November 26, 2025 • 35 mins
“Why do I have to tell your chatbot to do something? Just do it.”
In this episode, Jeff Seibert – founder of Digits (AI-native accounting platform), former Twitter Head of Product, and the engineer behind Crashlytics (now on 6 billion devices) – reveals what it actually takes to build AI-native companies from scratch. We explore why most companies are getting AI wrong by bolting chatbots onto old products, how to structure teams for extreme velocity, and why the accounting industry is about to experience its HP-35 calculator moment. Jeff’s bold prediction: the entire month-end close process will be automated within 12 months.
What You’ll Discover:
[02:45] Why Accounting Data Quality is Decades Behind Product Analytics → The genesis story of Digits: when Twitter’s 100-person finance team couldn’t answer a simple budget question in under three weeks
[08:28] Building Companies for AI From Day One → How ML-native architecture differs from traditional databases and why this matters more than the AI hype suggests
[10:31] The 65-Person Company That Runs All-Hands Every 48 Hours → Jeff’s radical approach to velocity: weekly sprints, fractal team structures, and why they’ll never hire “lone eagle” engineers
[15:20] Keeping Teams Intentionally Small at Scale → How to eliminate the “empire building” problem by dissociating engineering coaches from project staffing
[19:59] What CEOs Actually Do That AI Can’t Replace (Yet) → The 10%/90% leadership philosophy and why Sundar Pichai’s “AI will replace CEOs” take misses the point
[23:30] Disrupting QuickBooks: Technology vs. Distribution → Why accounting is uniquely suited for AI disruption and how startups can outpace 800-pound gorillas
[26:14] Why AI Isn’t Just Another Ajax Moment → The fundamental shift from “talk to our chatbot” to “the AI should just do it” – and what that means for software architecture
[30:47] The Architectural Wall Ahead for Large Language Models → Why current LLM architecture won’t reach AGI: the context window problem, lack of memory, and inability to backtrack during inference
[32:05] The Great Work Displacement: Data Entry is Dead by 2026 → Jeff’s evolved prediction on AI’s economic impact and why the “lump of labor fallacy” applies to automation fears
Key Takeaways:
- AI-native means redesigning your data architecture from scratch, not adding a chatbot interface to legacy systems
- Run your company on the shortest planning horizon you can see – for Digits, that’s 4-5 week “horizons”
- Hire senior people who are “chill” with strong opinions, loosely held – and actively filter out solo operators
- The most powerful AI products won’t ask users what to do – they’ll understand the goal and just execute
- Accounting’s month-end close will be automated by end of 2025, marking one of AI’s first complete workflow eliminations
About Jeff Seibert:
Jeff is the founder and CEO of Digits, the AI-native accounting platform. Previously, he served as Twitter’s Head of Consumer Product (launching the algorithmic timeline), co-founded Crashlytics (acquired by Twitter, now runs on 6 billion smartphones), and was featured in Netflix’s Emmy-winning documentary “The Social Dilemma.” He’s backed 100+ startups as an angel investor and has been building software since releasing his first app at age 12.
Related Links:
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:07):
Build for turbulence. This has always driven me crazy
because I don't want another thing to talk to.
But why do I have to tell your chat bot to do something?
Just do it. Hey friends, Pascal here.
Welcome to another episode of Built for Turbulence.
Yep, we've rebranded the. Podcast to match the title of
(00:28):
our next book and. Better capture what we're.
Talking about on the Pod and a fair warning, I was fighting a
bit of a cold during this recording, so if I sound a
little nasal, that's why. Today I'm talking with Jeff
Siebert, founder and CEO of Digits.
The world's first. AI native accounting platform
Jeff's one of those builders who've been in the arena for the
(00:50):
long haul. He.
Previously served as Twitter's head of consumer product.
And if you've seen the Emmy award-winning Netflix
documentary The Social Dilemma, you've seen Jeff.
He was one of the featured voices in that conversation
about technology's impact on society.
Before Digits and Twitter, Jeff Co founded and Let Crush Lytics,
(01:11):
the mobile performance analyticscompany that Twitter acquired in
2013. If you're not.
Familiar with Crashlytics? Here's the scale we're talking
about. It now runs on over 6 billion
monthly active smartphone. That's right, billions with AB.
Google owns it now and it's the market leading crash reporter
(01:31):
for both iOS and Android. Just a self-taught programmer
who released his first app at 12years old, then went on to
Stanford where he earned his BS in Computer Science.
He's also an active Angel investor.
He backed more than 100 startupsand Insider named him one of
their top 100 CPCS in 2021. What I love about this
(01:53):
conversation is that Jeff's beenon every side of the table.
Founder, operator at scale, investor, and now founder again.
He knows what it takes to build in turbulence and we had a
really rich conversation about AI product and what it means to
build companies that matter. Jeff, I'm generally stoked to
have you on the podcast. Welcome.
(02:14):
Pascal, it's so great to be here.
Thanks for having me. I have to start with a question.
So you run Digits and I want to hear a little bit more about
what Digits actually does. But it is an AI native
accounting platform. You come from a very different
background, Twitter and you did a start up before that which
(02:34):
very successful and also you were in the social dilemma.
We need to talk about that at one point, but how do you go
from that to accounting? Yes, great question.
So yeah, I had no background in finance or accounting.
My life has actually been code. I taught myself C when I was 12.
I fell in love with the concept of being able to program a
(02:56):
computer and write the code onceand have it run 1000 or
1,000,000 or a billion times andthe output was uncorrelated with
the input. It was like blow mind blowing to
me as a middle schooler. And so I just, I just spent my
whole life coding. And so Digits is actually my
third start up. I did a document collaboration
company back in O 7. We got acquired by Box.
I built Crashlytics in the mobile developer tool space.
(03:19):
We got acquired by Twitter and Digits actually came from that
journey of those companies. And then also what I saw at
Twitter and what struck me was the difference in data quality
between product engineering and finance.
And on the product engineering side, you have Google Analytics,
you have AB testing tools, you have Grafana, right?
(03:39):
All these live dashboards that tell you exactly what your users
are doing, what your servers aredoing, and you can figure out
what's going on. And then on the finance side,
like I, as the founder and CEO, was waiting two to three weeks
to get a black and white PNL andbalance sheet from our
accountant. And I was like, there's clearly
something broken here, right? Like, why don't I have a live
(04:00):
finance dashboard? And so that was honestly the
genesis of digits. And at first I thought it was
just a small business issue. And I was like, oh, I'm sure big
companies have got this sorted out.
And then at Twitter, as head of product, I went to corporate
finance 1/4 and I was like, what's our budget?
I'd like to run an event for theteam.
And their actual answer was, oh,we haven't run those books yet.
(04:21):
Give us 3 weeks and I was like, you have 100 people in corporate
finance, what are you doing? And so that that was, it
basically had to fall down the accounting rabbit hole, talk
with hundreds of accountants, quickly realized that it wasn't
their fault, It wasn't the people like accountants are very
hard working, very disciplined, diligent, etcetera.
They were being held back by thesoftware.
(04:43):
And so we realized in 2018, we had to start from scratch and
build a new general Ledger from the ground up designed for real
time finance. Let me ask you a question.
If I zoom out from the comment you made about data, I find this
actually an interesting question.
In a world which is increasinglyuncertain, how much does past?
(05:06):
Because most all of our data is past data, how much do you think
from a business perspective, from a leadership perspective,
from a leader's perspective, does past data actually help you
in a world which is increasinglyuncertain?
I think that is a really thoughtful question.
I, I think it's still an essential building block because
(05:27):
businesses tend to have momentum.
And so yes, the world is uncertain, but if you're going
to be informed by anything, you might as well anchor it in what
you have done over the past couple periods to at least give
you a starting point. And one of the challenge, one of
an example of this is you talk with small business owners all
across the country, almost none of them have a budget.
