents.
They are independent from ED,you can run them inside your
terminal.
And there is some value behindit.
First of all, you can run iteverything there.
Next, they are working this filesystem differently than ED
(49:49):
works.
It works uh per my experiencemore efficiently with um all
these uh commands which you canexecute inside your uh terminal,
like bash command, uh PowerShellcommand.
It's very easy to automate, andyou as a developer can easily
handle it.
Um so try some command-lineagents, and there are several
(50:11):
popular agents on the market orin general.
There's like uh compiler agentas well, there's uh codex from
OpenAA, and there's uh CloudCode uh from Anthropic.
Uh I try to use them and try tomix a little bit to understand
what tool is better now to whatis better than now because it's
(50:34):
changing dramatically, as Imentioned.
So try with Visual Studio codeED agent inside it, try to um
read the basics.
What is MCP, what is contextwindow, what is uh slash
comments, or uh uh what is um MDfile skills, it's it's actually
(50:55):
called skills now.
Before it was slash comments,now it's skills.
Uh how skills can affect your umgeneration, agent generation.
Try to next do next steps to usesome MCPs actually to on
practice.
Next steps try to use commentline agents, try to compare
different agents to see what uhvalue uh you can get from it.
(51:19):
I think uh also try businesscentral MCP, it's also a good
point to check uh how it works,how data is flowing, how you can
just tell to agent, please showme or delete customer numbers.
Of course, it's not forproduction, you just do your
daily work on some sandboxes oron some uh uh Docker
(51:40):
environment.
So I think those like mainentrance points that you can go.
SPEAKER_01 (51:50):
It does sound like there is a lot, but as with
anything else, it sounds worsethan it is.
And I think once you startworking with it, it all comes
into play.
And I think the important partis to start using it and not try
to know everything, just knowwhat you need to know to get
(52:11):
kind of work your way throughthe task and see what you need
to do for that task to to be uhuh be successful with it.
SPEAKER_04 (52:21):
Yeah, and I I can tell you practice and practice.
Like if you just will readtheory, it's will not be so
useful.
Try to do it in practice.
I realize by myself that when Ido practice, I um start to
understand things better.
But when you just read theory,yeah, you know MCP, but try to
it in practice, and you willrealize uh actually how it can
(52:44):
be useful for you and how it canuh give you some output,
valuable output.
SPEAKER_01 (52:53):
It's it's um yeah, it's a strange world now with
all of this output and uhbusiness central developments
coming along, and the businesscentral tools that they keep
adding are also nice to have,and it's nice to see that
they're adding all these toolsto assist with the development
point of view.
(53:13):
And in my opinion, it's alsoreinforcing that the days of
just development are gone.
I think that developers need tohave a good understanding of the
application and the function ofthe application so that they can
work with AI to develop thesesolutions.
(53:34):
And also, as you had mentionedat the beginning, the importance
of still understanding and theability to know the architecture
and the code.
So it's not it's not one ofthose things that, oh, AI, I can
create this extension, it'sgoing to do this thing.
I think you do need to have asense of understanding and
(53:56):
review how to review what getscreated.
Uh, I think I keep sayingbusiness central to me is a
little bit different becauseit's in a box.
It's a predefined applicationthat you need to work with
within the constraints of howthe process and the application
works.
We can't just change a completeapplication, every build is a
(54:18):
standalone application, it hasto work within a framework.
So it's important to be able tounderstand that framework so you
can review what gets generatedand then deliver something of
quality.
SPEAKER_04 (54:31):
It's is that I actually agree, and you bring a
great point.
Like as business-centereddevelopers, we are not just
developers like um Pythondevelopers who just know
language and know all these umkeywords and how to do and write
some algorithms, but withbusiness centered as
business-centered developers, weare also should and we actually
(54:54):
should really and must know thebasic functionality of business
center, how it actually worksright now, how it actually works
from a functionalityperspective, not just from code
how it's coded.
It's one part, and butunderstanding how it works for
different types of businesses,how you can configure it, it's
(55:14):
actually super important in uhany era, even before I it's also
was super important.
Understanding of how it works,it's actually because if you
without this understanding, youcan build efficient tools, you
can build efficient uh uhsolutions for your customers
because you will not aware howto handle it in the best way.
(55:36):
That uh you don't need to quoteanything what you hear what the
customer asks you.
You need to think, okay, is thisalready a functionality existing
business central?
Can you just implement just useit without coding anything else?
Because I saw in my experience,many poor developers try to
build something immediately whensomeone asks them, like, please
(55:58):
create me this and this.
No, try to think if this isactually required to build, if
it's actually maybe already apart of standard business
central library, maybe you'renot aware of it.
Try to do some research beforeyou say something.
