November 16, 2024 • 29 mins
In this episode, Bernhard Ritz (CEO & Founder, Ainzel) and Andreas Welsch discuss supercharging your business processes with collaborative AI agents. Bernhard shares his insights on introducing AI agents into business functions and provides valuable advice for listeners looking to elevate their business operations with the help of a virtual team of expert agents.
Key topics:
- Describe the opportunity to bring AI agents into your business processes
- Determine the challenges of bringing AI agents to human-AI teams
- Quantify the value of collaborative AI agents for business functions
- Recommend how to start exploring AI agents
Listen to the full episode to hear how you can:
- Take a business look at AI and re-imagining your business processes
- Learn to work with AI agent frameworks
- Define management as a function spanning humans and AI agents
- Articulate limitations and guardrails for using AI agents
Watch this episode on YouTube:
https://youtu.be/FnkeZMSIbHI
Questions or suggestions? Send me a Text Message.
***********
Disclaimer: Views are the participants’ own and do not represent those of any participant’s past, present, or future employers. Participation in this event is independent of any potential business relationship (past, present, or future) between the participants or between their employers.
Level up your AI Leadership game with the AI Leadership Handbook:
https://www.aileadershiphandbook.com
More details:
https://www.intelligence-briefing.com
All episodes:
https://www.intelligence-briefing.com/podcast
Get a weekly thought-provoking post in your inbox:
https://www.intelligence-briefing.com/newsletter
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Andreas Welsch (00:00):
Today, we'll talk about supercharging your
business processes withcollaborative AI agents.
And who better to talk about itthan someone who's actively
working on that.
Bernhard Ritz.
Hey Bernhard, thank you so muchfor joining.
Bernhard Ritz (00:11):
Hello Andreas, thanks for having me on the
show.
Andreas Welsch (00:14):
Awesome.
Hey, why don't you tell ouraudience a little bit about
yourself, who you are and whatyou do?
Bernhard Ritz (00:18):
Sure.
I'm the CEO and founder ofAinzel, which is an enterprise
AI platform that adds virtualemployees, AI agents to
businesses.
I worked for 22 years for SAP.
A couple of years in Germany,later here in the US.
And then I joined an SAP partnerto build up their North America
business.
And around 18 months ago, Istarted Ainzel, so I started my
(00:41):
own company.
Andreas Welsch (00:42):
That's awesome.
I'm really looking forward toour conversation today.
You've shown me Ainzel a coupleof weeks ago, and I was so
excited to see thesecollaborative agents and what it
actually means.
Looking forward to ourconversation.
And also don't forget to pick upthe AI Leadership Handbook on
AILeadershipHandbook.
com.
(01:03):
So you can learn how you cansuccessfully implement AI in
your business.
Because it's not just all abouttech.
With all of that out of the way,What do you say?
Should we play a little game tokick things off?
Bernhard Ritz (01:13):
Okay.
Andreas Welsch (01:15):
Wonderful.
So hang on.
This one is called In Your OwnWords.
And when I hit the buzzer, thewheels will start spinning.
When they stop, you'll see asentence and I'd like for you to
answer with the first thing thatcomes to mind and why.
In your own words.
To make it a little moreinteresting, you only have 60
seconds for your answer.
And for those of you who arewatching us live, drop your
(01:36):
answer in the chat and why too.
So are you ready for What's theBUZZ?
Bernhard Ritz (01:41):
Let's go.
Andreas Welsch (01:41):
Okay, here we go.
If AI were a movie, what wouldit be?
60 seconds on the clock.
Bernhard Ritz (01:49):
Movie, that's an easy one.
It's Inside Out.
You probably know the Disneymovie, where these little
figures are thinking about you,your different emotions and so
on.
So that's a great parallel thingto, AI agents.
If we had the point where wehave them reasoning and think,
(02:11):
so all these little differentaspects of your life together,
collaborating, so perfectcomparison to AI agents.
Andreas Welsch (02:20):
I love that.
That's wonderful.
Definitely on message and themefor the show.
let's jump right into it, right?
You already mentioned agentsworking together, but maybe even
before we talk about themcollaborating, what does it even
mean to bring agents into abusiness process?
(02:40):
I think agents have been the hottopic this summer.
But what does it actually meanin concrete terms when you do
that?
Bernhard Ritz (02:49):
The way I look at this is, and a lot of people ask
me and say, okay, AI agents,there's a lot of technology
involved.
And I tell them, step back,don't look at the technology,
look at what they do for thebusiness, these AI agents.
And I always compare it withvirtual employees, which is
scary for some people.
Some people say, Oh no, they'rejust augmenting, they're not
(03:09):
real employees and so on.
But step by step AI agentsworking towards this idea of.
being part of the team, being amember.
Now there are these littlevirtual helpers that are built
on generative AI.
They started with very simpletasks can you improve this
email?
And then it went to, can youwrite this email?
(03:31):
Can you run this email campaign?
Can you manage my email inboxfor me?
So step by step these littlehelpers got better and step by
step we're integrating themdeeper into our business
processes.
As I just described we use themto augment what we are doing and
(03:51):
now they are starting to takeover single activities, single
steps.
They're very knowledgeable, theyare trained in certain domains
at an expert level.
They are fast, they respond withinformation in seconds that
would take us minutes and hoursto compile.
So very helpful, affordable,knowledgeable trained little
(04:16):
helpers.
And this is actually how thename ANZL came along.
I thought, how can you namethese little helpers?
And in Germany, there issomething called
Mainzelmaennchen, which is aGerman TV show.
And I tried to find a domainname that fits to
Mainzelmaennchen, these littlehelpers, and I came up with
Ainzl.
And later recognized, boop,there is AI at the beginning of
it.
Yeah.
Andreas Welsch (04:37):
I love that analogy.
I've been wondering, too.
Now, when you introduce more ofthis autonomy, into your
business process.
I'm sure that there's somechallenges that you will
introduce with that as well,right?
I think in business, it'salready tough enough if you have
a team of human workers andemployees, getting them to
(04:59):
communicate, getting them tocollaborate and agree on
terminology and what is the goaland how do we break this down
while still benefiting fromeverybody's individual
experience and background.
