June 4, 2024 • 48 mins
In this episode, we dive deep into the world of AI-driven productivity with Gabriel Hubert, the Founding CEO of Dust. Dust is a Paris-based SaaS scale-up that's redefining team collaboration and efficiency through the power of generative AI. Supported by global venture capital giants like Sequoia, Dust is on a mission to break down information silos and empower teams to thrive in a fast-paced, data-driven world.
Gabriel shares the innovative ways Dust leverages large language models and text transformer models to act as the glue between various forms of unstructured text and data. By doing so, Dust ensures that the right people have access to the right information at the right time, eliminating the need for endless meetings and fostering a more autonomous and purposeful work environment.
We explore how Dust's adaptable platform meets the diverse needs of different teams and industries, providing high-performance, customizable AI assistants tailored to specific use cases. Gabriel discusses how the platform integrates seamlessly into existing workflows, whether as a web app, Slack bot, or API, making it an indispensable tool for fast-moving teams.
Discover the early adopters of Dust's revolutionary technology, including companies like Alan Healthcare, Qonto, and Payfit. These organizations, characterized by rapid growth and a commitment to innovation, are harnessing the power of Dust to navigate internal information chaos and stay ahead in their respective fields.
Gabriel also delves into the metrics used to measure Dust's impact on team performance and efficiency, emphasizing the importance of both quantitative and qualitative feedback. He highlights the potential productivity gains across various job functions, not just software development, and how Dust aims to free up employees' time for higher-value tasks.
The conversation takes an introspective turn as Gabriel reflects on his journey from Silicon Valley to Paris, drawing lessons from his Stanford experiences and also from his time at Stripe. He shares valuable insights on maintaining a high hiring bar, building a culture of transparency, and fostering a diverse and inclusive team environment.
Finally, we explore the vibrant startup ecosystem in Paris and the unique advantages it offers for companies like Dust. And Gabriel outlines his vision for the future of AI as one of collaboration and augmentation, where humans and machines work together to achieve greater efficiency and innovation.
Join us for an inspiring and thought-provoking discussion on how Dust is transforming team productivity and what it takes to build a successful AI-driven startup in today's dynamic tech landscape. For more AI driven productivity insights head over to https://dust.tt/ & for guidance on hiring senior talent at software and AI startups and scaleups check out https://alpinasearch.com/
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
Special guest today is Gabriel Hubert, founding CEO at Dust,
a Paris-based SaaS scale-up funded by major global VCs like Sequoia,
and they leverage generative AI to transform how teams collaborate and thrive.
So, Gabriel, welcome to the Startup to Scale-Up game plan. Thank you very much for having me on, Gary.
(00:22):
You're most welcome. Now, your tagline, or rather Dust's tagline,
is cracking team productivity with AI. It's a pretty bold statement.
So could you elaborate on how Dust leverages AI to enhance team productivity
and maybe a little about the way you set your approach apart from traditional productivity tools?
(00:48):
Yeah, I think the main difference that arises from this wave of technology that
everybody's been talking about, large language models,
and more specifically, text transformer models, is their ability to really act
(01:09):
as a glue between forms of unstructured text.
Context and so a lot of productivity in
companies is really given the right people
the right information for them to make decisions as they go on about
their day and i think if you've hired the right people then the best thing you
can do is get out of their way and let them do their work but they do need the
right piece of information to do their work and sometimes that information is
(01:32):
in a siloed piece of software that another function is using that they don't have access to and that
might prevent them from being able to make the decision without calling a meeting.
So I think one of the keys to really changing the way we approach autonomy, empowerment.
And just a more purposeful day-to-day life at the office is to break down those
(01:57):
knowledge silos and have people be able to access and transform at will the
information they might need to make decisions.
And that's really what Dust tries to do. So using the brick of generative artificial
intelligence, we try and make it possible for,
anybody in the company to have access to the right pieces of information they
(02:20):
need to go on about their day.
And how do you, or how does the DUST solution adapt to the diverse needs of
different teams and different industries?
Yeah, that's a great question. I think it's actually one of the reasons that
we've decided to build a product that's fairly horizontal in nature and has
some of the key building blocks that we think are necessary to support a wide range of use cases.
(02:47):
The way the Dust platform is built, it allows for unstructured data flows from a variety of
software solutions to be brought into context and then consumed by purpose-built assistants.
These assistants are customized around specific use cases.
(03:09):
And we've really moved away from the paradigm of having a single assistant try
and do everything for everybody within companies.
We're pretty much on the opposite end of the spectrum. We think that people are quite smart.
You know, when they have access to a toolbox, they can tell the difference between
a hammer and a screwdriver.
And so what they need is a high-performance tool to complete the task or the tasks that are at hand.
(03:35):
And customizing AI assistance to that specific set of use cases is going to
increase their performance and allow them to benefit from just a higher throughput set of tools.
So one of the really core ways we've thought about Dust is, what do fast-moving teams need?
(03:55):
They need fairly horizontal access to the data.
They need relatively easy-to-understand and apprehend building blocks to build
and customize these assistants.
And then they need that product to be able to live as close as possible to where people do their work.
That's why we're available as a web app, but also as a Slack bot in Slack,
(04:17):
or as an API where people can actually build for Dust to be really brought to
the piece of software or the specific tool that they're using day to day.
And who are some of the early adopters?
Yeah, we seem to have struck a chord with companies that fit three characteristics.
(04:37):
One, their size and pace of growth means that internal information chaos is
really a need to solve problem. It's not a vitamin problem.
It's a painkiller problem.
And that generally means that we're talking to companies at least 250,
often 500, sometimes a few thousand employees.
(04:58):
The second piece is that we do have connectors to certain families of software solutions.
And so that does restrict the customers that we're currently able to talk to.
We're a fit for teams that use Slack, GitHub, Notion, Google, Intercom, et cetera.
The list goes on, but it does paint a picture of maybe more recent and in many
(05:22):
cases, tech forward companies.
The third point, And I actually should have started by that because I do think
it's the most important one, is that we really are a fit for companies where leadership,
whether it's the founders or the C-suite, maybe the board has been a source
of inspiration in getting people moving in the right direction,
optimistically and fundamentally convinced that this technology is going to
(05:45):
change things very quickly.
And that they need to adapt, they need to react very quickly,
and they need to deploy generative artificial intelligence solutions within
their teams as quickly as possible.
If those three conditions are met, we seem to strike quite a bit of interest.
In the early group of customers that we're working with, I can mention Allen
(06:05):
Healthcare Insurance, focused on the European market.
Conto, a neobank, also focused on Europe.
At Penny Lane, Payfit, which is a payroll provider, and Penny Lane's accounting software.
And Watershed, for example, based in the US that focuses on carbon accounting
and helping large companies better prepare for the amount of regulation and
(06:26):
climate change risk mitigation that lies ahead of us.
These all have in common that they're generally less than 10 years old,
and the founders really see this as a disruption, not necessarily to their business,
but to the way people are going to be working.
They want to focus on understanding how their most empowered team members should
(06:47):
think about changing the way they work, which will have obviously an impact
on these co-workers, but also maybe on the way they need to plan hiring in the future.
And how do you actually measure the impact of dust on team performance and efficiency?
What metrics are you using?
Yeah, I'd start, first of all, by mentioning that this is a broad and sometimes complex topic.
(07:10):
Oversimplification on a few metrics can maybe miss some points.
I've managed support teams or engineering teams building tooling for support teams in my career.