(05:47):
And it's largely because they can't afford to pay their
account and to create a budget for them.
And even if they did, maintaining it is so tedious
today, right? Like you have to measure it
against the actuals every month and update it and adjust it and
so on. Imagine a world where actually
your AI was able to just give you a budget based on last
month's spend. And it's like, that's likely
(06:07):
what you're going to spend this month unless you change
something. And so then you could just drag
a little slider and be like, oh,I want to increase marketing
spend or I want to hire another person, or I want to sell off
one of my trucks or whatever it is for my business.
Like at least it gives you a starting point.
I think. Otherwise you're flying blind.
And yes, the rate of change in speed of business has increased
dramatically, but you still needto anchor yourself somewhere.
(06:30):
Interesting. And let me ask you another
question about digits, because you're playing in an arena which
has, I wouldn't even say £800 gorillas.
They're like probably beyond that, right?
The likes of the QuickBooks, etcetera.
What makes you compete? What makes you start a company
(06:50):
in an arena which feels so incredibly entrenched?
Yes. So definitely I don't know if
it's a rational choice, but it, this is my third company I had
sold the prior to Crashlytics did achieve insane scale.
So we grew to 300 million phonesin 12 months.
Today Crashlytics runs on 6 billion smartphones on Earth,
(07:12):
basically every active smartphone and we were in over
1,000,000 apps. And so we, we really were like
the de facto performance monitoring tool for the mobile
ecosystem. And I guess that gave me the
confidence that we can build something that truly scales and
what's a bigger market and like a, an even more compelling
opportunity to make changes. If you can make finance and
(07:34):
accounting real time and visual and intuitive, that becomes
pretty powerful. So I've never been afraid of
tackling a big challenge. I think if you can design A
better product and of course, pair it with distribution is the
key. And I wouldn't be worried about
the status quo. It's let's imagine what the
world has to look like in 10 years and just only focus on the
(07:55):
path to get there. Interesting.
And then I'd be curious to understand, you have built a
company. So first of all, your company is
AI native. So you literally built a company
just with a little bit of AI sprinkling on, like you built a
company from the ground up as being a company which is set to
thrive in an AI world. You've built companies before.
(08:18):
You also worked for a very largecompany as a leader.
How is leading different for youin a company which is AI native?
Oh, interesting. So this has been core back to
when we started digits in 2018. And I actually describe this as
ML native more than AI native, right?
Because AI now has been this hype term for the past couple
years. But way back to Twitter as had a
(08:41):
product, I launched the algorithmic timeline.
So the transition from just a straight chronological to now
it's ranked and sorted, right. And that was with Facebook,
which had done it right before we did one of the first global
deployments of machine learning tech.
And I saw the power of ML, I also saw all the downsides in
terms of misinformation and polarization and so on.
(09:03):
That became a social dilemma film.
But I wanted to build a company where we could use that
technology but avoid the downsides.
And I was like, accounting is great.
Like you, everyone would love itif you could automate the
tedium. And so there's not real risk of
like misinformation or whatever.It's let's just do the work.
And so we really structured morethan just the company, but the
(09:23):
actual architecture around beingML native.
So like going back to the fundamental data design of
Digits, we aren't a traditional relational database.
Like when you load the Digits dashboard, you're not hitting a
database. We actually have a whole
different data architecture and we render the dashboard in 100
milliseconds. We designed that for machine
learning algorithms to better manipulate.
(09:45):
And so I've really thought aboutit in terms of product
capability and innovation more so than how to run the company
approaches. We used to run the business I
actually think came from Crashlytics back in 2011.
So we do a bunch of stuff very differently.
We run the whole thing on a weekly Sprint.
I've run on all hands every 48 hours for the past seven years.
(10:07):
It's yeah. Remind me, how big is your
company today? We are 65 people.
OK. So if you project this out
orders of magnitude like I am very sure you will get to the
650 people company and beyond, how do you think you can
maintain that if you can maintain it?
(10:28):
Like the? Speed, the agility, etcetera.
It's a great question and something I think about
constantly. I actually don't know if we'll
need to get to 650 people. My goal is to have the smallest
team at every stage of the market opportunity.
And you look at teams like WhatsApp, I think we're around
20 people when they were acquired by Facebook Enterprise
(10:49):
Software. You do need a few more hands,
but I don't think you need hundreds or thousands.
And so we'll see. But I think it's, we've already
scaled 2 orders of magnitude. So like way back at the start of
the company, there's just a few folks around the table and every
single person speaks at every all hands sync, right and says
what they're doing and where they need help.
And then it fractalizes to individual like work teams, work
(11:12):
streams and then it fractalizes to sort of departments of, OK,
go to markets doing this, we need help with this, whatever.
So I think you can keep fractalizing it out and it works
pretty well. And it's more about a mindset of
being able to run the whole business on a weekly cadence
where we're all setting goal goals Monday morning.
We all do show and tell Friday afternoon and show off what we
(11:34):
did that week and that just mindset creates this momentum in
the business where and it allowsyou to just tack every single
week if you need to adapt to market conditions.
In this context, how do you do long term planning?
So we plan in periods we call horizons.
They are horizons because it is literally as far as we can see.
(11:56):
And right now a horizon is 4 to 5 weeks and Yep.
And so we plan out what looks like effectively the next month
of what are our goals, where do we want to be.
And it's very lightweight. It's just a couple bullets for
each team of what do you want tohave achieved in the next 4 to 5
weeks. And then we make sure that the
weekly goals you set each week are directionally aligned with
(12:18):
where you want to be in that horizon.
I'm curious for your people, I totally can see how this works
in your head. Does this work for your people?
I find one of the challenges I find with very short planning
horizons which deck you. And I love that some people like
to have more of a much longer planning horizon as a but how do
(12:41):
you know if I perform well in myperformance review?
Like that kind of thing, like sitting down at the 1st of
January and like determining what do I want to achieve this
year. Have you come across and said
like how do you manage that? It's a it's definitely a very
different culture and certainly for some of those folks, a
company like Digits won't be a fit for them.
We do not do performance reviewsperiod.
(13:03):
And that's because in my career I've never seen it effective.
Twitter did them at twice at once to twice a year and it was
a nightmare process and took everyone's time and then nothing
really happens. We prefer to do just continual
one on ones. We have a coaching mentality
where your manager is really responsible for sharing, like
helping guide you in your careerand what are your goals?
What do you want to achieve? OK, let's try to arrange
(13:25):
opportunities for you to get there.
And then we focus a lot on team collaboration and so do not
reward individual contribution whatsoever.
Everything at the company is done in minimum pairs of people
because like it allows you to beway more thoughtful in the
approach. It makes it way more fun to work
on projects together versus by yourself.
And so we don't reward any individual achievement.
(13:48):
We recognize folks who help other people in other teams.
Interesting. Presumably you've got a bunch of
engineers in your company though.
Yes, my experience was, particularly when I was at
Mozilla working with some reallysmart engineers, that there is a
certain class of engineers who are lone eagles.
(14:09):
Yes, they don't hire them. Oh, OK, interesting.
So yeah, we have it's, this is these are great line of
questions because it goes back to the whole philosophy of the
business and what you're aiming for.
So way back in our interview process, we actually spend the
bulk of the time in the interview process not on
technical skill, but on collaboration and career history
and progression and more of a culture style interview.
(14:32):
And so we actively select against the more cowboy coder
style engineers, the Lone Eagle style engineers.
And the way I describe what we look for is senior folks.
We generally only hire senior talent across the company.
The danger with that is you get people who are set in their ways
or want to do it by themselves or whatever.
And so we focus on senior folks who have shown a pattern of deep
(14:55):
collaboration and who are chill.And so it's still like strong
opinions, loosely held mentality.
It's, hey, I come with a stance,but oh, let's do something a
different way. OK, great, we'll do it that way.