How do you see this evolving inbusiness when we introduce now.
AI agents as a softwarecomponent that can do some of
those little more complex tasksthan just if then else at this
(05:22):
time.
Bernhard Ritz (05:22):
You brought up an interesting point and this is
language.
I think the beauty of largelanguage models and the
underlying vector databasesalso, they bring a lot of
flexibility.
To things that AI agents canreact to.
If you think about traditionalapplications, right?
You have databases and they havestrict fields and you need to
(05:43):
provide the information in theexact format required so that it
can handle it.
With AI there's ontologies,there's vector databases that do
similarity searches, things likethat.
So you can have a very flexibleway of entering information and
the AI agent is still able to,handle it.
So I think that's a greatadvantage.
(06:04):
You said autonomy, obviouslythat's the big point.
The more the AI agents take overin terms of functionality and,
parts of the business processes,the more autonomy they will get
to do things independent.
And this is where also the riskis, where the oversight needs to
kick in.
About hallucinations, aboutbias, about privacy, security,
(06:31):
compliance, I said they arevirtual employees.
And if you play that model, thenyou see virtual employees, they
have an NDA, they have anemployee contract, they need to
follow code of business conduct.
So for the AI agents, we need todesign them the same way.
And I think this is also thedifference from enterprise AI
platform, which takes that veryserious in comparison if you
(06:53):
feed information into a largelanguage model and it's working
with you, but this oversightcomponent that you expect in an
enterprise platform is probablynot there.
Information goes out there,information is shared that
shouldn't be there.
So enterprise platform dealingwith these AI agents and the
collaboration between theagents, is the go to solution in
(07:16):
this case.
Andreas Welsch (07:17):
That's, awesome that you mentioned this because
I think a lot of times that partis, underestimated or maybe not
even covered.
So I was thinking about that alittle more the other week after
our conversation and I realized,for example, in a lot of
organizations, HR is actuallyabsent in these discussions,
right?
It's technologists defining whatshould these new agents, these
(07:40):
new digital workers look likeand how should they behave?
And what are the policies wewant to need them to abide by?
So I think there's anotheropportunity for technology
leaders to bring in.
HR experts and say what do weneed our AI agents, to abide by
how do we need to ground them inwhat documents is there a single
(08:02):
source of truth, ideally, likethere's for employees, or is it
okay if every department definesit somewhat similarly, but has
the autonomy.
I think there's a lot more, tocover there.
Also on the human to humancollaboration with your HR team,
as you're defining that.
But what are you seeing?
(08:23):
What do these agents actuallybring to processes?
How does that work if they workin a collaborative setting?
How can I envision that?
Bernhard Ritz (08:34):
I think the simplest comparison in AI terms
is these agents generatesomething like a chain of
thought.
In a traditional homeenvironment where you have a
ChatGPT, you ask something, youget a response, it has a memory
that follows what you're doing.
But here it's like this multiagent systems.
You have a bunch of agentstrained in different, in this
(08:56):
case, business roles, somebodywho's a compliant agent, a
finance analyst, an industryspecialist, you name it and they
start to figure out certaintopics and answer questions.
So instead of one prompt thatgives you an answer, You run a
full, call it a conversation, adiscussion between these agents.
(09:17):
They are contributing to theconversation.
They are reacting to eachother's answers.
So this guy said this.
Hey, the finance analyst saidthis.
Hey, entrepreneur, what do youthink about this?
Yeah, and so you run adiscussion.
You run a very complex chain ofthoughts from different
perspectives.
Design based on the role, andsome of these agents are trained
(09:37):
with internal information orexternal information.
So if you have somebody who istrained on your own information,
they bring perspective of thecompany.
Stuff based on customermaterial, or let's say you're a
pharmaceutical company and youhave 20 years of information in
clinical trials, you convertthis in a chatbot, you bring him
into the conversation andsuddenly you're bringing that
(09:59):
knowledge into a broaderconversation with a compliance
officer, with an entrepreneur,with a financial specialist.
So the idea is you can put,that's my philosophy, people
often say, oh, is it is AI hereto make things cheaper?
I say it's a mix.
It's more affordable,definitely, but it's also you
can put more efforts into eachtask.
(10:22):
And this is exactly this chainof thought.
So imagine every question thatyou answer, every step that
you're doing, you do a littleworkshop around it.
Let's figure this out, right?
Let's have a one hourconversation as a group how we
solve this.
How can we customize this tothis specific customer so it
makes a lot of sense to them?
Let's run this chain of thought.
Let's have a conversation fromdifferent perspectives.
(10:44):
I think you get the picture thatI'm drawing, right?
It's a, little mind dimension,these little angels that are
figuring out stuff.
And that's also something that alot of people say reasoning, can
AI reason?
Single prompt reasoning isgetting much better.
No question about that.
But if you run this chain ofthought and they have some
guardrails around whichdirection they are going up,
(11:06):
what are the key questions theyshould answer?
Instructions that they'refollowing, reasoning why they do
a specific task.
That gives them a lot ofguidance around how the team of
agents then generates responses.
Andreas Welsch (11:20):
Now, I think that the part that you mentioned
around accomplishing more in, inthe same amount of time or
having a more in depthdiscussion and evaluation of of
options, I think that's a keyaspect that I haven't seen a lot
of people talk about in, in thatsense.
It's not just about how can youbecome more efficient, meaning
(11:44):
how can you reduce the mundanethings, but what can you
accomplish in the same amount oftime now that we can scale it
and we can have all of thesevirtual experts.
come together and have thatdiscussion, which would us, or
which would take us an hour, twohours, half a day, right?
Great.
Bernhard Ritz (12:01):
I go back to what I said before, if you look at
this model of virtual employeesI developed a strategy framework
for AI together with WilliamFaye, and we went in and said,
how would things change?
How would your businessprocesses change?
If you could put 10x the effortinto what you're doing.
Imagine you have 10 times theresources to do something.
(12:24):
How would it change?
And this goes to some stuff thatI've also seen from you.
You you have to reimagine yourbusiness processes.
If you can customize.