And the number of tickets processed per minute
is a very tempting metric to try and
check that we're increasing efficiency but it can hide sometimes
a lot of complexity in the fact that you know if that's
(07:31):
really the only the only incentive teams have they might
they might feel the need to reply to a
ticket that's already been solved with a one last email saying thank you very
much and have a lovely day because that can get them another another score on
the tickets per hour so quantifying everything is a little tough but we do work
on quantifying the ability for people to get access to information their need
(07:52):
and generally do their work better so in our pilot phases with customers.
We'll have a quantitative survey with the teams to ask them how much time they
feel they've saved on average per week.
And we think that it's a relatively descriptive metric of what we're trying to do.
At the end of the day, if you're focusing on productivity, you're trying to
(08:13):
get people to do the same amount of stuff in less time or more stuff in the same amount of time.
And we take the results with a pinch of salt. It's also important to try and
understand the qualitative feedback.
In some cases, the gains may not yet be substantial across all teams,
but the way in which they're able to delegate the more mundane tasks to an assistant
(08:36):
or an automated system gives people the impression that they're taking back
control of their day and able to focus on higher value-add tasks.
And that's also something that we try and look at.
Our ambition, if you look at most commonly shared surveys among software developers
that have used Copilot for almost more than a year now,
(08:56):
is that we can get within 25%, 30%, sometimes 35% gains in productivity.
And we really think the number of jobs that are going to see those gains is
not limited to software development. Of course, coding is an obvious candidate
because the human-machine interface is what these people spend their day working in.
(09:18):
And so making the human-machine interface that much snappier directly correlates
to productivity gains for them.
But we really think that the amount of time people spend looking for information,
formatting information that's been collected.
Translating between internal languages pieces of information that they might
not initially understand,
The marketing team that's trying to understand what the engineering team has
(09:40):
shipped or the engineering team that's trying to pass through a sales call and
see if there are actually some new and unsuspected feature requests from a key customer.
These are all use cases where the unstructured flows of information can be structured,
can be structured on the fly upon request with a special kind of filtering.
(10:03):
And that's what we really think is powerful. So again, North Star is to be shaving
significant amounts of time off people's day-to-day tasks and hopefully moving
them to a higher value-add mix of tasks day in, day out.
Now, much of your career was spent in product management in Silicon Valley.
You also did your master's at Stanford.
(10:24):
So you followed a somewhat similar path to Steve Jobs.
I'm not sure if you also dropped out at Stanford. Obviously, Steve famously did.
How have you leveraged your product management experience and your Silicon Valley
experience to become a startup founder in Europe?
Yeah. I am. Well, so historically, I was actually a startup founder in Europe
(10:46):
before my product management experience in the US.
Graduating from Stanford was a pretty unique experience.
I was in applied mathematics there. And it really gave me a taste of just how
close the academic and private sector ecosystems were in that specific part of the world,
and how open and dynamic the conversations were.
(11:08):
Being able to go to talks with startup and scale-up founders that have been
very successful, meeting investors and demystifying the job of a venture capitalist early on.
Those were amazing opportunities that really helped me understand that the people
starting some of the most successful companies that we now know of or use the
products of every day are generally very ambitious individuals,
(11:30):
but also people who have access to an ecosystem and in many cases,
reflexes and principles that help them think differently about the way to to solve problems.
In my second experience in Silicon Valley, so after the acquisition of a company
that we'd started in Europe in 2012, the company was acquired in 2014 by Stripe,
(11:51):
a payments processor based in San Francisco, and we'd just moved to San Francisco.
So we spent a few years working at Stripe after that acquisition.
I think what was quite clear and very obvious was that the scale of your ambition
does forge the way in which you make certain decisions.
The same situation can be resolved quite differently depending on the attitude
(12:16):
and the mindset that you bring to the problem and the ambition that you bring
to solving it. Who are you solving it for?
How often are you going to be solving it? How automated do you want the solution to be?
All these are sub questions that really can be drawn from some original questions
like, all right, what kind of company do you want to build?
I think one of the obvious separations is that in many situations,
(12:38):
you could have a consulting style answer to people's problems.
I'll build you this one-off piece of software that fits your needs,
and I'll be able to bill you by the hour, bill you by the day,
and I'll have a very happy customer if I keep working on that.
But the ability to design a product that is going to solve the problems of this
one customer, but maybe also the problems of many other customers to be is really
(13:01):
where I'd say product thinking comes into play.
So to give some.
Often quoted examples of the mindset that I think was very much put in the spotlight
in Silicon Valley, an obsession about the customer, an obsession about the user.
What are the users actually doing? What problem are they trying to solve?
(13:22):
And how does your solution fit in their way of seeing how to solve the problem?
Clay Christensen, one of the deep thinkers on product discovery,
Discovery always mentioned the example of the smoothie at Starbucks,
which was an elegant solution to the problem of being able to have a breakfast
on the go while not getting out of your car and not getting your clothes dirty.
(13:46):
So the problem is, you know, you want to get your protein on the way to work.
The solution is a smoothie that fits in the cup holder in your car and get to that mix.
And I think that's a really vivid illustration of, you know,
how close you have to be to your users to try and not mistake what they're describing
as their problem for what their actual problem is.
(14:06):
Because otherwise, the risk is obviously that, you know, Ford famously,
I think, was quoted saying, if I built what people were asking,
I would have built faster horses.
But of course, you need to try and extrapolate from the description that people
have of the job they're trying to get done and move to a solution that elegantly covers that,
but also maybe takes out some of the intermediary steps for them.
(14:30):
So yeah, that's one example.
I think another one that's pretty key is the importance of hiring a very high
quality team And the fact that maintaining the bar at recruiting and during
the development of people's career is one of the best favors you can do to your team as it grows.
(14:51):
Great people have many options on the table. They could be doing anything.
And so if they choose to come and work with you, they're going to do it because
they like the product that you're building.
They like the mission that you're trying to achieve, but also they like the
people that they're working with. And the people that they're working with is
really one of these assets that can deteriorate over time if you don't pay attention
to continuously maintaining that high hiring bar.
(15:15):
You know, the famous quote about A's hire A's and B's tend to hire C's.
And so you can see how that could lead to a much lower quality team over time.
Sure. And how will you keep the bar high as you continue to scale?
I mean, it's relatively easy when it's just a team of 10 or 15 to make sure
that everyone's an A player.
(15:35):
A player is a bit of a cliched phrase, but I guess we sort of understand what
we're each talking about here.
But once you're scaling to 30, 40, 50 employees, which I guess is the direction
you're heading in, then it's much harder to maintain the quality control,
shall we say, over the talent you're bringing in.
(15:56):
So how will you go about that how have you dealt with that in the past.
Yeah, I challenge the premise.
I think that you're always faced
with a certain number of constraints or incentives when growing a team.
One of the obvious incentives is that you want to grow it fast and you want
(16:19):
to grow it well. And those seem to be an apparent contradiction.
But given the amount of leverage that many people can have in a software development
company and a company that's building software for hundreds,
thousands, potentially millions of customers,
the amount of impact one single individual or a small team of people can have
(16:40):
on a very large number of customers is something to keep in mind.
And that's really where I feel that the desire to make or to sort of give up
on certain hiring constraints in the name of Pace can be easy to put back to bed.
If you remember that these people will be writing code that is going to impact,
(17:03):
you know, years of the company's life and thousands or potentially hundreds
of thousands of customers, it gives you a bit of perspective on how you should focus on that.
I'd say that in today's working culture in a 50-person team,
in a 100-person team, Stripe was a little over 100 people when I joined it.
It's quite easy to build a culture of mutual respect and internal transparency
(17:30):
that essentially makes it hard for people to hide behind obfuscated processes
or complex systems of delegation and,
you know, mountains of meetings where nobody's actually doing anything and people
are talking about making decisions to do things.