And so as a result, in seven years, there's never been like
an argument in the business. The mindset is the only thing
that matters is what does the customer want?
Great, let's throw out some ideas.
(15:15):
Great, let's pick one. If that one didn't work, we'll
try the next one next week. So it doesn't matter.
Interesting. And be curious, what's something
you said which really struck me,which I'm 100% believer, but I
think the market doesn't value quite as much.
You said like this idea of I want to keep the company at the
exact right size, like I don't need to hire people for the sake
(15:37):
of hiring people. I'm 100% believer in that.
And I think it's really hard to do.
What I found, at least in my career, was even if you as the
leader have their mentality, eventually you have this creep
where you hire a couple people and then for them to show their
worth, they need to grow their teams.
We had this at Mozilla and it was crazy.
(15:59):
This is the pattern I saw at Twitter.
It's a complete disaster. And so I was at Twitter through
their hyperscale period, I was at Box through its hyperscale
period. And you have just take for
example, engineering managers like your worth as an
engineering manager is the scopeand size of your team.
And so you are highly motivated to grow your team.
We took an explicit step to eliminate that.
(16:22):
So at digits every engineer has a coach.
That coach does not determine what project they work on and
the totally separates out the sort of one on ones HR like org
chart from the actual teams. Because the amount of like
manpower staffing you have on a given project should depend on
(16:43):
how important that project is, not how good that engineering
manager is at hiring. And so we completely dissociate
the two. We create Ng work streams based
and just pick engineers from across the team for each
project. And then we'll blow those up in
a month or two and create a new team.
And so you can move across the code base, across the company,
(17:04):
work on different things and youkeep your one-on-one with your
coach. That's what's your career
conversation. Interesting.
It sounds like you're overcomingsome of the challenges we saw
with the Valve has famously these like very loosely coupled
teams. And essentially, if you can
convince your colleagues that whatever you do is a great idea,
like you get the resources, etcetera, and it turned into
(17:26):
complete 5 DOMS and ultimately. I will say we are top down in
that respect and this I started my career as an engineer at
Apple and Apple has huge pros and cons as you're well aware.
One of the things they did well is very intentional planning of
what priorities they take on when, with what team etcetera.
And so we do top down planning in terms of we are going to
(17:47):
take, we are going to work on these these set projects in
parallel. Right now we usually pick five
projects in parallel and then wehandpick the teams for each and
then it's off and running. Interesting.
A lot of what you just said in terms of the structure reminds
me to when I was in Mozilla. We build Mozilla essentially
after the chaotic system invented by D Hawk.
(18:07):
So this all like idea of like loosely coupled etcetera,
there's a lot of remnants in there and it sounds like it's an
updated version of what D built with VISA originally.
Interesting. Yeah, I'm less familiar with
that history, but yeah, this hasbeen an organic evolution
really. We started experimenting this at
Crashlytics in 2011, and the core engineering team at Digits
(18:29):
is the Crashlytics team. I've worked with these engineers
for 15 years now, and so we've just naturally iterated it as
the different teams have scaled and so on.
And we followed this at Twitter.Twitter's entire developer
platform team ran using this model, and then other teams at
Twitter started copying us and adopting the model.
So it's been really interesting to see it flow through different
orgs. Interesting.
(18:51):
Stepping back a little bit and talking about you as the leader,
as the CEO, as a founder, be curious to hear from you.
How do you think about the evolution of the of your
leadership, the evolution of leading your company as you're
looking into the future as a little bit of inside baseball?
You and I basically just joked alittle bit before we got on the
(19:12):
recording. Literally yesterday I read yet
another quote from Sundar, the CEO of Google, saying that AI is
now so good it will replace he'she will not be surprised if AI
replaces the CEO in just a couple of years and agents are
so good. And then of course in this whole
interview piece, he doesn't get into any of the details what
that goddamn agent is actually even going to do.
(19:35):
But nonetheless, as someone who's building AI, someone I'm
sure is using AII, personally use AI as half my day I spend in
front of an AI now doing all kinds of interesting stuff and
sometimes stupid stuff. But I'd be curious, like, how do
you think about the evolution ofyour role as a leader in an
environment which now has AI data, all kinds of stuff?
(19:58):
Available to you. Yeah, I think this obviously
varies by stage of business. I'd, I'd be curious what Sundar
thinks his job is and how replaceable that is at their
scale. But I'd say for like for most
entrepreneurs, initially your job as the founder CEO is to
secure capital and build a team and build the best team you can.
And those I think are uniquely hard for an AI to replace,
(20:20):
right. Like are investors going to
blindly invest in an AI? Is the AI going to go out and
recruit and get great people whoare going to work for an AI?
So they. Can't just go and meet with it
and hang out. And so is the AI going to go and
win customers on sales calls? Maybe.
Is it going to sit down for the in person dinner?
Questionable. I think in practice, the AI
(20:41):
replacing the CEO is pretty unlikely in terms of as we
scale. Those definitely were my core
jobs in the early days as also with coding, like I spent the
first number of years of digits just heads down coding with the
team because that was the most important thing.
Now I'm much more focused on thego to market and getting our
name out there, talking with customers and so on.
(21:01):
I still try to be involved in everything we build and ship.
And the way I frame it to our team is basically for every
project across the company, I want to be involved at the 10%
phase and the 90% phase. So at 10%, like I want to be
aware of what we're working on, why, what the goal is, right?
Let's make sure we're directionally aligned and then
(21:23):
they're off and running. And then I want to be involved
at the 90% phase where I critique the pixels, we edit the
website copy, I draft some of the blog posts.
We get really in the weeds on some technical architecture
review and so on. Because I feel like that's where
you can add the differentiated Polish and like what makes the
company really special and hold the sort of quality bar and
(21:46):
speed bar as we're focused on racing forward into market.
And that so far has gone really well.
And it it makes sense to the team.
It's an easy concept to remember.
And it's been great to allow me to be directionally aligned and
then get those little pixels right at the end.
And I'm curious about your business, your industry.
You mentioned earlier this thereis a established, very
(22:07):
established players in the industry.
This also seems to be an enormous amount of start up
activity in this space. A lot of three people, three
guys in the garage throwing an NLM together and creating
whatever and then anything in inin between.
I'd be curious, what is your perspective like?
Where do you think? And I'm using the accounting and
(22:29):
finance industry as a proxy because I think it's happening
in pretty much every industry looking at CRM and ERP systems
and God knows what. What do you think is the
evolution of this? I just recently read the
beautiful story of the HP35 pocket calculator, which came
out of an experiment where someone at HP built the thing
for themselves because they wanted to have a pocket
(22:51):
calculator. Classic entrepreneurial story,
but within essentially a three-year period, because the
pocket calculator was so good, it utterly decimated the slide
ruler market. Every company in the slide ruler
market basically went out of business.
I wonder where do you see the dynamics playing in the software
industry? Is it the start-ups will
(23:13):
actually eat the big companies because it is such a
fundamentally different thing tobuild, for example, an AI native
accounting system versus a, oh, we throw some AI on top of the
old thing. Or do you think it's more like a
coexistence? What's your prediction?
This is a really crazy time in the market there.
I agree. There's been a Cambrian
(23:33):
explosion of startups. We saw the same to a similar
extent with the rise of mobile and everyone all of a sudden was
doing mobile stuff with the crypto waves.
Every company had a crypto thingor bolted on a crypto thing.
Of course, most of those went nowhere.
With AII think you're seeing that same sort of just energy
around the space. It's like it creates this whole
(23:54):
new realm of startup possibility.
The vast majority will fail. And I think it is, it's going to
really come down to who is serious about it and going into
detail about it versus basicallydoing a wrapper around open AI
or whatever it might be. Historically, in these sessions,
everyone thinks, oh, the incumbents are going to get
disrupted some of a lot of timesthey just get more powerful and
(24:17):
stronger. And there were questions about
Google survival and then of course, they released Gemini 3
and everyone said, oh, OK, they've got this.