Things much more than as oftoday, if you can put much more
effort into it, the result is,different.
Andreas Welsch (12:40):
Yeah, absolutely.
And I see Josh has put a commentin the chat.
We're going to see a need forleaders to manage two different
workforces, humans and AIagents.
I think that's spot on, right?
I think in, in many cases, Asleaders, we need to be prepared
(13:00):
to also now guide agents andgive them feedback.
And also think more criticallyagain when we're presented with
a proposal, whether our teammembers have used AI to generate
that.
And again, we don't know howthey prompted it or how what the
instructions were for, agents tocreate something, but.
(13:22):
If it's on us to make a decisionand say, should we go left?
Should we go right?
With these options, right?
We need to think a lot morecritical about that as well,
which I believe is getting moreimportant, but curious what
you're seeing.
Bernhard Ritz (13:36):
There's an interesting aspect that you
brought up, right?
This is this oversight of whatgoes into a problem.
Then do we understand why thesystem is answering the way it's
answering?
And I think multi agent systemshave an interesting advantage.
You're not just seeing theresult of, here's the answer,
it's 42.
It's more like you can followthat conversation between the
agents.
Ah, this is what they haveargued about.
(13:58):
This is what one agent asked theother agent, and then if you run
some monitoring on top of this,and you say is this compliant?
Can I see some bias in thesequestions and answers?
Then you have much more You havemuch better transparency of how
a decision is made and how wecame to a certain point instead
of just responding with the 42.
Andreas Welsch (14:18):
That's a great point.
And I think also comparing thatto spending an hour or two in a
room behind closed doors andcoming out with an answer, you
also don't know that the exactconversation flows, but here you
can actually audit it or take alook.
Bernhard Ritz (14:34):
It's a good point.
Looking at Josh's question weneed to manage two workforces.
I think a good start is if wesay, okay, let's apply the same
standards that we have forhumans to the AI agents, as I
said, right?
If we have a, code of businessconduct, we don't wanna see
anything in these conversationsthat doesn't need be, code of
(14:55):
business conduct, things likethat everything that goes on.
So the human standards is a goodstarting point, and then we need
additional measures foroversight on top.
Andreas Welsch (15:04):
Great point.
Yeah.
I think that'll become only moreimportant as more and more
businesses are exploring AIagents and how to deploy them
and do that in a way that it iscompliant.
We've seen that before withmachine learning where
algorithms have a differentoptimization function,
basically, and optimize for thatone outcome, but it can have
(15:25):
adversarial effects and negativeside effects to what you
actually want to do.
So to your point, grounding themin.
Here is the binding documentthat you need to abide by and
then leave your guardrails.
I think that's only getting moreand more important.
Now, I'm curious, when you usesystems that have multiple
(15:48):
agents collaborating, you cansee The conversation flow,
basically.
What did one say?
What did the other say?
How did they arrive at thisdecision?
What's really left for humansthen to do in terms of decision?
And where do we still needpeople in that process?
Bernhard Ritz (16:07):
There's probably a short term answer and there's
a long term answer, right?
You remember 12, 12 months ago,people said the humans are the
creatives and AI agents are not.
I guess we are now over thatpoint where we see a lot of
creativity in AI agents and howthey bring together topics that
are typically not together andthe new outcome is quite
creative.
(16:28):
I see functions or job rolesthat are easier to handle by AI
agents.
So you're probably not asoftware developer anymore, but
you are a software architect andyou're orchestrating, but at the
same time we see that more andmore these AI agents climb up
the ladder to more complextasks, more complex roles.
(16:51):
Yesterday they coded, nowthey're architecting solutions.
Tomorrow, who knows?
Do I see a limit where I say AIwill never be able to X, Y, Z?
Actually not.
I don't see it.
It's more a question of how dowe deal with these capabilities.
And this is, I'm going back alittle bit to AI strategy, where
(17:12):
I say this gives you newcapabilities, you have to
rethink the way you do business,you have to rethink and
reimagine your businessprocesses, as we saw with this
HR question, you have to rethinkwhat are the skills required in
your company in this new way.
And you have to rethink themarket as well, because it's not
just you who has these newcapabilities, all the other
(17:33):
participants in the market, yourcompetitors, your customers,
your suppliers, you name it,have it, have the same
capabilities.
Could they suddenly substituteyour products or services?
And so we can do this in housewith AI.
We don't need company expert.
We don't burn out to do this.
Same for you.
You probably asking the samequestion.
Can I do stuff in house that Ihaven't done?
(17:54):
So that goes to the strategicidea.
It's we see the technology.
Technology has evolvedsignificantly over the last two
years and now more and more thebusiness aspect shines.
Andreas Welsch (18:07):
Now, that makes me wonder if you have your
agents, your multi agent systemsin your own company.
Your customers, your partners,your vendors, they do the same
thing.
Where's this going?
Do you envision a future whereAgents of different companies
then collaborate, much like wewould collaborate with our
(18:30):
suppliers, with our partners,with our customers and say, what
products do you need?
When can we actually build orship them?
What material do we need tosource?
Again, collaborate with with thepartner or with the supplier.
Where's this going and where,does it stop or where does it
end if it is?
Bernhard Ritz (18:49):
You're already drawing the next generation,
Andreas.
Quite interesting.
So it's a marketplace whereagents from different companies
start to collaborate.
You have a little workshop goingon with your customers and
suppliers.
Interesting thought.
I haven't thought about it, butwhy not?
We're not there yet.
So we're still in the process ofcompany internal agents that are
(19:12):
collaborating.
But what you can do today is,you can put agents into the role
of customers and try to get testcertain things you could put an
AI agent in the role of acompetitor.
And see how they react tocertain activities that you're
doing.
Again it's, crazy how fast thewhole thing develops.
(19:33):
I like the idea of intercompanyagent collaboration.
I haven't seen it.
Here's a business opportunityfor somebody on the call.
Andreas Welsch (19:40):
That's true.
Now, I obviously see it as, andyou're most likely seeing this
too, all of the large vendors inthe market have announced that
they're Some sort of agentstrategy, agent features,
whether it's Salesforce, orOracle, or SAP, or Microsoft,
and many others.