So one of the cultural aspects
(17:50):
of Stripe that has since been imitated by many companies was the degree to which
internal transparency allowed people to be in the know of what was going on
because most emails were shared internally in a transparent way or meeting notes were
expected very promptly after a meeting ended,
(18:10):
and that created an incentive for people to challenge if they needed to be in
the meeting in the first place, because they knew they were going to get high
quality notes 10 or 15 minutes after the meeting took place.
All these small operational tactics,
I think, fed very positively into the overall strategy of making it easy to
focus on what you do best and rewarded to do what you do best most of the time.
(18:36):
And I think that other companies, I was a product leader at Allen for a few
years in my last role before starting Dust,
the degree of transparency on total compensation that the team still has today
is essentially an invitation and a standard by which to live,
(18:58):
which I think helps people avoid micro-optimizations that are actually detrimental
to the whole system because the penalty of that micro-decision is more urgently or more keenly felt.
When a manager accepts to let a person
team member that's not really pulling their weight stay on the team,
but compensations are not public or performance reviews are not shared,
(19:21):
there's obviously less of a penalty to them as a manager to be doing that.
Their team isn't performing quite as well, but maybe they have 10 or 12 people
on the team that's going to get diluted. It's probably going to be easier for them to.
But if each member of the team gets to know how people are performing,
how people are rewarded, it.
I think it could also trigger some interesting questions in the manager's thinking process.
(19:44):
Am I going to be judged for the decisions that I've made here?
I think some amazing Silicon Valley companies or American companies like Netflix
that early on had a culture and a philosophy of making sure that managers always
felt comfortable with what they were paying their team because they should never
cave in to a negotiation pressure.
If somebody came in and said, I've got an offer with a high compensation somewhere
(20:06):
else, could you match it?
The manager I just should always be in a position saying, you know what?
No, if you've got a higher offer somewhere else, you should go there because
we're really trying to compensate you with the marginal value that we think
you're worth to the company.
I think the inevitable advent of some form of either mediocrity or sub stellar standard is likely.
(20:27):
It happens. I think that in large tech scale ups,
the frantic hiring pace of
2020 and 2021 are often quoted as examples where people hired too fast i think
there was a combination of pace and the cultural shock of hiring people to work
remotely those two probably played off of each other and a lot of mistakes were
(20:48):
made and i think some companies are still finishing to pay for for.
Those mistakes and setting the the company on another track but i you know we're
sub 15 people at dust right now i really don't think that the the bar is something
we could find excuses for lowering at 50.
And I certainly hope we can keep it to 500.
I'd say that there's a tremendous amount of pride to take from having each candidate
(21:12):
that comes in through the door, realizing that this is the entire team asking
us if we use vendors or subcontractors, hearing that we don't,
hearing that we do things internally.
I think it also generates a certain amount of interest from certain people who
want to join that kind of team because great things can be done by fairly small teams, I think.
(21:34):
And I don't believe that the number of humans that are participating in the
project is a very relevant metric for many things.
Yeah, I'm on the same page there. That's for sure. Tell me about today's startup ecosystem in Paris.
And what are the advantages that that offers for companies like Dust compared
(21:55):
to other tech hubs, either in Europe or further afield?
Yeah, so obviously a couple of contrasting points.
I think that it's likely Silicon Valley is still the acme point of the startup
culture and development potential.
And some of the reasons I would invoke for that are that there's layers and
(22:19):
layers and generations and generations of repeated game microeconomics at play
in Silicon Valley, where people have had to be very collaborative,
have had to maintain a reputation across jobs, across companies,
in the high highs and the low lows, through busts and bubbles.
And it has done two things. One, I think it's created an ecosystem of people
(22:42):
that have many more referenceable co-workers, which is always nice.
And two, it's really created a cohort of people who have first-hand operational
experience in scale-ups, in high-paced growth environments, which I think Europe
is just a bit of a disadvantage on.
That said, I think that France today versus France 10 years ago is night and day.
(23:07):
I'd mention three points, maybe. One, the degree to which it is recognized as
a promising industry sector
by the press, the government, society at large.
I think people can no longer ignore the fact that some of the largest companies
(23:28):
in the world have been technology companies, that they've provided many jobs
to people in different markets.
And so that awareness has helped really reshape how people thought about their
options graduating from a business school or an engineering school versus 10 years ago.
And when I graduated, you were going into finance, you were going into consulting,
(23:49):
you were maybe gonna go into the
automotive industry or other leading industries, oil and gas, et cetera.
And I think that tech has made it into that mix for selective students or selective
people beginning their career.
The second point is that we have the first elements of the members of this cohort
(24:11):
of people who've done it before, who've experienced what it is to work at a
scale up, who are now starting companies.
And I'd say that the level of ambition of these founders is in many ways different
and higher than what you could more commonly observe 10 or 12 years ago.
It was more common 10 or 12 years ago to see people that were trying to execute
a me too version of an idea that they'd seen in the US with a hope that they'd
(24:35):
be acquired maybe by the Samuel brothers from Germany or a company coming in
and looking to do their European expansion in some faster way.
But many companies started now are started by founders who have no intention
of wavering or selling early and who are very happy to attract talent and know
that the level of the ambition they put in their idea is going to positively
(24:58):
impact the quality of the team and product and company they're able to build.
So I'd say that that mindset change is really supported by people who've worked
at Airbnb for five or seven years and then come back to France.
People who've worked at Uber that was one of the largest offices in Europe for
a while and are now starting their own companies.
I'd say that the third point is in the specific context of generative artificial
(25:22):
intelligence or more broadly artificial intelligence.
Whereas I think the web and mobile revolutions really favored London,
I think a very healthy ecosystem of agencies, very close to branding,
very close to making snappy, snazzy apps and projects and websites,
either because they were working for a WPP company or because they were working
(25:45):
for some of the top brands that all had their offices in London.
London, some of the companies that are emerging this time around focused on
more mathematically driven technology and the training that students have in
the engineering or engineering schools in France is one of those types of university
(26:05):
that really favors this.
Very STEM focused, quite rigorous, very comfortable with a theoretical approach
to problem solving and a systemic approach to problem solving,
which does provide a bit of an edge.
I'd say some really successful companies from 20 years ago that maybe not everybody
remembers, but Exalead was a very successful company in search 20 years ago.
Early employees of Exalead have become a very successful mafia.
(26:28):
Some have gone on to start companies like Algolia, which is a very popular search solution.
Or other relatively technical companies where scaling is an intrinsic part of
how the company can be successful.
I think that the problems that the field of generative artificial intelligence
faces today are really scaling problems,
(26:49):
and in some cases, very theoretical problems that are on the frontier of research and development.
Obviously, I'd be remiss not to mention Mistral, which is an amazing team based
here in France that's managed to attract global talent,
that's managed to create a brand and a spot for itself in a very competitive
industry, and that is doing that
(27:12):
by leveraging, I'd say, all three of the previous aspects I mentioned.
A government that looks with a more enthusiastic eye on what's being done in this sector,
researchers and engineers that have had a successful beginning to their careers,
whether at Meta or at DeepMind, and that are eager to carry those lessons and
those experiences forward in their own experience.
(27:33):
And thirdly, training and for their early employees, just a general appetite
for hard theoretical problems. That's very encouraging.
You got your one, two, three punch in there. Okay.
When we last spoke, you shared an intriguing perspective on AI.
You said to me, I think this technology
(27:54):
is going to force us to turn our brains on, not off as many fear.