So I think in accounting specifically it's a really
unique industry because it's sufficiently complex where the
none of the foundation companiesare going to focus on this,
right. Open AI can't just spin up an
(24:38):
accounting product next month. And that's not the case in a lot
of other more just text in, textout things like code where it is
a natural capability of their model.
Working with structured double entry accounting data is very
different. And accounting is not a
generative problem space in general.
You don't want your accountant fabricating A transaction.
(24:59):
It's a predictive problem space.You're taking data in and then
structuring it and booking it and classifying it.
And so I think the key question is whether the £800 gorillas in
the space of which there are a couple can adapt quickly enough
for whether start-ups like us can outpace them in
distribution. It's really a technology versus
distribution rate. Interesting, right?
(25:21):
Because they they have their distribution channels locked in.
How defensible are those versus how quickly can the tech
overcome that? And then be curious to hear your
perspective when it comes to AI.There is an interesting
conversation at the moment for awhile now where people make the
argument that AI is such a fundamental shift in how
(25:44):
software operates, how it's being built, what the back ends
are. You mentioned, for example, you
might replace a database, which was the backbone of what we did
in computer science for God knows, 4050 years with something
different like vector databases and like very different
approaches. Is it, are you sharing that
perspective that it's like such a fundamental thing or is it
(26:06):
much more of a, an evolutionary thing where we move from, I
don't know, Sybase to D base to whatever?
Yeah, I think it is a pretty fundamental thing in terms of
new capability. It's not going to make software
go away, obviously. It's going to change the form
factor of the software. And if you look back before Web
(26:26):
2O, before XMLHCTP request, right?
Like you had desktop apps that worked and then you have web
pages that were just static pages, right?
And you had to reload the site in order to get new data.
And that dramatically changed Ajax, dramatically changed how
software worked with AII think that is exactly what's
happening. But we're not seeing it yet
(26:46):
because most companies are just bolting it on and it's here is
my product plus my chat bot. And this has always driven me
crazy because I don't want another thing to talk to, right?
Like. Why do I have to tell your chat
bot to do something? Just do it.
And. So that's what we're trying to
push with Digits. It's like the accounting, the
(27:07):
month end close process is a known process and a known goal
state, right for these businesses.
They want closed books that are complete and accurate.
The AI should just do that right?
Just get me there, don't make mesit there and tell.
Ask your chat bot. Can you please help me close the
books? Of course you want to close the
books. Like what else are you trying to
do? Interesting.
(27:27):
And then you do think I'd be curious to hear your thoughts on
you've built software now you mentioned you started when you
were 12, something we share by the way.
So for me it was C and then assembler which is like really
dates me. No, I was doing a bit of
assembler as well and then the actually The funny thing is I
was learning C but it was on Mac, all the documentation was
(27:50):
in Pascal and so I was reading Pascal and writing C at 12
trying to like translate betweenthe two it.
Just that is hard. That is really hard.
So my my question goes into the direction of as you're building
software now that you have AI outside of what everybody
currently does is which is clearly like vibe coding and
(28:11):
agents and what have you. Are you actually building
software differently? Has something changed in the way
you build the software? That's a really great question.
In some sense, yes to the data model side of things, like you
use different types of data structures and so on.
But I would say largely no, because the vast majority of
(28:33):
software is still systems engineering, right?
And so if you don't hook up yourservers correctly and your APIs
correctly and your authentication layer correctly
and all that stuff, everything else is going to fall apart
anyway. So you expose these tools to the
AI agents, and that's great, butyou still need to build the
software infrastructure to have those tools and have them do the
(28:53):
right thing and not fail and logtheir errors and all of that
stuff that software engineers are used to.
Interesting. And then I'm like curious to
hear looking out into the futurea little bit.
So the predominant computer interface for us is the thing I
still stare at, which still looks exactly like very much
exactly like my very first computer.
And then granted we got phones. Are you envisioning that is
(29:16):
going to change? There's a lot of talk about
yesterday, I was at a ramen place and saw a bunch of younger
folks in the like early 20s playing around with the voice
interface of ChatGPT and they literally did not touch their
phone anymore. They had an hour long, we were
in this restaurant an hour long where they're like in
(29:37):
conversation with ChatGPT just purely in voice.
And I thought this really interesting.
It's like they didn't look at the screens anymore.
Like a very interesting behavioural change.
So I wonder. If you are envisioning it with
this AI transition that we're moving, if we're moving into a
different modality. Definitely to an extent.
Have you played with Whisper? Yes.
(29:59):
So effectively perfect voice dictation.
And that I would say has changedmy life over the past year
because I now voice dictate all of my emails, all my slacks, I
messages. I dictated our 30 page board doc
last quarter and it's brilliant because I can speak at 200 words
a minute and I can't type that fast.
And so it saves so much time. What's really interesting
(30:22):
though, is that's brilliant if you work from home like I do,
right? What a disaster.
Yes. And so it's actually more
constraining. So yes, I think there will be
continued bounds pushing on what's acceptable there.
Interesting. In the same vein, I'd be curious
anything you're particularly excited about at the moment?
(30:43):
Other than your own company, of course.
Other than the company, I'll just say that the capability of
the models have obviously dramatically improved just in
two years. And even flashback to January of
this year, Vibe coding was actually pretty terrible and
like most of the code it wrote was horrible.
And now it's like decent for some things.
And so it's still not perfect byany sense, but the trajectory is
(31:06):
incredible. So I'm very excited about that.
I do think we will hit an architectural wall with the
models. I don't think the current large
language model architecture willget us to AGI.
The context window thing is crazy.
The lack of memory is crazy. Once the model starts
generating, it can't really backtrack.
It's it can pretend to, but likea human, you can say something
(31:27):
and then stop and catch yourselfand think more and then say
something different. And the models can't do that in
forward inference right now. So there's going to have to be
some changes. But I think we're now really on
the path to like very smart models, which is pretty cool.
Interesting. Curious.
There's a big conversation around the large, potentially
(31:51):
large economic impact. And at the same time you hear
anything from, oh, it's eating jobs.
No, it's not eating jobs. It's like it's adding X percent
to GDP. No, it's not adding anything to
GDP. I'd be curious.
And you're in the accounting industry.
You're in the finance industry. What is your like gut?
Where's this going? So it is going to have a
dramatic impact and my belief onthis has actually evolved just
(32:13):
in the past six months. I now believe in accounting the
month end close process, which historically takes accountants
many hours every month per client will be effectively
automated by end of next year. So like 12 months from now and
really crazy, right? Like basically instantaneous 0
day close for most businesses. And but stepping back from that,
(32:34):
there's a macroeconomic principle called the lump of
Labor fallacy, which is right. There's this distinct chunk of
Labor that needs to be done. And then beyond that, there's no
more jobs. That's been proven wrong every
single tech evolution in history, right?
Like people were worried when electricity came out that it
would take jobs, that cars wouldtake jobs, that airplanes would
take jobs, etcetera, etcetera, etcetera.
(32:55):
And so I do think AI will resultin one of the greatest work
displacements or transitions that we've seen, right?
Because a lot of jobs, a lot of work won't need to be done.
You don't need to sit there doing data entry anymore.
Data entry is definitely going away.
That doesn't mean there's not going to be employment.
I think there's going to be a period of a number of years of a
(33:15):
lot of sort of structural friction in the job market and
then it'll get sorted out of, OK, what are the new careers
that people do? Interesting.
This is a really fascinating conversation.
I could go literally for hours, But let me bring this to a
close. And I want to close with a nod
to the incredible Kyle Ryssdale.Mollywood, formerly Molly, I
(33:37):
don't think is at NPR anymore. Kyle still is.
They had a podcast called Make Me Smart, which was my favorite
podcast. They ended that podcast with an
interesting question, which is what's something you thought you
knew that you later found out were wrong about?