But I also see them buildingtheir capabilities very tightly.
(20:02):
into their applications, right?
It's all about creating, on onehand, more value, on the other
hand, more lock in, if you will.
But what I haven't seen a lot ofis interagent, or not interagent
operability, but basicallyinteroperability between these
different solutions.
Because in a business processand in a company, you most
(20:23):
likely don't just have onevendor wall as much as Mentors
would love that to be true.
So I think there's anotheropportunity when it comes to
integration, if you have an HRworkflow and there, there are
some budget questions.
You can see the differencebetween, say, Workday to SAP.
What does that integration looklike?
(20:43):
But I think there are alsoadvantages, right?
If you're not locked into any ofthese verticals, if you do have
an open platform.
What are you seeing there?
Bernhard Ritz (20:54):
I think a lot of companies are in a position
where they try to find outWhat's a good starting point for
us with AI?
How do we get started, right?
And when you look at yourbusiness system, your
salesforce, your ServiceNow,your SAP, your Oracle, you name
it, yeah these vendors areadding these agent capabilities
(21:14):
to their solutions.
So they will handle part of theprocesses or the processes
within these systems.
What I also saw is that they arenow open to connecting to other
systems.
Salesforce agent connecting toan SAP system and vice versa.
So there's some level ofopenness, but not to the point,
probably how a third partycould, do this.
(21:34):
When I look at Ansel, obviouslyI say, yeah, we are connecting
to multiple systems and you havethis third party that can
independently orchestrate it,the capabilities that I see from
the different vendors.
Everybody goes into the samedirection of this agent and what
we discussed, how they haveevolved and how they will
evolve.
It's just a question of what'sthe orchestration framework for
(21:57):
all of these agents and howflexible and how open is it to
the different vendors.
And again, it will be anevolution.
The vendors, the current vendorshave a big advantage because
their processes and theirsolutions are already
implemented in companies.
Why wouldn't you start with justadding AI capabilities that
these vendors come?
(22:18):
But then when it comes to theinteroperability between the
systems, that will be aninteresting situation how this
works over there.
Andreas Welsch (22:26):
Now, let's bring this back to one of the earlier
points that you made, thatideally we start with viewing
agents like a human worker,right?
When it comes to policies andguidelines.
Now imagine Again, your HRdepartment and your finance
department and your salesdepartment all saying, here's
the scope of what I do.
(22:47):
If you want to talk to me again,you need to fill out a form or
you need some kind oforchestration or integration.
So I know what I need to do withthe information that you give to
me, or even worse saying, All Ican do is, this little box,
don't come to me for anythingelse, right?
To your point, I think that'sthe next evolution that we see
and that we will need to see.
(23:08):
We it's fine to have silos whenit comes to business systems,
but for business and the processthat goes horizontal, it doesn't
really matter if it's vendor oneor two or three.
From a business point of view.
So I think there's another bigopportunity.
And like Milind is posting inthe chat here, right?
Perhaps we need to developstandards or open systems that
(23:31):
allow intercompany agentcollaboration and even
interdepartmental collaboration,I would say.
Now, we've, covered a lot ofground in, in, in the last 25
minutes.
I'm curious where are you seeingthings go?
What's your recommendation forleaders looking to learn more
about AI agents and starting onthis journey?
Bernhard Ritz (23:53):
I think we touched some of these points.
So step back and look at this asa business capability and what
does it do to your business, andmost companies have experience
with.
Little projects, little petproject from the IT department
or some other departments, howto use AI.
I haven't seen a lot ofcompanies who really stepped
back and said, okay, how do wereimagine our business?
(24:16):
How does this influence ourposition in the market?
So there's a bigger picture tothis.
And especially when you havethis picture of these virtual
employees and you can reimagine,then that's a good metaphor to
do that.
When it comes to the technology,a lot of companies have.
Started with internal chatbots,where they have trained agents
(24:37):
on their own data, theirproprietary data.
That's the starting point,that's probably the pre version
of an agent, yeah?
Look at existing AI agentframeworks, and take a look if
there's something, like aninstant platform, that helps you
to orchestrate this, agentcommunication.
(24:58):
Other than that, we're still inearly stages, we're just at the
phase where, We are adding AI tothe business processes.
Companies need to look how theyhandle this, oversight, this
compliance.
I think topics likehallucinations, they're starting
to go away because there's moreand more done in terms of, okay,
give me the sources.
Please validate this withsources.
(25:19):
So hallucination, in my opinion,is something that goes away in
the next 12 months where we havechain of thoughts and all kinds
of processes that, that help usto overcome this.
But when it comes to.
Topics like, the bias and givingmore autonomy to the AI agents
and taking over bigger parts ofthe business processes, I think
(25:40):
this will be some of thechallenges and things we have to
look at when we go step by stepinto this AI ready world, AI
enabled world.
Andreas Welsch (25:49):
No, that, that sounds exciting.
I can't wait for hallucinationsto go away because if we
starting to put LLMs and Gen AIinto so many critical places and
having the certainty that doeswhat it says and says what it
does is really important.
Now we're getting close to theend of the show, and I was
wondering if you can summarizethe key three takeaways for our
audience today.
Bernhard Ritz (26:10):
Take a business look at AI.
Yeah.
Think about re imagining yourbusiness and your market
situation is one thing.
Learn to work with AI agentframeworks, because they're the
evolution of large languagemodels.
It's not about individualpromptings anymore.
But it's trained agents incertain domains who are really
(26:33):
world class experts in what theyare doing and a whole bunch of
them collaborating together tofigure out stuff.
Companies have to get theirfeedback in learning this new
way of doing business.
Okay, there's a whole, I take anindividual contributor and I put
this virtual team around him andnow he's a manager of a virtual
team and has suddenly all thiscapability available to him.
(26:54):
And the third one is, yeah, allof these.
changes to, to let's sayautonomy.
How far do I go?
How do I step by step unleash AIand do I even want to unleash
it?
How does a company first want toproceed with these capabilities
in a responsible and ethicalway?
I think that these are othertopics that should be
(27:15):
considered.