So to expand on this perspective, because I think there are many people out
there who hold the opposite viewpoint.
Yes. And you know, the jury's out. Let's see how things go.
But the observation is as follows. The current paradigm for generative artificial
(28:18):
artificial intelligence, not to go into too many details, but these models,
these large language models are stochastic in nature.
They are non-deterministic. So you could ask the same question to an LLM and
you could get a different answer depending on some other settings that you might have put in.
And that's disturbing. That's in some ways not reassuring at all.
(28:40):
We've all heard about the issue of hallucinations, you know,
is the model actually responding truthfully?
And and and that's a valid but slightly
different pocket but i think if we just focus on the fact that these
are non-deterministic models what's happening is that we're getting tools that
allow us to do many things much faster at the expense of having some variance
(29:02):
in the actual result and so you know you have the option to do something very
precisely and meticulously but in a very very
slow manner, or something that may be right x percent of the time,
you know, x may be 90, maybe 95, it may be 98, it may be 99,
but that will get something wrong y percent of the time.
And they can do it almost immediately or in a matter of seconds or sometimes
(29:25):
minutes for the most complex cases.
And so I think the question is, rather than say that one is always better than the other,
it's to realize that we're going to have to reason in terms of distributions
and that many of the tasks that we've been doing in our jobs,
in knowledge worker jobs,
have been reasoning, information gathering, information synthesis,
(29:48):
information refactoring.
Some of these have a pretty high tolerance
to minor mistakes it doesn't really matter if the
way in which you write your meeting notes has you know
a slightly verbose conclusion it doesn't really matter
if you can spot an obvious
repetition of something in a paragraph that you just delete before sending and
(30:09):
so we have to really try and compare the roi we're getting from the much shorter
time invested and the actual value of the job and so what i was saying when
I was talking about turning our brains on is that I think we'll be collaborating
with this technology rather than delegating in many cases,
and that the augmentation that we will get from having these tools at our disposal
will also require that we have our brains turned on and that we're quite focused
(30:33):
on bringing out our critical sense to the decision as to whether or not the
decision is the right one,
the context was brought in correctly, a certain number of checkpoints have been validated.
This notion of human in the loop is, I think, pretty key.
For the tasks where there really is no value in even considering the fact that
sometimes the answer is correct or right and it just gets it 100% of the time
(30:57):
or close enough to 100% that you could say it's not distinguishable,
I think what we have to ask ourselves is.
What were these tasks and what was the perceived value add that humans were
bringing to it before machines could do it?
Because we used to be able to take ourselves from A to B in cities by horse
(31:19):
and carriage, and we found it great.
And I think the skills of being able to tend to a horse while riding a carriage
are a little less valued today than they were back then, because you really
couldn't get very far if you didn't know how to manage a horse and carriage then.
And I think we have technology that just removes that problem and looks at it differently.
So I'm trying to not be naive and I'm trying to not be too gleefully optimistic
(31:42):
in the fact that there will be disruption.
And I think there will be situations where we'll have to ask hard questions
about the ethics of using automated systems and decision making.
And there's many cases where automating it has been either bad or a downright catastrophe.
I always think about Virginia Eubanks' book, Automating Inequality,
where I think in data analysis in the United States, certain minority populations
(32:06):
were at a systemic disadvantage because of the way credit reporting was poorly done.
And in some cases, when credit reporting was not available, people would put
in a zero instead of putting not available.
And that would damage averages as they were computed and have,
obviously, very bad effects.
So there are situations where automated systems will be in critical paths,
(32:28):
and it'll be important to really isolate those and understand that.
But by and large, we are just getting a new tool that is going to make bringing
up to the surface what is left for humans to decide, which can be in some cases
due to aesthetic issues.
Faith-based, principles-based, ethically-based decisions on a number of steps
(32:49):
that have been taken for us, which I don't know people were waking up in a particularly
excited way to go and do in the morning.
I don't think great people or the best team members wake up in the morning and
just like in such a hurry to go and type out meeting notes or just to collect
the file from the CRM database and understand which people we've emailed in the last three months.
I just don't, I don't believe that that's what people were excited to do.
(33:12):
You mentioned earlier on that this is not your first startup,
but I guess this is even so an interesting learning experience for you.
We're always learning, always modifying, always overcoming fresh challenges.
So what are some of the biggest challenges you faced this time around?
And what are some of your major learnings?
(33:35):
Well, I mean, I'm certainly not done. done i think new challenges every day
i think it goes back to you know the the framing and the mindset that you approach
problem with in the first company and by the way my first company was started
with the same business partner stan and i have known each other since undergrad
and we started our previous company together and the starting dust together again so.
(33:55):
I i'd say that by setting the bar
of your ambition higher you're going to see as
potential problems problems some things that you might not have even
considered as problems if if that wasn't if
that wasn't the mindset you were entering the problem with so
you're thinking about how to raise capital who to raise capital from how to
(34:16):
optimize the terms at which you raise that money 12 years ago we were just happy
to have investors return our calls and offer us a term sheet and I we weren't
particularly picky on the terms that came with that term sheet because we didn't
really have leverage to do any better.
But seeing firsthand from my experiences at Allen or at Stripe,
how very thoughtful founders were trying to play a few steps out what the consequences
(34:42):
of those decisions might be on themselves,
their independence of decision making with the board, the ability to attract
and retain top talent with aggressive compensation packages.
All these things come back to bite you if you haven't anticipated them.
And so sometimes I think one of the challenges is trying to also recognize that,
(35:04):
you know, it's still very hard to make perfect decisions.
And the time in which you make those decisions is still one of your assets.
Compared to larger companies, compared to incumbents, the one asset a startup
has is the velocity with which it can move.
And so you don't want to give that asset up by over-optimizing or over-calculating
on too many decisions. That's one challenge.
(35:26):
But other than that, I'd say, you know, it's broadly a lot easier the second time around.
There are certain things that you just don't spend time on or don't worry about
as much when you try and understand that it's all very simple.
You're trying to make something that people really want.
You're trying to get it in front of the kind of people that are likely to want it.
(35:46):
And you have to make sure that when they want it, they're willing to pay something
that is higher than what it costs you to make it.
That's basically what it is. And as soon as you depart apart from those considerations,
you're probably deluding yourself and wasting time on things that matter a lot less.
It's really focusing on those key things day in, day out and not getting too
distracted that is the major lesson, I think.
(36:07):
I love your ability to break complex things down into typically three components.
So now I'm going to put you on the spot. Who are the three people from the world
of tech and the world of the world of tech and the world of entrepreneurship
three people who've most inspired you to be who you are and why.
(36:32):
Good question i think i think one of them is and i'll start with him although
uh saturn adela i think is an inspirational character.
Satya Nadella has added more market cap to Microsoft than Bill Gates ever did.
He's made a few very ambitious decisions or has let very ambitious decisions
(36:57):
be made under his stewardship of Microsoft and has turned a very large ship around,
quote unquote, or at least has made it into one of the most exciting technology
companies to talk about in 2024, which I think if we were having this conversation
in 2010 10 would not seem like a like a likely scenario and and i think that speaks to a tremendous,
(37:20):
yeah an inspirational ability to not be overwhelmed by something that seems
impossible to do and just getting on and doing it i think.
Yeah, actually, I'd say that my co-founder is a source of inspiration.
My co-founder started coding at a very young age and trying to find business
(37:42):
applications for technological hacks and ideas that he would have.
And he started his first company in high school, which was a completely stupid idea.
It was to sell storage online and allow people to have a hard drive in the clouds.