So let me ask you that question.All right, I will pick something
completely change of topic out of tech AI, everything.
(33:58):
Let's talk about ice cubes. So my wife's a chemical engineer
and we initially had a huge argument about this.
And then she of course proved mewrong, which is I thought, oh,
you can just cool your drink faster by adding more ice to it
or adding larger ice cubes to itand it won't get diluted as
much. And you even see like a cocktail
bars that use like some massive cube, right?
(34:19):
With that, it's a complete lie. Ice itself does not cool
Anything like frozen ice doesn'ttransfer heat.
It is the melting process that cools your drink.
Ice cool up when it melts, and so by definition, to get a
volume of liquid from X temperature to Y temperature,
(34:40):
you need Z volume of ice to melt.
So it doesn't matter what shape,size, anything your ice is, it's
always going to be the same. Wow, I did not know that.
Yep. OK that is a good one.
Jeff. On that note, thank you so much.
This was an incredible conversation for anyone anyone
(35:01):
running your books. We will put a link to Jeff's
company Digits into the show notes.
Of course. Highly recommend checking that
product out. It will blow your mind.
Jeff, this was incredible. Thank you so much for making
that time and all the best for digits.
I can't wait to see what's up next for you.
Thank you so much, super fun conversation.
This was a blast. Thanks Pascal.
Build for turbulence. This has always driven me crazy
because I don't want another thing to talk to.
But why do I have to tell your chat bot to do something?
Just do it. Hey friends, Pascal here.
Welcome to another episode of Built for Turbulence.
Yep, we've rebranded the. Podcast to match the title of
(00:28):
our next book and. Better capture what we're.
Talking about on the Pod and a fair warning, I was fighting a
bit of a cold during this recording, so if I sound a
little nasal, that's why. Today I'm talking with Jeff
Siebert, founder and CEO of Digits.
The world's first. AI native accounting platform
Jeff's one of those builders who've been in the arena for the
(00:50):
long haul. He.
Previously served as Twitter's head of consumer product.
And if you've seen the Emmy award-winning Netflix
documentary The Social Dilemma, you've seen Jeff.
He was one of the featured voices in that conversation
about technology's impact on society.
Before Digits and Twitter, Jeff Co founded and Let Crush Lytics,
(01:11):
the mobile performance analyticscompany that Twitter acquired in
2013. If you're not.
Familiar with Crashlytics? Here's the scale we're talking
about. It now runs on over 6 billion
monthly active smartphone. That's right, billions with AB.
Google owns it now and it's the market leading crash reporter
(01:31):
for both iOS and Android. Just a self-taught programmer
who released his first app at 12years old, then went on to
Stanford where he earned his BS in Computer Science.
He's also an active Angel investor.
He backed more than 100 startupsand Insider named him one of
their top 100 CPCS in 2021. What I love about this
(01:53):
conversation is that Jeff's beenon every side of the table.
Founder, operator at scale, investor, and now founder again.
He knows what it takes to build in turbulence and we had a
really rich conversation about AI product and what it means to
build companies that matter. Jeff, I'm generally stoked to
have you on the podcast. Welcome.
(02:14):
Pascal, it's so great to be here.
Thanks for having me. I have to start with a question.
So you run Digits and I want to hear a little bit more about
what Digits actually does. But it is an AI native
accounting platform. You come from a very different
background, Twitter and you did a start up before that which
(02:34):
very successful and also you were in the social dilemma.
We need to talk about that at one point, but how do you go
from that to accounting? Yes, great question.
So yeah, I had no background in finance or accounting.
My life has actually been code. I taught myself C when I was 12.
I fell in love with the concept of being able to program a
(02:56):
computer and write the code onceand have it run 1000 or
1,000,000 or a billion times andthe output was uncorrelated with
the input. It was like blow mind blowing to
me as a middle schooler. And so I just, I just spent my
whole life coding. And so Digits is actually my
third start up. I did a document collaboration
company back in O 7. We got acquired by Box.
I built Crashlytics in the mobile developer tool space.
(03:19):
We got acquired by Twitter and Digits actually came from that
journey of those companies. And then also what I saw at
Twitter and what struck me was the difference in data quality
between product engineering and finance.
And on the product engineering side, you have Google Analytics,
you have AB testing tools, you have Grafana, right?
(03:39):
All these live dashboards that tell you exactly what your users
are doing, what your servers aredoing, and you can figure out
what's going on. And then on the finance side,
like I, as the founder and CEO, was waiting two to three weeks
to get a black and white PNL andbalance sheet from our
accountant. And I was like, there's clearly
something broken here, right? Like, why don't I have a live
(04:00):
finance dashboard? And so that was honestly the
genesis of digits. And at first I thought it was
just a small business issue. And I was like, oh, I'm sure big
companies have got this sorted out.
And then at Twitter, as head of product, I went to corporate
finance 1/4 and I was like, what's our budget?
I'd like to run an event for theteam.
And their actual answer was, oh,we haven't run those books yet.
(04:21):
Give us 3 weeks and I was like, you have 100 people in corporate
finance, what are you doing? And so that that was, it
basically had to fall down the accounting rabbit hole, talk
with hundreds of accountants, quickly realized that it wasn't
their fault, It wasn't the people like accountants are very
hard working, very disciplined, diligent, etcetera.
They were being held back by thesoftware.
(04:43):
And so we realized in 2018, we had to start from scratch and
build a new general Ledger from the ground up designed for real
time finance. Let me ask you a question.
If I zoom out from the comment you made about data, I find this
actually an interesting question.
In a world which is increasinglyuncertain, how much does past?
(05:06):
Because most all of our data is past data, how much do you think
from a business perspective, from a leadership perspective,
from a leader's perspective, does past data actually help you
in a world which is increasinglyuncertain?
I think that is a really thoughtful question.
I, I think it's still an essential building block because
(05:27):
businesses tend to have momentum.
And so yes, the world is uncertain, but if you're going
to be informed by anything, you might as well anchor it in what
you have done over the past couple periods to at least give
you a starting point. And one of the challenge, one of
an example of this is you talk with small business owners all
across the country, almost none of them have a budget.
(05:47):
And it's largely because they can't afford to pay their
account and to create a budget for them.
And even if they did, maintaining it is so tedious
today, right? Like you have to measure it
against the actuals every month and update it and adjust it and
so on. Imagine a world where actually
your AI was able to just give you a budget based on last
month's spend. And it's like, that's likely
(06:07):
what you're going to spend this month unless you change
something. And so then you could just drag
a little slider and be like, oh,I want to increase marketing
spend or I want to hire another person, or I want to sell off
one of my trucks or whatever it is for my business.
Like at least it gives you a starting point.
I think. Otherwise you're flying blind.
And yes, the rate of change in speed of business has increased
dramatically, but you still needto anchor yourself somewhere.
(06:30):
Interesting. And let me ask you another
question about digits, because you're playing in an arena which
has, I wouldn't even say £800 gorillas.
They're like probably beyond that, right?
The likes of the QuickBooks, etcetera.
What makes you compete? What makes you start a company
(06:50):
in an arena which feels so incredibly entrenched?
Yes. So definitely I don't know if
it's a rational choice, but it, this is my third company I had
sold the prior to Crashlytics did achieve insane scale.
So we grew to 300 million phonesin 12 months.
Today Crashlytics runs on 6 billion smartphones on Earth,
(07:12):
basically every active smartphone and we were in over
1,000,000 apps. And so we, we really were like
the de facto performance monitoring tool for the mobile
ecosystem. And I guess that gave me the
confidence that we can build something that truly scales and
what's a bigger market and like a, an even more compelling
opportunity to make changes. If you can make finance and
(07:34):
accounting real time and visual and intuitive, that becomes
pretty powerful. So I've never been afraid of
tackling a big challenge. I think if you can design A
better product and of course, pair it with distribution is the
key. And I wouldn't be worried about
the status quo. It's let's imagine what the
world has to look like in 10 years and just only focus on the
(07:55):
path to get there. Interesting.