Andreas Welsch (27:16):
Awesome.
For sharing that and forsummarizing.
our discussion today.
Folks, for those of you in theaudience, if you would like to
learn more about AI agents, atleast on a high level recently
released two courses on LinkedInlearning that help you get
started with AI agents and putin the basics of what it is all
about and where theopportunities and the challenges
(27:36):
are.
And I thank you so much forjoining us and for sharing your
expertise with us.
It's been great having thisconversation with you and
learning from you as well, whereyou see things be today and in
the future.
Bernhard Ritz (27:47):
Thanks for having me on the show.
Today, we'll talk about supercharging your
business processes withcollaborative AI agents.
And who better to talk about itthan someone who's actively
working on that.
Bernhard Ritz.
Hey Bernhard, thank you so muchfor joining.
Bernhard Ritz (00:11):
Hello Andreas, thanks for having me on the
show.
Andreas Welsch (00:14):
Awesome.
Hey, why don't you tell ouraudience a little bit about
yourself, who you are and whatyou do?
Bernhard Ritz (00:18):
Sure.
I'm the CEO and founder ofAinzel, which is an enterprise
AI platform that adds virtualemployees, AI agents to
businesses.
I worked for 22 years for SAP.
A couple of years in Germany,later here in the US.
And then I joined an SAP partnerto build up their North America
business.
And around 18 months ago, Istarted Ainzel, so I started my
(00:41):
own company.
Andreas Welsch (00:42):
That's awesome.
I'm really looking forward toour conversation today.
You've shown me Ainzel a coupleof weeks ago, and I was so
excited to see thesecollaborative agents and what it
actually means.
Looking forward to ourconversation.
And also don't forget to pick upthe AI Leadership Handbook on
AILeadershipHandbook.
com.
(01:03):
So you can learn how you cansuccessfully implement AI in
your business.
Because it's not just all abouttech.
With all of that out of the way,What do you say?
Should we play a little game tokick things off?
Bernhard Ritz (01:13):
Okay.
Andreas Welsch (01:15):
Wonderful.
So hang on.
This one is called In Your OwnWords.
And when I hit the buzzer, thewheels will start spinning.
When they stop, you'll see asentence and I'd like for you to
answer with the first thing thatcomes to mind and why.
In your own words.
To make it a little moreinteresting, you only have 60
seconds for your answer.
And for those of you who arewatching us live, drop your
(01:36):
answer in the chat and why too.
So are you ready for What's theBUZZ?
Bernhard Ritz (01:41):
Let's go.
Andreas Welsch (01:41):
Okay, here we go.
If AI were a movie, what wouldit be?
60 seconds on the clock.
Bernhard Ritz (01:49):
Movie, that's an easy one.
It's Inside Out.
You probably know the Disneymovie, where these little
figures are thinking about you,your different emotions and so
on.
So that's a great parallel thingto, AI agents.
If we had the point where wehave them reasoning and think,
(02:11):
so all these little differentaspects of your life together,
collaborating, so perfectcomparison to AI agents.
Andreas Welsch (02:20):
I love that.
That's wonderful.
Definitely on message and themefor the show.
let's jump right into it, right?
You already mentioned agentsworking together, but maybe even
before we talk about themcollaborating, what does it even
mean to bring agents into abusiness process?
(02:40):
I think agents have been the hottopic this summer.
But what does it actually meanin concrete terms when you do
that?
Bernhard Ritz (02:49):
The way I look at this is, and a lot of people ask
me and say, okay, AI agents,there's a lot of technology
involved.
And I tell them, step back,don't look at the technology,
look at what they do for thebusiness, these AI agents.
And I always compare it withvirtual employees, which is
scary for some people.
Some people say, Oh no, they'rejust augmenting, they're not
(03:09):
real employees and so on.
But step by step AI agentsworking towards this idea of.
being part of the team, being amember.
Now there are these littlevirtual helpers that are built
on generative AI.
They started with very simpletasks can you improve this
email?
And then it went to, can youwrite this email?
(03:31):
Can you run this email campaign?
Can you manage my email inboxfor me?
So step by step these littlehelpers got better and step by
step we're integrating themdeeper into our business
processes.
As I just described we use themto augment what we are doing and
(03:51):
now they are starting to takeover single activities, single
steps.
They're very knowledgeable, theyare trained in certain domains
at an expert level.
They are fast, they respond withinformation in seconds that
would take us minutes and hoursto compile.
So very helpful, affordable,knowledgeable trained little
(04:16):
helpers.
And this is actually how thename ANZL came along.
I thought, how can you namethese little helpers?
And in Germany, there issomething called
Mainzelmaennchen, which is aGerman TV show.
And I tried to find a domainname that fits to
Mainzelmaennchen, these littlehelpers, and I came up with
Ainzl.
And later recognized, boop,there is AI at the beginning of
it.
Yeah.
Andreas Welsch (04:37):
I love that analogy.
I've been wondering, too.
Now, when you introduce more ofthis autonomy, into your
business process.
I'm sure that there's somechallenges that you will
introduce with that as well,right?
I think in business, it'salready tough enough if you have
a team of human workers andemployees, getting them to
(04:59):
communicate, getting them tocollaborate and agree on
terminology and what is the goaland how do we break this down
while still benefiting fromeverybody's individual
experience and background.
How do you see this evolving inbusiness when we introduce now.
AI agents as a softwarecomponent that can do some of
those little more complex tasksthan just if then else at this
(05:22):
time.
Bernhard Ritz (05:22):
You brought up an interesting point and this is
language.
I think the beauty of largelanguage models and the
underlying vector databasesalso, they bring a lot of
flexibility.
To things that AI agents canreact to.
If you think about traditionalapplications, right?
You have databases and they havestrict fields and you need to
(05:43):
provide the information in theexact format required so that it
can handle it.
With AI there's ontologies,there's vector databases that do
similarity searches, things likethat.
So you can have a very flexibleway of entering information and
the AI agent is still able to,handle it.
So I think that's a greatadvantage.
(06:04):
You said autonomy, obviouslythat's the big point.
The more the AI agents take overin terms of functionality and,
parts of the business processes,the more autonomy they will get
to do things independent.