And who on earth would ever want that? and and and
i think understanding that there are many different ways
(38:04):
to see the same reality and that that's okay and then
being able to talk about the differences in those viewpoints and
really break down with no to low to
no ego why you believe the things
you do is one of the most constructive working relationships you can have and
i think that and that you know 20 plus years on the fact that we're working
(38:25):
together again is is is in part because of that i think because of the continuous
desire to continue a relationship that I think is very positive in that way.
And then you asked for three, so I'm going to have to come up with a third one.
And I'd say, I'll cheat a bit, but I'd say that what I'm increasingly seeing
(38:49):
is that people are complex, people are nuanced to every complicated problem.
There's a solution that's simple, elegant and false.
And so rather than take people wholesale and accept the things that they might
do very differently from what you prefer, I'd like to spot what people were
(39:10):
great at and see that as a source of inspiration.
So, you know, I think Steve Jobs was a horrendous manager close up,
but I do think that he was brilliant in pushing for product quality and bringing
in inspiration from fields that,
that had previously not really communicated to each other to maintain a bar
of simplicity and beauty in some of the products that he built. I think that.
(39:36):
I think, you know, John and Patrick Collison,
the founders of Stripe, were always quite impressive in their ability to elevate
and bring back to first principles a certain number of situations that they were faced with.
And, you know, they were obviously very well surrounded and had a lot of people
(39:58):
to go and get that inspiration from, but to try and, you know,
challenge pretty much everything that needed to be challenged rather than just
go ahead with what had been done previously.
And so, you know, the reverse auction for IPO price of Google is another example of that.
I think, you know, it's always cited as one of those moments where the legal
and banking industry is like, what are you actually going to say?
(40:19):
It's like, well, why not?
You know, if people really want to bid for this, they should give their best number.
And more recently, I think I've found myself,
really appreciating some of the ways in which Xavier Niel here in France is
happy to put his reputation and focus in the balance on big ambitious projects.
(40:44):
I think 42 was, you know, people remember him for free. People know him for
some of his more recent projects, obviously, Scaleway and Qtai in artificial intelligence.
But I think 42 was a brilliant move in actually going back to the roots of why
the French educational system is relatively well known for being very elitist
(41:06):
and having a hard selection at many steps of the process, especially after high school. all.
And that comes with pros and cons. And I think what you get when you see people
from that particular track is people who are able to resist a certain amount of stress and pressure.
It doesn't mean that they're great at the things you're hiring them for,
but that's what they were able to do when they were between the ages of 18 and 20.
(41:27):
And I think using that principle to bring people into an environment where you
could actually bring back that high bar of competitive competitive mindset in
a much shorter timeframe,
focused on software development, build a brand around that, that people in less
than a decade now reference as a solid brand for software engineering is amazing.
(41:49):
I think it's hard to, I've always got a lot of respect for ideas that are hard
to imagine the world without once they've been put in the world,
but that nobody had thought of in that obvious a way before,
at least in that ecosystem.
So yeah, there's realizing that I've only mentioned men and so of course my
wife's voice is ringing in the back of my head right now,
(42:13):
but yeah I think we're actually at Dust excited to bring back some of the creative aspects of.
Research and not just development and allowing people
to not just follow a gradient descent in optimizing everything but
allowing them to have the space and time to find
(42:34):
new solutions to problems that we we
were just confronted with and i think marie curie
in france of course i'll go and wave the french
flag again here a bit it's one of those you know one of
those pioneers that proved everybody around her
with her husband of course but okay things that have been previously mentioned
as being impossible which is a question of grit
(42:55):
and perseverance and and
just following on that point you you made reference to females and made reference
to your your wife there how challenging is it for you to build a truly diverse
team at at dust is that something that you've paid much attention to?
(43:16):
Yes, so if I'm not mistaken, it's the 1999 Stanford paper by O'Reilly,
and I'm forgetting a number of the other authors on diversity and team performance.
And I think it's not always quoted the way I think it was intended.
I get a lot of inspiration from the way in which Commando teams were designed
and invented around World War II.
(43:37):
And as a startup, you're dead until proven otherwise.
You're really a nonprofit that's got a few months wants to make a point.
And so what you need is to increase your chances of survival.
And I think NASA or the SAS have a lot in that toolbox.
And what you need is diversity of skills. You need diversity of skills and mindsets
to approach problems in a way that is novel and often superior.
(44:01):
And so that's something that we really try and do.
Pragmatic and modest way in which we do it is that we don't pay too much attention to people's.
Cvs when they come in for interviews and it's not
even recorded in the hiring process right now we
try and have chats a few chats and a screening
on on skills and what we call technical screening before really
spending too much time thinking about what people did
(44:24):
when they were just after high school when it comes
to diversity gender i i think
it just you know you just have to have a
hiring bar and try and increase the top of
the funnel to increase the chances of you meeting a
certain number of diversity goals in in scaling engineering
teams at stripe one of the engineering leaders there as a woman i have a lot
(44:46):
of respect for pretty just pragmatically said you know if if we want to have
more women in seats we just need to see a lot more women at the interview stage
not to adapt our interview bar but force ourselves at the very first steps of recruiting.
So the second, if I'm not mistaken, the third engineer to join Dust is a woman.
(45:08):
She was a fantastic software developer that I'd met in a previous part of my
career and that I was excited to have join us.
The first person to lead the business operations team, also a woman who had
had a very successful career scaling the French branch of an international company.
And I think that makes it easier. when there's seven people in a room and two
(45:30):
of them are women, then it just makes it easier for anybody walking in the door
to find the environment maybe a little less tough to be the first in.
It's a longer topic, and I don't want to skim over the other forms of diversity
that need to be contemplated.
I think in software development, we're fortunate in that it's by and large a new industry.
I think if you're very determined, hardworking, and curious,
(45:54):
there's not a lot that you can't learn and practice online.
And so some of the more traditional ways in which,
elitist groups have favoured themselves and their own are a little challenged by that.
I think we've met some incredible candidates who came to software development
or the tech world more broadly by very surprising ways.
(46:16):
And one of them was always curious to, you know, their ability to just be in
wonder and or at something new that they just wanted to get their hands dirty with.
So I think as we grow, we'll also have to continue to understand what for the
brand, the employer brand that we build, all the role that we want to play in
the ecosystem that we're in, we want to do. I think there's a time and place for everything.
(46:38):
I don't believe that a 12- or 13-person company can do everything,
but I think we can always challenge ourselves to do more.
So we incorporated the company in France. That wasn't something that a lot of
people told us to do. We decided to build a team here.
That wasn't necessarily something that a lot of people encouraged us to do.
I think we're trying to challenge ourselves to be positive contributors in the
(47:01):
ecosystems that we want to live in.
And when I look at some of the ways in which this industry has virtualized and.
Digitized almost everything, I am left with the question of,
you know, how does it positively contribute to the streets that you live on
and the parks that you walk by and the places that you actually want your kids to go to school in?
And so building teams more locally
(47:24):
in places that you want to live in seems like
a pretty sure way to redistribute uh obviously
hopefully positive outcomes where uh where you're going
to to spend your time that's uh one of the ways in which
i think it's a lovely lovely expressive way of getting your points across i
wish you and your diverse team a dust huge never diverse enough but you know
(47:50):
diverse in skills for sure and and and and hopefully more diverse in many more ways over time.
Well, in your increasingly diverse team, wish you huge success cracking team productivity with AI.
And thank you so much for joining me on today's show.
Thank you very much for having me, Gary. It was a pleasure to chat.