And then I'd be curious to understand, you have built a
company. So first of all, your company is
AI native. So you literally built a company
just with a little bit of AI sprinkling on, like you built a
company from the ground up as being a company which is set to
thrive in an AI world. You've built companies before.
(08:18):
You also worked for a very largecompany as a leader.
How is leading different for youin a company which is AI native?
Oh, interesting. So this has been core back to
when we started digits in 2018. And I actually describe this as
ML native more than AI native, right?
Because AI now has been this hype term for the past couple
years. But way back to Twitter as had a
(08:41):
product, I launched the algorithmic timeline.
So the transition from just a straight chronological to now
it's ranked and sorted, right. And that was with Facebook,
which had done it right before we did one of the first global
deployments of machine learning tech.
And I saw the power of ML, I also saw all the downsides in
terms of misinformation and polarization and so on.
(09:03):
That became a social dilemma film.
But I wanted to build a company where we could use that
technology but avoid the downsides.
And I was like, accounting is great.
Like you, everyone would love itif you could automate the
tedium. And so there's not real risk of
like misinformation or whatever.It's let's just do the work.
And so we really structured morethan just the company, but the
(09:23):
actual architecture around beingML native.
So like going back to the fundamental data design of
Digits, we aren't a traditional relational database.
Like when you load the Digits dashboard, you're not hitting a
database. We actually have a whole
different data architecture and we render the dashboard in 100
milliseconds. We designed that for machine
learning algorithms to better manipulate.
(09:45):
And so I've really thought aboutit in terms of product
capability and innovation more so than how to run the company
approaches. We used to run the business I
actually think came from Crashlytics back in 2011.
So we do a bunch of stuff very differently.
We run the whole thing on a weekly Sprint.
I've run on all hands every 48 hours for the past seven years.
(10:07):
It's yeah. Remind me, how big is your
company today? We are 65 people.
OK. So if you project this out
orders of magnitude like I am very sure you will get to the
650 people company and beyond, how do you think you can
maintain that if you can maintain it?
(10:28):
Like the? Speed, the agility, etcetera.
It's a great question and something I think about
constantly. I actually don't know if we'll
need to get to 650 people. My goal is to have the smallest
team at every stage of the market opportunity.
And you look at teams like WhatsApp, I think we're around
20 people when they were acquired by Facebook Enterprise
(10:49):
Software. You do need a few more hands,
but I don't think you need hundreds or thousands.
And so we'll see. But I think it's, we've already
scaled 2 orders of magnitude. So like way back at the start of
the company, there's just a few folks around the table and every
single person speaks at every all hands sync, right and says
what they're doing and where they need help.
And then it fractalizes to individual like work teams, work
(11:12):
streams and then it fractalizes to sort of departments of, OK,
go to markets doing this, we need help with this, whatever.
So I think you can keep fractalizing it out and it works
pretty well. And it's more about a mindset of
being able to run the whole business on a weekly cadence
where we're all setting goal goals Monday morning.
We all do show and tell Friday afternoon and show off what we
(11:34):
did that week and that just mindset creates this momentum in
the business where and it allowsyou to just tack every single
week if you need to adapt to market conditions.
In this context, how do you do long term planning?
So we plan in periods we call horizons.
They are horizons because it is literally as far as we can see.
(11:56):
And right now a horizon is 4 to 5 weeks and Yep.
And so we plan out what looks like effectively the next month
of what are our goals, where do we want to be.
And it's very lightweight. It's just a couple bullets for
each team of what do you want tohave achieved in the next 4 to 5
weeks. And then we make sure that the
weekly goals you set each week are directionally aligned with
(12:18):
where you want to be in that horizon.
I'm curious for your people, I totally can see how this works
in your head. Does this work for your people?
I find one of the challenges I find with very short planning
horizons which deck you. And I love that some people like
to have more of a much longer planning horizon as a but how do
(12:41):
you know if I perform well in myperformance review?
Like that kind of thing, like sitting down at the 1st of
January and like determining what do I want to achieve this
year. Have you come across and said
like how do you manage that? It's a it's definitely a very
different culture and certainly for some of those folks, a
company like Digits won't be a fit for them.
We do not do performance reviewsperiod.
(13:03):
And that's because in my career I've never seen it effective.
Twitter did them at twice at once to twice a year and it was
a nightmare process and took everyone's time and then nothing
really happens. We prefer to do just continual
one on ones. We have a coaching mentality
where your manager is really responsible for sharing, like
helping guide you in your careerand what are your goals?
What do you want to achieve? OK, let's try to arrange
(13:25):
opportunities for you to get there.
And then we focus a lot on team collaboration and so do not
reward individual contribution whatsoever.
Everything at the company is done in minimum pairs of people
because like it allows you to beway more thoughtful in the
approach. It makes it way more fun to work
on projects together versus by yourself.
And so we don't reward any individual achievement.
(13:48):
We recognize folks who help other people in other teams.
Interesting. Presumably you've got a bunch of
engineers in your company though.
Yes, my experience was, particularly when I was at
Mozilla working with some reallysmart engineers, that there is a
certain class of engineers who are lone eagles.
(14:09):
Yes, they don't hire them. Oh, OK, interesting.
So yeah, we have it's, this is these are great line of
questions because it goes back to the whole philosophy of the
business and what you're aiming for.
So way back in our interview process, we actually spend the
bulk of the time in the interview process not on
technical skill, but on collaboration and career history
and progression and more of a culture style interview.
(14:32):
And so we actively select against the more cowboy coder
style engineers, the Lone Eagle style engineers.
And the way I describe what we look for is senior folks.
We generally only hire senior talent across the company.
The danger with that is you get people who are set in their ways
or want to do it by themselves or whatever.
And so we focus on senior folks who have shown a pattern of deep
(14:55):
collaboration and who are chill.And so it's still like strong
opinions, loosely held mentality.
It's, hey, I come with a stance,but oh, let's do something a
different way. OK, great, we'll do it that way.
And so as a result, in seven years, there's never been like
an argument in the business. The mindset is the only thing
that matters is what does the customer want?
Great, let's throw out some ideas.
(15:15):
Great, let's pick one. If that one didn't work, we'll
try the next one next week. So it doesn't matter.
Interesting. And be curious, what's something
you said which really struck me,which I'm 100% believer, but I
think the market doesn't value quite as much.
You said like this idea of I want to keep the company at the
exact right size, like I don't need to hire people for the sake
(15:37):
of hiring people. I'm 100% believer in that.
And I think it's really hard to do.
What I found, at least in my career, was even if you as the
leader have their mentality, eventually you have this creep
where you hire a couple people and then for them to show their
worth, they need to grow their teams.
We had this at Mozilla and it was crazy.
(15:59):
This is the pattern I saw at Twitter.
It's a complete disaster. And so I was at Twitter through
their hyperscale period, I was at Box through its hyperscale
period. And you have just take for
example, engineering managers like your worth as an
engineering manager is the scopeand size of your team.
And so you are highly motivated to grow your team.
We took an explicit step to eliminate that.
(16:22):
So at digits every engineer has a coach.
That coach does not determine what project they work on and
the totally separates out the sort of one on ones HR like org
chart from the actual teams. Because the amount of like
manpower staffing you have on a given project should depend on
(16:43):
how important that project is, not how good that engineering
manager is at hiring. And so we completely dissociate
the two. We create Ng work streams based
and just pick engineers from across the team for each
project. And then we'll blow those up in
a month or two and create a new team.
And so you can move across the code base, across the company,
(17:04):
work on different things and youkeep your one-on-one with your
coach. That's what's your career
conversation. Interesting.