And this is where also the riskis, where the oversight needs to
kick in.
About hallucinations, aboutbias, about privacy, security,
(06:31):
compliance, I said they arevirtual employees.
And if you play that model, thenyou see virtual employees, they
have an NDA, they have anemployee contract, they need to
follow code of business conduct.
So for the AI agents, we need todesign them the same way.
And I think this is also thedifference from enterprise AI
platform, which takes that veryserious in comparison if you
(06:53):
feed information into a largelanguage model and it's working
with you, but this oversightcomponent that you expect in an
enterprise platform is probablynot there.
Information goes out there,information is shared that
shouldn't be there.
So enterprise platform dealingwith these AI agents and the
collaboration between theagents, is the go to solution in
(07:16):
this case.
Andreas Welsch (07:17):
That's, awesome that you mentioned this because
I think a lot of times that partis, underestimated or maybe not
even covered.
So I was thinking about that alittle more the other week after
our conversation and I realized,for example, in a lot of
organizations, HR is actuallyabsent in these discussions,
right?
It's technologists defining whatshould these new agents, these
(07:40):
new digital workers look likeand how should they behave?
And what are the policies wewant to need them to abide by?
So I think there's anotheropportunity for technology
leaders to bring in.
HR experts and say what do weneed our AI agents, to abide by
how do we need to ground them inwhat documents is there a single
(08:02):
source of truth, ideally, likethere's for employees, or is it
okay if every department definesit somewhat similarly, but has
the autonomy.
I think there's a lot more, tocover there.
Also on the human to humancollaboration with your HR team,
as you're defining that.
But what are you seeing?
(08:23):
What do these agents actuallybring to processes?
How does that work if they workin a collaborative setting?
How can I envision that?
Bernhard Ritz (08:34):
I think the simplest comparison in AI terms
is these agents generatesomething like a chain of
thought.
In a traditional homeenvironment where you have a
ChatGPT, you ask something, youget a response, it has a memory
that follows what you're doing.
But here it's like this multiagent systems.
You have a bunch of agentstrained in different, in this
(08:56):
case, business roles, somebodywho's a compliant agent, a
finance analyst, an industryspecialist, you name it and they
start to figure out certaintopics and answer questions.
So instead of one prompt thatgives you an answer, You run a
full, call it a conversation, adiscussion between these agents.
(09:17):
They are contributing to theconversation.
They are reacting to eachother's answers.
So this guy said this.
Hey, the finance analyst saidthis.
Hey, entrepreneur, what do youthink about this?
Yeah, and so you run adiscussion.
You run a very complex chain ofthoughts from different
perspectives.
Design based on the role, andsome of these agents are trained
(09:37):
with internal information orexternal information.
So if you have somebody who istrained on your own information,
they bring perspective of thecompany.
Stuff based on customermaterial, or let's say you're a
pharmaceutical company and youhave 20 years of information in
clinical trials, you convertthis in a chatbot, you bring him
into the conversation andsuddenly you're bringing that
(09:59):
knowledge into a broaderconversation with a compliance
officer, with an entrepreneur,with a financial specialist.
So the idea is you can put,that's my philosophy, people
often say, oh, is it is AI hereto make things cheaper?
I say it's a mix.
It's more affordable,definitely, but it's also you
can put more efforts into eachtask.
(10:22):
And this is exactly this chainof thought.
So imagine every question thatyou answer, every step that
you're doing, you do a littleworkshop around it.
Let's figure this out, right?
Let's have a one hourconversation as a group how we
solve this.
How can we customize this tothis specific customer so it
makes a lot of sense to them?
Let's run this chain of thought.
Let's have a conversation fromdifferent perspectives.
(10:44):
I think you get the picture thatI'm drawing, right?
It's a, little mind dimension,these little angels that are
figuring out stuff.
And that's also something that alot of people say reasoning, can
AI reason?
Single prompt reasoning isgetting much better.
No question about that.
But if you run this chain ofthought and they have some
guardrails around whichdirection they are going up,
(11:06):
what are the key questions theyshould answer?
Instructions that they'refollowing, reasoning why they do
a specific task.
That gives them a lot ofguidance around how the team of
agents then generates responses.
Andreas Welsch (11:20):
Now, I think that the part that you mentioned
around accomplishing more in, inthe same amount of time or
having a more in depthdiscussion and evaluation of of
options, I think that's a keyaspect that I haven't seen a lot
of people talk about in, in thatsense.
It's not just about how can youbecome more efficient, meaning
(11:44):
how can you reduce the mundanethings, but what can you
accomplish in the same amount oftime now that we can scale it
and we can have all of thesevirtual experts.
come together and have thatdiscussion, which would us, or
which would take us an hour, twohours, half a day, right?
Great.
Bernhard Ritz (12:01):
I go back to what I said before, if you look at
this model of virtual employeesI developed a strategy framework
for AI together with WilliamFaye, and we went in and said,
how would things change?
How would your businessprocesses change?
If you could put 10x the effortinto what you're doing.
Imagine you have 10 times theresources to do something.
(12:24):
How would it change?
And this goes to some stuff thatI've also seen from you.
You you have to reimagine yourbusiness processes.
If you can customize.
Things much more than as oftoday, if you can put much more
effort into it, the result is,different.
Andreas Welsch (12:40):
Yeah, absolutely.
And I see Josh has put a commentin the chat.
We're going to see a need forleaders to manage two different
workforces, humans and AIagents.
I think that's spot on, right?
I think in, in many cases, Asleaders, we need to be prepared
(13:00):
to also now guide agents andgive them feedback.
And also think more criticallyagain when we're presented with
a proposal, whether our teammembers have used AI to generate
that.
And again, we don't know howthey prompted it or how what the
instructions were for, agents tocreate something, but.
(13:22):
If it's on us to make a decisionand say, should we go left?
Should we go right?
With these options, right?
We need to think a lot morecritical about that as well,
which I believe is getting moreimportant, but curious what
you're seeing.
Bernhard Ritz (13:36):
There's an interesting aspect that you
brought up, right?
This is this oversight of whatgoes into a problem.