Special guest today is Gabriel Hubert, founding CEO at Dust,
a Paris-based SaaS scale-up funded by major global VCs like Sequoia,
and they leverage generative AI to transform how teams collaborate and thrive.
So, Gabriel, welcome to the Startup to Scale-Up game plan. Thank you very much for having me on, Gary.
(00:22):
You're most welcome. Now, your tagline, or rather Dust's tagline,
is cracking team productivity with AI. It's a pretty bold statement.
So could you elaborate on how Dust leverages AI to enhance team productivity
and maybe a little about the way you set your approach apart from traditional productivity tools?
(00:48):
Yeah, I think the main difference that arises from this wave of technology that
everybody's been talking about, large language models,
and more specifically, text transformer models, is their ability to really act
(01:09):
as a glue between forms of unstructured text.
Context and so a lot of productivity in
companies is really given the right people
the right information for them to make decisions as they go on about
their day and i think if you've hired the right people then the best thing you
can do is get out of their way and let them do their work but they do need the
right piece of information to do their work and sometimes that information is
(01:32):
in a siloed piece of software that another function is using that they don't have access to and that
might prevent them from being able to make the decision without calling a meeting.
So I think one of the keys to really changing the way we approach autonomy, empowerment.
And just a more purposeful day-to-day life at the office is to break down those
(01:57):
knowledge silos and have people be able to access and transform at will the
information they might need to make decisions.
And that's really what Dust tries to do. So using the brick of generative artificial
intelligence, we try and make it possible for,
anybody in the company to have access to the right pieces of information they
(02:20):
need to go on about their day.
And how do you, or how does the DUST solution adapt to the diverse needs of
different teams and different industries?
Yeah, that's a great question. I think it's actually one of the reasons that
we've decided to build a product that's fairly horizontal in nature and has
some of the key building blocks that we think are necessary to support a wide range of use cases.
(02:47):
The way the Dust platform is built, it allows for unstructured data flows from a variety of
software solutions to be brought into context and then consumed by purpose-built assistants.
These assistants are customized around specific use cases.
(03:09):
And we've really moved away from the paradigm of having a single assistant try
and do everything for everybody within companies.
We're pretty much on the opposite end of the spectrum. We think that people are quite smart.
You know, when they have access to a toolbox, they can tell the difference between
a hammer and a screwdriver.
And so what they need is a high-performance tool to complete the task or the tasks that are at hand.
(03:35):
And customizing AI assistance to that specific set of use cases is going to
increase their performance and allow them to benefit from just a higher throughput set of tools.
So one of the really core ways we've thought about Dust is, what do fast-moving teams need?
(03:55):
They need fairly horizontal access to the data.
They need relatively easy-to-understand and apprehend building blocks to build
and customize these assistants.
And then they need that product to be able to live as close as possible to where people do their work.
That's why we're available as a web app, but also as a Slack bot in Slack,
(04:17):
or as an API where people can actually build for Dust to be really brought to
the piece of software or the specific tool that they're using day to day.
And who are some of the early adopters?
Yeah, we seem to have struck a chord with companies that fit three characteristics.
(04:37):
One, their size and pace of growth means that internal information chaos is
really a need to solve problem. It's not a vitamin problem.
It's a painkiller problem.
And that generally means that we're talking to companies at least 250,
often 500, sometimes a few thousand employees.
(04:58):
The second piece is that we do have connectors to certain families of software solutions.
And so that does restrict the customers that we're currently able to talk to.
We're a fit for teams that use Slack, GitHub, Notion, Google, Intercom, et cetera.
The list goes on, but it does paint a picture of maybe more recent and in many
(05:22):
cases, tech forward companies.
The third point, And I actually should have started by that because I do think
it's the most important one, is that we really are a fit for companies where leadership,
whether it's the founders or the C-suite, maybe the board has been a source
of inspiration in getting people moving in the right direction,
optimistically and fundamentally convinced that this technology is going to
(05:45):
change things very quickly.
And that they need to adapt, they need to react very quickly,
and they need to deploy generative artificial intelligence solutions within
their teams as quickly as possible.
If those three conditions are met, we seem to strike quite a bit of interest.
In the early group of customers that we're working with, I can mention Allen
(06:05):
Healthcare Insurance, focused on the European market.
Conto, a neobank, also focused on Europe.
At Penny Lane, Payfit, which is a payroll provider, and Penny Lane's accounting software.
And Watershed, for example, based in the US that focuses on carbon accounting
and helping large companies better prepare for the amount of regulation and
(06:26):
climate change risk mitigation that lies ahead of us.
These all have in common that they're generally less than 10 years old,
and the founders really see this as a disruption, not necessarily to their business,
but to the way people are going to be working.
They want to focus on understanding how their most empowered team members should
(06:47):
think about changing the way they work, which will have obviously an impact
on these co-workers, but also maybe on the way they need to plan hiring in the future.
And how do you actually measure the impact of dust on team performance and efficiency?
What metrics are you using?
Yeah, I'd start, first of all, by mentioning that this is a broad and sometimes complex topic.
(07:10):
Oversimplification on a few metrics can maybe miss some points.
I've managed support teams or engineering teams building tooling for support teams in my career.
And the number of tickets processed per minute
is a very tempting metric to try and
check that we're increasing efficiency but it can hide sometimes
a lot of complexity in the fact that you know if that's
(07:31):
really the only the only incentive teams have they might
they might feel the need to reply to a
ticket that's already been solved with a one last email saying thank you very
much and have a lovely day because that can get them another another score on
the tickets per hour so quantifying everything is a little tough but we do work
on quantifying the ability for people to get access to information their need
(07:52):
and generally do their work better so in our pilot phases with customers.
We'll have a quantitative survey with the teams to ask them how much time they
feel they've saved on average per week.
And we think that it's a relatively descriptive metric of what we're trying to do.
At the end of the day, if you're focusing on productivity, you're trying to
(08:13):
get people to do the same amount of stuff in less time or more stuff in the same amount of time.
And we take the results with a pinch of salt. It's also important to try and
understand the qualitative feedback.
In some cases, the gains may not yet be substantial across all teams,
but the way in which they're able to delegate the more mundane tasks to an assistant
(08:36):
or an automated system gives people the impression that they're taking back
control of their day and able to focus on higher value-add tasks.
And that's also something that we try and look at.
Our ambition, if you look at most commonly shared surveys among software developers
that have used Copilot for almost more than a year now,
(08:56):
is that we can get within 25%, 30%, sometimes 35% gains in productivity.
And we really think the number of jobs that are going to see those gains is
not limited to software development. Of course, coding is an obvious candidate
because the human-machine interface is what these people spend their day working in.
(09:18):
And so making the human-machine interface that much snappier directly correlates
to productivity gains for them.
But we really think that the amount of time people spend looking for information,
formatting information that's been collected.
Translating between internal languages pieces of information that they might
not initially understand,
The marketing team that's trying to understand what the engineering team has
(09:40):
shipped or the engineering team that's trying to pass through a sales call and
see if there are actually some new and unsuspected feature requests from a key customer.
These are all use cases where the unstructured flows of information can be structured,
can be structured on the fly upon request with a special kind of filtering.
(10:03):
And that's what we really think is powerful. So again, North Star is to be shaving
significant amounts of time off people's day-to-day tasks and hopefully moving
them to a higher value-add mix of tasks day in, day out.
Now, much of your career was spent in product management in Silicon Valley.
You also did your master's at Stanford.
(10:24):
So you followed a somewhat similar path to Steve Jobs.
I'm not sure if you also dropped out at Stanford. Obviously, Steve famously did.
How have you leveraged your product management experience and your Silicon Valley
experience to become a startup founder in Europe?