It sounds like you're overcomingsome of the challenges we saw
with the Valve has famously these like very loosely coupled
teams. And essentially, if you can
convince your colleagues that whatever you do is a great idea,
like you get the resources, etcetera, and it turned into
(17:26):
complete 5 DOMS and ultimately. I will say we are top down in
that respect and this I started my career as an engineer at
Apple and Apple has huge pros and cons as you're well aware.
One of the things they did well is very intentional planning of
what priorities they take on when, with what team etcetera.
And so we do top down planning in terms of we are going to
(17:47):
take, we are going to work on these these set projects in
parallel. Right now we usually pick five
projects in parallel and then wehandpick the teams for each and
then it's off and running. Interesting.
A lot of what you just said in terms of the structure reminds
me to when I was in Mozilla. We build Mozilla essentially
after the chaotic system invented by D Hawk.
(18:07):
So this all like idea of like loosely coupled etcetera,
there's a lot of remnants in there and it sounds like it's an
updated version of what D built with VISA originally.
Interesting. Yeah, I'm less familiar with
that history, but yeah, this hasbeen an organic evolution
really. We started experimenting this at
Crashlytics in 2011, and the core engineering team at Digits
(18:29):
is the Crashlytics team. I've worked with these engineers
for 15 years now, and so we've just naturally iterated it as
the different teams have scaled and so on.
And we followed this at Twitter.Twitter's entire developer
platform team ran using this model, and then other teams at
Twitter started copying us and adopting the model.
So it's been really interesting to see it flow through different
orgs. Interesting.
(18:51):
Stepping back a little bit and talking about you as the leader,
as the CEO, as a founder, be curious to hear from you.
How do you think about the evolution of the of your
leadership, the evolution of leading your company as you're
looking into the future as a little bit of inside baseball?
You and I basically just joked alittle bit before we got on the
(19:12):
recording. Literally yesterday I read yet
another quote from Sundar, the CEO of Google, saying that AI is
now so good it will replace he'she will not be surprised if AI
replaces the CEO in just a couple of years and agents are
so good. And then of course in this whole
interview piece, he doesn't get into any of the details what
that goddamn agent is actually even going to do.
(19:35):
But nonetheless, as someone who's building AI, someone I'm
sure is using AII, personally use AI as half my day I spend in
front of an AI now doing all kinds of interesting stuff and
sometimes stupid stuff. But I'd be curious, like, how do
you think about the evolution ofyour role as a leader in an
environment which now has AI data, all kinds of stuff?
(19:58):
Available to you. Yeah, I think this obviously
varies by stage of business. I'd, I'd be curious what Sundar
thinks his job is and how replaceable that is at their
scale. But I'd say for like for most
entrepreneurs, initially your job as the founder CEO is to
secure capital and build a team and build the best team you can.
And those I think are uniquely hard for an AI to replace,
(20:20):
right. Like are investors going to
blindly invest in an AI? Is the AI going to go out and
recruit and get great people whoare going to work for an AI?
So they. Can't just go and meet with it
and hang out. And so is the AI going to go and
win customers on sales calls? Maybe.
Is it going to sit down for the in person dinner?
Questionable. I think in practice, the AI
(20:41):
replacing the CEO is pretty unlikely in terms of as we
scale. Those definitely were my core
jobs in the early days as also with coding, like I spent the
first number of years of digits just heads down coding with the
team because that was the most important thing.
Now I'm much more focused on thego to market and getting our
name out there, talking with customers and so on.
(21:01):
I still try to be involved in everything we build and ship.
And the way I frame it to our team is basically for every
project across the company, I want to be involved at the 10%
phase and the 90% phase. So at 10%, like I want to be
aware of what we're working on, why, what the goal is, right?
Let's make sure we're directionally aligned and then
(21:23):
they're off and running. And then I want to be involved
at the 90% phase where I critique the pixels, we edit the
website copy, I draft some of the blog posts.
We get really in the weeds on some technical architecture
review and so on. Because I feel like that's where
you can add the differentiated Polish and like what makes the
company really special and hold the sort of quality bar and
(21:46):
speed bar as we're focused on racing forward into market.
And that so far has gone really well.
And it it makes sense to the team.
It's an easy concept to remember.
And it's been great to allow me to be directionally aligned and
then get those little pixels right at the end.
And I'm curious about your business, your industry.
You mentioned earlier this thereis a established, very
(22:07):
established players in the industry.
This also seems to be an enormous amount of start up
activity in this space. A lot of three people, three
guys in the garage throwing an NLM together and creating
whatever and then anything in inin between.
I'd be curious, what is your perspective like?
Where do you think? And I'm using the accounting and
(22:29):
finance industry as a proxy because I think it's happening
in pretty much every industry looking at CRM and ERP systems
and God knows what. What do you think is the
evolution of this? I just recently read the
beautiful story of the HP35 pocket calculator, which came
out of an experiment where someone at HP built the thing
for themselves because they wanted to have a pocket
(22:51):
calculator. Classic entrepreneurial story,
but within essentially a three-year period, because the
pocket calculator was so good, it utterly decimated the slide
ruler market. Every company in the slide ruler
market basically went out of business.
I wonder where do you see the dynamics playing in the software
industry? Is it the start-ups will
(23:13):
actually eat the big companies because it is such a
fundamentally different thing tobuild, for example, an AI native
accounting system versus a, oh, we throw some AI on top of the
old thing. Or do you think it's more like a
coexistence? What's your prediction?
This is a really crazy time in the market there.
I agree. There's been a Cambrian
(23:33):
explosion of startups. We saw the same to a similar
extent with the rise of mobile and everyone all of a sudden was
doing mobile stuff with the crypto waves.
Every company had a crypto thingor bolted on a crypto thing.
Of course, most of those went nowhere.
With AII think you're seeing that same sort of just energy
around the space. It's like it creates this whole
(23:54):
new realm of startup possibility.
The vast majority will fail. And I think it is, it's going to
really come down to who is serious about it and going into
detail about it versus basicallydoing a wrapper around open AI
or whatever it might be. Historically, in these sessions,
everyone thinks, oh, the incumbents are going to get
disrupted some of a lot of timesthey just get more powerful and
(24:17):
stronger. And there were questions about
Google survival and then of course, they released Gemini 3
and everyone said, oh, OK, they've got this.
So I think in accounting specifically it's a really
unique industry because it's sufficiently complex where the
none of the foundation companiesare going to focus on this,
right. Open AI can't just spin up an
(24:38):
accounting product next month. And that's not the case in a lot
of other more just text in, textout things like code where it is
a natural capability of their model.
Working with structured double entry accounting data is very
different. And accounting is not a
generative problem space in general.
You don't want your accountant fabricating A transaction.
(24:59):
It's a predictive problem space.You're taking data in and then
structuring it and booking it and classifying it.
And so I think the key question is whether the £800 gorillas in
the space of which there are a couple can adapt quickly enough
for whether start-ups like us can outpace them in
distribution. It's really a technology versus
distribution rate. Interesting, right?
(25:21):
Because they they have their distribution channels locked in.
How defensible are those versus how quickly can the tech
overcome that? And then be curious to hear your
perspective when it comes to AI.There is an interesting
conversation at the moment for awhile now where people make the
argument that AI is such a fundamental shift in how
(25:44):
software operates, how it's being built, what the back ends
are. You mentioned, for example, you
might replace a database, which was the backbone of what we did
in computer science for God knows, 4050 years with something
different like vector databases and like very different
approaches. Is it, are you sharing that
perspective that it's like such a fundamental thing or is it
(26:06):
much more of a, an evolutionary thing where we move from, I
don't know, Sybase to D base to whatever?
Yeah, I think it is a pretty fundamental thing in terms of
new capability. It's not going to make software
go away, obviously. It's going to change the form
factor of the software. And if you look back before Web
(26:26):
2O, before XMLHCTP request, right?
Like you had desktop apps that worked and then you have web
pages that were just static pages, right?