Then do we understand why thesystem is answering the way it's
answering?
And I think multi agent systemshave an interesting advantage.
You're not just seeing theresult of, here's the answer,
it's 42.
It's more like you can followthat conversation between the
agents.
Ah, this is what they haveargued about.
(13:58):
This is what one agent asked theother agent, and then if you run
some monitoring on top of this,and you say is this compliant?
Can I see some bias in thesequestions and answers?
Then you have much more You havemuch better transparency of how
a decision is made and how wecame to a certain point instead
of just responding with the 42.
Andreas Welsch (14:18):
That's a great point.
And I think also comparing thatto spending an hour or two in a
room behind closed doors andcoming out with an answer, you
also don't know that the exactconversation flows, but here you
can actually audit it or take alook.
Bernhard Ritz (14:34):
It's a good point.
Looking at Josh's question weneed to manage two workforces.
I think a good start is if wesay, okay, let's apply the same
standards that we have forhumans to the AI agents, as I
said, right?
If we have a, code of businessconduct, we don't wanna see
anything in these conversationsthat doesn't need be, code of
(14:55):
business conduct, things likethat everything that goes on.
So the human standards is a goodstarting point, and then we need
additional measures foroversight on top.
Andreas Welsch (15:04):
Great point.
Yeah.
I think that'll become only moreimportant as more and more
businesses are exploring AIagents and how to deploy them
and do that in a way that it iscompliant.
We've seen that before withmachine learning where
algorithms have a differentoptimization function,
basically, and optimize for thatone outcome, but it can have
(15:25):
adversarial effects and negativeside effects to what you
actually want to do.
So to your point, grounding themin.
Here is the binding documentthat you need to abide by and
then leave your guardrails.
I think that's only getting moreand more important.
Now, I'm curious, when you usesystems that have multiple
(15:48):
agents collaborating, you cansee The conversation flow,
basically.
What did one say?
What did the other say?
How did they arrive at thisdecision?
What's really left for humansthen to do in terms of decision?
And where do we still needpeople in that process?
Bernhard Ritz (16:07):
There's probably a short term answer and there's
a long term answer, right?
You remember 12, 12 months ago,people said the humans are the
creatives and AI agents are not.
I guess we are now over thatpoint where we see a lot of
creativity in AI agents and howthey bring together topics that
are typically not together andthe new outcome is quite
creative.
(16:28):
I see functions or job rolesthat are easier to handle by AI
agents.
So you're probably not asoftware developer anymore, but
you are a software architect andyou're orchestrating, but at the
same time we see that more andmore these AI agents climb up
the ladder to more complextasks, more complex roles.
(16:51):
Yesterday they coded, nowthey're architecting solutions.
Tomorrow, who knows?
Do I see a limit where I say AIwill never be able to X, Y, Z?
Actually not.
I don't see it.
It's more a question of how dowe deal with these capabilities.
And this is, I'm going back alittle bit to AI strategy, where
(17:12):
I say this gives you newcapabilities, you have to
rethink the way you do business,you have to rethink and
reimagine your businessprocesses, as we saw with this
HR question, you have to rethinkwhat are the skills required in
your company in this new way.
And you have to rethink themarket as well, because it's not
just you who has these newcapabilities, all the other
(17:33):
participants in the market, yourcompetitors, your customers,
your suppliers, you name it,have it, have the same
capabilities.
Could they suddenly substituteyour products or services?
And so we can do this in housewith AI.
We don't need company expert.
We don't burn out to do this.
Same for you.
You probably asking the samequestion.
Can I do stuff in house that Ihaven't done?
(17:54):
So that goes to the strategicidea.
It's we see the technology.
Technology has evolvedsignificantly over the last two
years and now more and more thebusiness aspect shines.
Andreas Welsch (18:07):
Now, that makes me wonder if you have your
agents, your multi agent systemsin your own company.
Your customers, your partners,your vendors, they do the same
thing.
Where's this going?
Do you envision a future whereAgents of different companies
then collaborate, much like wewould collaborate with our
(18:30):
suppliers, with our partners,with our customers and say, what
products do you need?
When can we actually build orship them?
What material do we need tosource?
Again, collaborate with with thepartner or with the supplier.
Where's this going and where,does it stop or where does it
end if it is?
Bernhard Ritz (18:49):
You're already drawing the next generation,
Andreas.
Quite interesting.
So it's a marketplace whereagents from different companies
start to collaborate.
You have a little workshop goingon with your customers and
suppliers.
Interesting thought.
I haven't thought about it, butwhy not?
We're not there yet.
So we're still in the process ofcompany internal agents that are
(19:12):
collaborating.
But what you can do today is,you can put agents into the role
of customers and try to get testcertain things you could put an
AI agent in the role of acompetitor.
And see how they react tocertain activities that you're
doing.
Again it's, crazy how fast thewhole thing develops.
(19:33):
I like the idea of intercompanyagent collaboration.
I haven't seen it.
Here's a business opportunityfor somebody on the call.
Andreas Welsch (19:40):
That's true.
Now, I obviously see it as, andyou're most likely seeing this
too, all of the large vendors inthe market have announced that
they're Some sort of agentstrategy, agent features,
whether it's Salesforce, orOracle, or SAP, or Microsoft,
and many others.
But I also see them buildingtheir capabilities very tightly.
(20:02):
into their applications, right?
It's all about creating, on onehand, more value, on the other
hand, more lock in, if you will.
But what I haven't seen a lot ofis interagent, or not interagent
operability, but basicallyinteroperability between these
different solutions.
Because in a business processand in a company, you most
(20:23):
likely don't just have onevendor wall as much as Mentors
would love that to be true.
So I think there's anotheropportunity when it comes to
integration, if you have an HRworkflow and there, there are
some budget questions.
You can see the differencebetween, say, Workday to SAP.
What does that integration looklike?
(20:43):
But I think there are alsoadvantages, right?
If you're not locked into any ofthese verticals, if you do have
an open platform.
What are you seeing there?
Bernhard Ritz (20:54):
I think a lot of companies are in a position
where they try to find outWhat's a good starting point for
us with AI?