Yeah. I am. Well, so historically, I was actually a startup founder in Europe
(10:46):
before my product management experience in the US.
Graduating from Stanford was a pretty unique experience.
I was in applied mathematics there. And it really gave me a taste of just how
close the academic and private sector ecosystems were in that specific part of the world,
and how open and dynamic the conversations were.
(11:08):
Being able to go to talks with startup and scale-up founders that have been
very successful, meeting investors and demystifying the job of a venture capitalist early on.
Those were amazing opportunities that really helped me understand that the people
starting some of the most successful companies that we now know of or use the
products of every day are generally very ambitious individuals,
(11:30):
but also people who have access to an ecosystem and in many cases,
reflexes and principles that help them think differently about the way to to solve problems.
In my second experience in Silicon Valley, so after the acquisition of a company
that we'd started in Europe in 2012, the company was acquired in 2014 by Stripe,
(11:51):
a payments processor based in San Francisco, and we'd just moved to San Francisco.
So we spent a few years working at Stripe after that acquisition.
I think what was quite clear and very obvious was that the scale of your ambition
does forge the way in which you make certain decisions.
The same situation can be resolved quite differently depending on the attitude
(12:16):
and the mindset that you bring to the problem and the ambition that you bring
to solving it. Who are you solving it for?
How often are you going to be solving it? How automated do you want the solution to be?
All these are sub questions that really can be drawn from some original questions
like, all right, what kind of company do you want to build?
I think one of the obvious separations is that in many situations,
(12:38):
you could have a consulting style answer to people's problems.
I'll build you this one-off piece of software that fits your needs,
and I'll be able to bill you by the hour, bill you by the day,
and I'll have a very happy customer if I keep working on that.
But the ability to design a product that is going to solve the problems of this
one customer, but maybe also the problems of many other customers to be is really
(13:01):
where I'd say product thinking comes into play.
So to give some.
Often quoted examples of the mindset that I think was very much put in the spotlight
in Silicon Valley, an obsession about the customer, an obsession about the user.
What are the users actually doing? What problem are they trying to solve?
(13:22):
And how does your solution fit in their way of seeing how to solve the problem?
Clay Christensen, one of the deep thinkers on product discovery,
Discovery always mentioned the example of the smoothie at Starbucks,
which was an elegant solution to the problem of being able to have a breakfast
on the go while not getting out of your car and not getting your clothes dirty.
(13:46):
So the problem is, you know, you want to get your protein on the way to work.
The solution is a smoothie that fits in the cup holder in your car and get to that mix.
And I think that's a really vivid illustration of, you know,
how close you have to be to your users to try and not mistake what they're describing
as their problem for what their actual problem is.
(14:06):
Because otherwise, the risk is obviously that, you know, Ford famously,
I think, was quoted saying, if I built what people were asking,
I would have built faster horses.
But of course, you need to try and extrapolate from the description that people
have of the job they're trying to get done and move to a solution that elegantly covers that,
but also maybe takes out some of the intermediary steps for them.
(14:30):
So yeah, that's one example.
I think another one that's pretty key is the importance of hiring a very high
quality team And the fact that maintaining the bar at recruiting and during
the development of people's career is one of the best favors you can do to your team as it grows.
(14:51):
Great people have many options on the table. They could be doing anything.
And so if they choose to come and work with you, they're going to do it because
they like the product that you're building.
They like the mission that you're trying to achieve, but also they like the
people that they're working with. And the people that they're working with is
really one of these assets that can deteriorate over time if you don't pay attention
to continuously maintaining that high hiring bar.
(15:15):
You know, the famous quote about A's hire A's and B's tend to hire C's.
And so you can see how that could lead to a much lower quality team over time.
Sure. And how will you keep the bar high as you continue to scale?
I mean, it's relatively easy when it's just a team of 10 or 15 to make sure
that everyone's an A player.
(15:35):
A player is a bit of a cliched phrase, but I guess we sort of understand what
we're each talking about here.
But once you're scaling to 30, 40, 50 employees, which I guess is the direction
you're heading in, then it's much harder to maintain the quality control,
shall we say, over the talent you're bringing in.
(15:56):
So how will you go about that how have you dealt with that in the past.
Yeah, I challenge the premise.
I think that you're always faced
with a certain number of constraints or incentives when growing a team.
One of the obvious incentives is that you want to grow it fast and you want
(16:19):
to grow it well. And those seem to be an apparent contradiction.
But given the amount of leverage that many people can have in a software development
company and a company that's building software for hundreds,
thousands, potentially millions of customers,
the amount of impact one single individual or a small team of people can have
(16:40):
on a very large number of customers is something to keep in mind.
And that's really where I feel that the desire to make or to sort of give up
on certain hiring constraints in the name of Pace can be easy to put back to bed.
If you remember that these people will be writing code that is going to impact,
(17:03):
you know, years of the company's life and thousands or potentially hundreds
of thousands of customers, it gives you a bit of perspective on how you should focus on that.
I'd say that in today's working culture in a 50-person team,
in a 100-person team, Stripe was a little over 100 people when I joined it.
It's quite easy to build a culture of mutual respect and internal transparency
(17:30):
that essentially makes it hard for people to hide behind obfuscated processes
or complex systems of delegation and,
you know, mountains of meetings where nobody's actually doing anything and people
are talking about making decisions to do things.
So one of the cultural aspects
(17:50):
of Stripe that has since been imitated by many companies was the degree to which
internal transparency allowed people to be in the know of what was going on
because most emails were shared internally in a transparent way or meeting notes were
expected very promptly after a meeting ended,
(18:10):
and that created an incentive for people to challenge if they needed to be in
the meeting in the first place, because they knew they were going to get high
quality notes 10 or 15 minutes after the meeting took place.
All these small operational tactics,
I think, fed very positively into the overall strategy of making it easy to
focus on what you do best and rewarded to do what you do best most of the time.
(18:36):
And I think that other companies, I was a product leader at Allen for a few
years in my last role before starting Dust,
the degree of transparency on total compensation that the team still has today
is essentially an invitation and a standard by which to live,
(18:58):
which I think helps people avoid micro-optimizations that are actually detrimental
to the whole system because the penalty of that micro-decision is more urgently or more keenly felt.
When a manager accepts to let a person
team member that's not really pulling their weight stay on the team,
but compensations are not public or performance reviews are not shared,
(19:21):
there's obviously less of a penalty to them as a manager to be doing that.
Their team isn't performing quite as well, but maybe they have 10 or 12 people
on the team that's going to get diluted. It's probably going to be easier for them to.
But if each member of the team gets to know how people are performing,
how people are rewarded, it.
I think it could also trigger some interesting questions in the manager's thinking process.
(19:44):
Am I going to be judged for the decisions that I've made here?
I think some amazing Silicon Valley companies or American companies like Netflix
that early on had a culture and a philosophy of making sure that managers always
felt comfortable with what they were paying their team because they should never
cave in to a negotiation pressure.
If somebody came in and said, I've got an offer with a high compensation somewhere
(20:06):
else, could you match it?
The manager I just should always be in a position saying, you know what?
No, if you've got a higher offer somewhere else, you should go there because
we're really trying to compensate you with the marginal value that we think
you're worth to the company.
I think the inevitable advent of some form of either mediocrity or sub stellar standard is likely.
(20:27):
It happens. I think that in large tech scale ups,
the frantic hiring pace of
2020 and 2021 are often quoted as examples where people hired too fast i think
there was a combination of pace and the cultural shock of hiring people to work
remotely those two probably played off of each other and a lot of mistakes were
(20:48):
made and i think some companies are still finishing to pay for for.