Try to think.
And yeah, so I thinkunderstanding of functionality
is important, and now with AI,it's even more important because
now we have more time forarchitecture and also more time
(56:23):
for understanding functionalityof business central as L
developers.
Like now you cannot say, Oh, I'mjust code.
Noah, it's not on, man.
SPEAKER_01 (56:31):
You hit that well, and you answered the question,
and I think it's a great way, Ihate to say it, it's a great way
to wrap it up.
What do you do with all thatfree time?
That's the question thateverybody's saying is okay, with
all this free time, I don't havefree time.
Don't look at it as you havefree time to go uh you know, not
(56:53):
work as much.
Now you have free time to, asyou had mentioned, learn the
application.
You have more time to understandthe architect, the application,
more time to architect and plana solution.
So it's not necessarily that wehave all this free time.
I can get work done in twominutes and now I'm done for the
(57:14):
day.
It's I can deliver betterquality work.
And this is where that questiongets asked.
What do you do with the time?
You take the extra time to addthat extra quality to it.
And again, not to say thatquality wasn't important before.
I'm not trying to discount it,but now you can deliver a better
(57:34):
quality product in a shorterperiod of time.
That to me is the answer to allof this.
It's better quality in a shorterperiod of time.
SPEAKER_04 (57:46):
I fully agree.
SPEAKER_03 (57:49):
That's a good one.
That was a good, good with that.
SPEAKER_01 (57:53):
I think uh would like to thank you for taking the
time to speak with us today.
I could talk about this stuff,and I'll probably now I'll just
have to put you on my list ofcalls to have that I could talk
about this stuff all day, everyday.
SPEAKER_04 (58:03):
Because we can spend days and hours to speak about
it.
SPEAKER_01 (58:06):
It's actually board topics like oh yes, and it's uh
you're in some of the same chatgroups we're in, and we get to
talk about this a lot, and uh,you know, it's the wonderful
chat groups because I'm in theUnited States, most of you are
in Europe.
I can wake up in the morning andthere's like 65, 70 messages I
have to read through and screenthrough.
(58:28):
And it's just like you know,really what's going on.
But it's it's it's a it's anexciting time for everybody.
And again, thank you very muchfor taking the time to speak
with us today and sharing yourvaluable insights.
I think anybody who's in the ALjourney, uh, we're all in the
beginning of this journey.
Anybody who says they're anexpert on it, I mean, how can
(58:49):
you be an expert fortechnologies that's so new and
it's changing?
So anybody that's uh currentlyusing AI, hopefully that uh
thank you for sharing so manyinsights.
Somebody that's new to it, youalso shared some insights for
that as well.
And if anybody would like totalk to you more to learn about
your open source projects ormaybe any other AI for AL
(59:10):
development tips that you canshare with them, what's the best
way to contact you?
SPEAKER_04 (59:15):
Sure, um there are several ways.
Um you can have just uh commentand put a comment in uh my blog.
I have a blog uh where I uh tryto share my um knowledge, what I
learned from uh my experienceabout development specifically
for business center development,also some some IA insights and
tools.
Uh you can comment it.
(59:36):
You can contact on LinkedIn ifyou want to ask anything.
I'm very open-minded.
If you just just ask me onLinkedIn, I will immediately
answer you, or when I will havetime, of course.
Um, so LinkedIn, my blog, I havealso um uh Blue Sky.
Anything of it will be okay tocontact me.
SPEAKER_01 (59:59):
Uh great, thank you.
And we'll have your contactinformation with the show as
well.
Uh again, thank you for takingthe time to speak with us.
Look forward to speaking withyou again soon.
Uh enjoy the rest of your dayand have a great weekend.
Talk to you soon.
Chao ciao.
SPEAKER_04 (01:00:12):
Take care.
SPEAKER_01 (01:00:13):
Bye.
SPEAKER_00 (01:00:15):
Thank you, Chris, for your time for another
episode of In the DynamicsCorner Chair, and thank you to
our guests for participating.
SPEAKER_03 (01:00:23):
Thank you, Brad, for your time.
It is a wonderful episode ofDynamics Corner Chair.
I would also like to thank ourguests for joining us.
Thank you for all of ourlisteners tuning in as well.
You can find Brad atdeveloperlife.com.
That is D V L P R L I F E dotcom.
(01:00:43):
And you can interact with themvia Twitter, D V L P R L I F E.
You can also find me atmattalino.io, m-a-t-a-l-in-o.io,
and my Twitter handle ismattalino16.
And see you can see those linksdown below in the show notes.
(01:01:06):
Again, thank you everyone.
Thank you, and take care.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
The Bobby Bones Show
Listen to 'The Bobby Bones Show' by downloading the daily full replay.