And you had to reload the site in order to get new data.
And that dramatically changed Ajax, dramatically changed how
software worked with AII think that is exactly what's
happening. But we're not seeing it yet
(26:46):
because most companies are just bolting it on and it's here is
my product plus my chat bot. And this has always driven me
crazy because I don't want another thing to talk to, right?
Like. Why do I have to tell your chat
bot to do something? Just do it.
And. So that's what we're trying to
push with Digits. It's like the accounting, the
(27:07):
month end close process is a known process and a known goal
state, right for these businesses.
They want closed books that are complete and accurate.
The AI should just do that right?
Just get me there, don't make mesit there and tell.
Ask your chat bot. Can you please help me close the
books? Of course you want to close the
books. Like what else are you trying to
do? Interesting.
(27:27):
And then you do think I'd be curious to hear your thoughts on
you've built software now you mentioned you started when you
were 12, something we share by the way.
So for me it was C and then assembler which is like really
dates me. No, I was doing a bit of
assembler as well and then the actually The funny thing is I
was learning C but it was on Mac, all the documentation was
(27:50):
in Pascal and so I was reading Pascal and writing C at 12
trying to like translate betweenthe two it.
Just that is hard. That is really hard.
So my my question goes into the direction of as you're building
software now that you have AI outside of what everybody
currently does is which is clearly like vibe coding and
(28:11):
agents and what have you. Are you actually building
software differently? Has something changed in the way
you build the software? That's a really great question.
In some sense, yes to the data model side of things, like you
use different types of data structures and so on.
But I would say largely no, because the vast majority of
(28:33):
software is still systems engineering, right?
And so if you don't hook up yourservers correctly and your APIs
correctly and your authentication layer correctly
and all that stuff, everything else is going to fall apart
anyway. So you expose these tools to the
AI agents, and that's great, butyou still need to build the
software infrastructure to have those tools and have them do the
(28:53):
right thing and not fail and logtheir errors and all of that
stuff that software engineers are used to.
Interesting. And then I'm like curious to
hear looking out into the futurea little bit.
So the predominant computer interface for us is the thing I
still stare at, which still looks exactly like very much
exactly like my very first computer.
And then granted we got phones. Are you envisioning that is
(29:16):
going to change? There's a lot of talk about
yesterday, I was at a ramen place and saw a bunch of younger
folks in the like early 20s playing around with the voice
interface of ChatGPT and they literally did not touch their
phone anymore. They had an hour long, we were
in this restaurant an hour long where they're like in
(29:37):
conversation with ChatGPT just purely in voice.
And I thought this really interesting.
It's like they didn't look at the screens anymore.
Like a very interesting behavioural change.
So I wonder. If you are envisioning it with
this AI transition that we're moving, if we're moving into a
different modality. Definitely to an extent.
Have you played with Whisper? Yes.
(29:59):
So effectively perfect voice dictation.
And that I would say has changedmy life over the past year
because I now voice dictate all of my emails, all my slacks, I
messages. I dictated our 30 page board doc
last quarter and it's brilliant because I can speak at 200 words
a minute and I can't type that fast.
And so it saves so much time. What's really interesting
(30:22):
though, is that's brilliant if you work from home like I do,
right? What a disaster.
Yes. And so it's actually more
constraining. So yes, I think there will be
continued bounds pushing on what's acceptable there.
Interesting. In the same vein, I'd be curious
anything you're particularly excited about at the moment?
(30:43):
Other than your own company, of course.
Other than the company, I'll just say that the capability of
the models have obviously dramatically improved just in
two years. And even flashback to January of
this year, Vibe coding was actually pretty terrible and
like most of the code it wrote was horrible.
And now it's like decent for some things.
And so it's still not perfect byany sense, but the trajectory is
(31:06):
incredible. So I'm very excited about that.
I do think we will hit an architectural wall with the
models. I don't think the current large
language model architecture willget us to AGI.
The context window thing is crazy.
The lack of memory is crazy. Once the model starts
generating, it can't really backtrack.
It's it can pretend to, but likea human, you can say something
(31:27):
and then stop and catch yourselfand think more and then say
something different. And the models can't do that in
forward inference right now. So there's going to have to be
some changes. But I think we're now really on
the path to like very smart models, which is pretty cool.
Interesting. Curious.
There's a big conversation around the large, potentially
(31:51):
large economic impact. And at the same time you hear
anything from, oh, it's eating jobs.
No, it's not eating jobs. It's like it's adding X percent
to GDP. No, it's not adding anything to
GDP. I'd be curious.
And you're in the accounting industry.
You're in the finance industry. What is your like gut?
Where's this going? So it is going to have a
dramatic impact and my belief onthis has actually evolved just
(32:13):
in the past six months. I now believe in accounting the
month end close process, which historically takes accountants
many hours every month per client will be effectively
automated by end of next year. So like 12 months from now and
really crazy, right? Like basically instantaneous 0
day close for most businesses. And but stepping back from that,
(32:34):
there's a macroeconomic principle called the lump of
Labor fallacy, which is right. There's this distinct chunk of
Labor that needs to be done. And then beyond that, there's no
more jobs. That's been proven wrong every
single tech evolution in history, right?
Like people were worried when electricity came out that it
would take jobs, that cars wouldtake jobs, that airplanes would
take jobs, etcetera, etcetera, etcetera.
(32:55):
And so I do think AI will resultin one of the greatest work
displacements or transitions that we've seen, right?
Because a lot of jobs, a lot of work won't need to be done.
You don't need to sit there doing data entry anymore.
Data entry is definitely going away.
That doesn't mean there's not going to be employment.
I think there's going to be a period of a number of years of a
(33:15):
lot of sort of structural friction in the job market and
then it'll get sorted out of, OK, what are the new careers
that people do? Interesting.
This is a really fascinating conversation.
I could go literally for hours, But let me bring this to a
close. And I want to close with a nod
to the incredible Kyle Ryssdale.Mollywood, formerly Molly, I
(33:37):
don't think is at NPR anymore. Kyle still is.
They had a podcast called Make Me Smart, which was my favorite
podcast. They ended that podcast with an
interesting question, which is what's something you thought you
knew that you later found out were wrong about?
So let me ask you that question.All right, I will pick something
completely change of topic out of tech AI, everything.
(33:58):
Let's talk about ice cubes. So my wife's a chemical engineer
and we initially had a huge argument about this.
And then she of course proved mewrong, which is I thought, oh,
you can just cool your drink faster by adding more ice to it
or adding larger ice cubes to itand it won't get diluted as
much. And you even see like a cocktail
bars that use like some massive cube, right?
(34:19):
With that, it's a complete lie. Ice itself does not cool
Anything like frozen ice doesn'ttransfer heat.
It is the melting process that cools your drink.
Ice cool up when it melts, and so by definition, to get a
volume of liquid from X temperature to Y temperature,
(34:40):
you need Z volume of ice to melt.
So it doesn't matter what shape,size, anything your ice is, it's
always going to be the same. Wow, I did not know that.
Yep. OK that is a good one.
Jeff. On that note, thank you so much.
This was an incredible conversation for anyone anyone
(35:01):
running your books. We will put a link to Jeff's
company Digits into the show notes.
Of course. Highly recommend checking that
product out. It will blow your mind.
Jeff, this was incredible. Thank you so much for making
that time and all the best for digits.
I can't wait to see what's up next for you.
Thank you so much, super fun conversation.
This was a blast. Thanks Pascal.
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.
Las Culturistas with Matt Rogers and Bowen Yang
Ding dong! Join your culture consultants, Matt Rogers and Bowen Yang, on an unforgettable journey into the beating heart of CULTURE. Alongside sizzling special guests, they GET INTO the hottest pop-culture moments of the day and the formative cultural experiences that turned them into Culturistas. Produced by the Big Money Players Network and iHeartRadio.
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by audiochuck Media Company.