How do we get started, right?
And when you look at yourbusiness system, your
salesforce, your ServiceNow,your SAP, your Oracle, you name
it, yeah these vendors areadding these agent capabilities
(21:14):
to their solutions.
So they will handle part of theprocesses or the processes
within these systems.
What I also saw is that they arenow open to connecting to other
systems.
Salesforce agent connecting toan SAP system and vice versa.
So there's some level ofopenness, but not to the point,
probably how a third partycould, do this.
(21:34):
When I look at Ansel, obviouslyI say, yeah, we are connecting
to multiple systems and you havethis third party that can
independently orchestrate it,the capabilities that I see from
the different vendors.
Everybody goes into the samedirection of this agent and what
we discussed, how they haveevolved and how they will
evolve.
It's just a question of what'sthe orchestration framework for
(21:57):
all of these agents and howflexible and how open is it to
the different vendors.
And again, it will be anevolution.
The vendors, the current vendorshave a big advantage because
their processes and theirsolutions are already
implemented in companies.
Why wouldn't you start with justadding AI capabilities that
these vendors come?
(22:18):
But then when it comes to theinteroperability between the
systems, that will be aninteresting situation how this
works over there.
Andreas Welsch (22:26):
Now, let's bring this back to one of the earlier
points that you made, thatideally we start with viewing
agents like a human worker,right?
When it comes to policies andguidelines.
Now imagine Again, your HRdepartment and your finance
department and your salesdepartment all saying, here's
the scope of what I do.
(22:47):
If you want to talk to me again,you need to fill out a form or
you need some kind oforchestration or integration.
So I know what I need to do withthe information that you give to
me, or even worse saying, All Ican do is, this little box,
don't come to me for anythingelse, right?
To your point, I think that'sthe next evolution that we see
and that we will need to see.
(23:08):
We it's fine to have silos whenit comes to business systems,
but for business and the processthat goes horizontal, it doesn't
really matter if it's vendor oneor two or three.
From a business point of view.
So I think there's another bigopportunity.
And like Milind is posting inthe chat here, right?
Perhaps we need to developstandards or open systems that
(23:31):
allow intercompany agentcollaboration and even
interdepartmental collaboration,I would say.
Now, we've, covered a lot ofground in, in, in the last 25
minutes.
I'm curious where are you seeingthings go?
What's your recommendation forleaders looking to learn more
about AI agents and starting onthis journey?
Bernhard Ritz (23:53):
I think we touched some of these points.
So step back and look at this asa business capability and what
does it do to your business, andmost companies have experience
with.
Little projects, little petproject from the IT department
or some other departments, howto use AI.
I haven't seen a lot ofcompanies who really stepped
back and said, okay, how do wereimagine our business?
(24:16):
How does this influence ourposition in the market?
So there's a bigger picture tothis.
And especially when you havethis picture of these virtual
employees and you can reimagine,then that's a good metaphor to
do that.
When it comes to the technology,a lot of companies have.
Started with internal chatbots,where they have trained agents
(24:37):
on their own data, theirproprietary data.
That's the starting point,that's probably the pre version
of an agent, yeah?
Look at existing AI agentframeworks, and take a look if
there's something, like aninstant platform, that helps you
to orchestrate this, agentcommunication.
(24:58):
Other than that, we're still inearly stages, we're just at the
phase where, We are adding AI tothe business processes.
Companies need to look how theyhandle this, oversight, this
compliance.
I think topics likehallucinations, they're starting
to go away because there's moreand more done in terms of, okay,
give me the sources.
Please validate this withsources.
(25:19):
So hallucination, in my opinion,is something that goes away in
the next 12 months where we havechain of thoughts and all kinds
of processes that, that help usto overcome this.
But when it comes to.
Topics like, the bias and givingmore autonomy to the AI agents
and taking over bigger parts ofthe business processes, I think
(25:40):
this will be some of thechallenges and things we have to
look at when we go step by stepinto this AI ready world, AI
enabled world.
Andreas Welsch (25:49):
No, that, that sounds exciting.
I can't wait for hallucinationsto go away because if we
starting to put LLMs and Gen AIinto so many critical places and
having the certainty that doeswhat it says and says what it
does is really important.
Now we're getting close to theend of the show, and I was
wondering if you can summarizethe key three takeaways for our
audience today.
Bernhard Ritz (26:10):
Take a business look at AI.
Yeah.
Think about re imagining yourbusiness and your market
situation is one thing.
Learn to work with AI agentframeworks, because they're the
evolution of large languagemodels.
It's not about individualpromptings anymore.
But it's trained agents incertain domains who are really
(26:33):
world class experts in what theyare doing and a whole bunch of
them collaborating together tofigure out stuff.
Companies have to get theirfeedback in learning this new
way of doing business.
Okay, there's a whole, I take anindividual contributor and I put
this virtual team around him andnow he's a manager of a virtual
team and has suddenly all thiscapability available to him.
(26:54):
And the third one is, yeah, allof these.
changes to, to let's sayautonomy.
How far do I go?
How do I step by step unleash AIand do I even want to unleash
it?
How does a company first want toproceed with these capabilities
in a responsible and ethicalway?
I think that these are othertopics that should be
(27:15):
considered.
Andreas Welsch (27:16):
Awesome.
For sharing that and forsummarizing.
our discussion today.
Folks, for those of you in theaudience, if you would like to
learn more about AI agents, atleast on a high level recently
released two courses on LinkedInlearning that help you get
started with AI agents and putin the basics of what it is all
about and where theopportunities and the challenges
(27:36):
are.
And I thank you so much forjoining us and for sharing your
expertise with us.
It's been great having thisconversation with you and
learning from you as well, whereyou see things be today and in
the future.
Bernhard Ritz (27:47):
Thanks for having me on the show.
Popular Podcasts
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
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Therapy Gecko
An unlicensed lizard psychologist travels the universe talking to strangers about absolutely nothing. TO CALL THE GECKO: follow me on https://www.twitch.tv/lyleforever to get a notification for when I am taking calls. I am usually live Mondays, Wednesdays, and Fridays but lately a lot of other times too. I am a gecko.