Those mistakes and setting the the company on another track but i you know we're
sub 15 people at dust right now i really don't think that the the bar is something
we could find excuses for lowering at 50.
And I certainly hope we can keep it to 500.
I'd say that there's a tremendous amount of pride to take from having each candidate
(21:12):
that comes in through the door, realizing that this is the entire team asking
us if we use vendors or subcontractors, hearing that we don't,
hearing that we do things internally.
I think it also generates a certain amount of interest from certain people who
want to join that kind of team because great things can be done by fairly small teams, I think.
(21:34):
And I don't believe that the number of humans that are participating in the
project is a very relevant metric for many things.
Yeah, I'm on the same page there. That's for sure. Tell me about today's startup ecosystem in Paris.
And what are the advantages that that offers for companies like Dust compared
(21:55):
to other tech hubs, either in Europe or further afield?
Yeah, so obviously a couple of contrasting points.
I think that it's likely Silicon Valley is still the acme point of the startup
culture and development potential.
And some of the reasons I would invoke for that are that there's layers and
(22:19):
layers and generations and generations of repeated game microeconomics at play
in Silicon Valley, where people have had to be very collaborative,
have had to maintain a reputation across jobs, across companies,
in the high highs and the low lows, through busts and bubbles.
And it has done two things. One, I think it's created an ecosystem of people
(22:42):
that have many more referenceable co-workers, which is always nice.
And two, it's really created a cohort of people who have first-hand operational
experience in scale-ups, in high-paced growth environments, which I think Europe
is just a bit of a disadvantage on.
That said, I think that France today versus France 10 years ago is night and day.
(23:07):
I'd mention three points, maybe. One, the degree to which it is recognized as
a promising industry sector
by the press, the government, society at large.
I think people can no longer ignore the fact that some of the largest companies
(23:28):
in the world have been technology companies, that they've provided many jobs
to people in different markets.
And so that awareness has helped really reshape how people thought about their
options graduating from a business school or an engineering school versus 10 years ago.
And when I graduated, you were going into finance, you were going into consulting,
(23:49):
you were maybe gonna go into the
automotive industry or other leading industries, oil and gas, et cetera.
And I think that tech has made it into that mix for selective students or selective
people beginning their career.
The second point is that we have the first elements of the members of this cohort
(24:11):
of people who've done it before, who've experienced what it is to work at a
scale up, who are now starting companies.
And I'd say that the level of ambition of these founders is in many ways different
and higher than what you could more commonly observe 10 or 12 years ago.
It was more common 10 or 12 years ago to see people that were trying to execute
a me too version of an idea that they'd seen in the US with a hope that they'd
(24:35):
be acquired maybe by the Samuel brothers from Germany or a company coming in
and looking to do their European expansion in some faster way.
But many companies started now are started by founders who have no intention
of wavering or selling early and who are very happy to attract talent and know
that the level of the ambition they put in their idea is going to positively
(24:58):
impact the quality of the team and product and company they're able to build.
So I'd say that that mindset change is really supported by people who've worked
at Airbnb for five or seven years and then come back to France.
People who've worked at Uber that was one of the largest offices in Europe for
a while and are now starting their own companies.
I'd say that the third point is in the specific context of generative artificial
(25:22):
intelligence or more broadly artificial intelligence.
Whereas I think the web and mobile revolutions really favored London,
I think a very healthy ecosystem of agencies, very close to branding,
very close to making snappy, snazzy apps and projects and websites,
either because they were working for a WPP company or because they were working
(25:45):
for some of the top brands that all had their offices in London.
London, some of the companies that are emerging this time around focused on
more mathematically driven technology and the training that students have in
the engineering or engineering schools in France is one of those types of university
(26:05):
that really favors this.
Very STEM focused, quite rigorous, very comfortable with a theoretical approach
to problem solving and a systemic approach to problem solving,
which does provide a bit of an edge.
I'd say some really successful companies from 20 years ago that maybe not everybody
remembers, but Exalead was a very successful company in search 20 years ago.
Early employees of Exalead have become a very successful mafia.
(26:28):
Some have gone on to start companies like Algolia, which is a very popular search solution.
Or other relatively technical companies where scaling is an intrinsic part of
how the company can be successful.
I think that the problems that the field of generative artificial intelligence
faces today are really scaling problems,
(26:49):
and in some cases, very theoretical problems that are on the frontier of research and development.
Obviously, I'd be remiss not to mention Mistral, which is an amazing team based
here in France that's managed to attract global talent,
that's managed to create a brand and a spot for itself in a very competitive
industry, and that is doing that
(27:12):
by leveraging, I'd say, all three of the previous aspects I mentioned.
A government that looks with a more enthusiastic eye on what's being done in this sector,
researchers and engineers that have had a successful beginning to their careers,
whether at Meta or at DeepMind, and that are eager to carry those lessons and
those experiences forward in their own experience.
(27:33):
And thirdly, training and for their early employees, just a general appetite
for hard theoretical problems. That's very encouraging.
You got your one, two, three punch in there. Okay.
When we last spoke, you shared an intriguing perspective on AI.
You said to me, I think this technology
(27:54):
is going to force us to turn our brains on, not off as many fear.
So to expand on this perspective, because I think there are many people out
there who hold the opposite viewpoint.
Yes. And you know, the jury's out. Let's see how things go.
But the observation is as follows. The current paradigm for generative artificial
(28:18):
artificial intelligence, not to go into too many details, but these models,
these large language models are stochastic in nature.
They are non-deterministic. So you could ask the same question to an LLM and
you could get a different answer depending on some other settings that you might have put in.
And that's disturbing. That's in some ways not reassuring at all.
(28:40):
We've all heard about the issue of hallucinations, you know,
is the model actually responding truthfully?
And and and that's a valid but slightly
different pocket but i think if we just focus on the fact that these
are non-deterministic models what's happening is that we're getting tools that
allow us to do many things much faster at the expense of having some variance
(29:02):
in the actual result and so you know you have the option to do something very
precisely and meticulously but in a very very
slow manner, or something that may be right x percent of the time,
you know, x may be 90, maybe 95, it may be 98, it may be 99,
but that will get something wrong y percent of the time.
And they can do it almost immediately or in a matter of seconds or sometimes
(29:25):
minutes for the most complex cases.
And so I think the question is, rather than say that one is always better than the other,
it's to realize that we're going to have to reason in terms of distributions
and that many of the tasks that we've been doing in our jobs,
in knowledge worker jobs,
have been reasoning, information gathering, information synthesis,
(29:48):
information refactoring.
Some of these have a pretty high tolerance
to minor mistakes it doesn't really matter if the
way in which you write your meeting notes has you know
a slightly verbose conclusion it doesn't really matter
if you can spot an obvious
repetition of something in a paragraph that you just delete before sending and
(30:09):
so we have to really try and compare the roi we're getting from the much shorter
time invested and the actual value of the job and so what i was saying when
I was talking about turning our brains on is that I think we'll be collaborating
with this technology rather than delegating in many cases,
and that the augmentation that we will get from having these tools at our disposal
will also require that we have our brains turned on and that we're quite focused
(30:33):
on bringing out our critical sense to the decision as to whether or not the
decision is the right one,
the context was brought in correctly, a certain number of checkpoints have been validated.
This notion of human in the loop is, I think, pretty key.
For the tasks where there really is no value in even considering the fact that
sometimes the answer is correct or right and it just gets it 100% of the time
(30:57):
or close enough to 100% that you could say it's not distinguishable,
I think what we have to ask ourselves is.
What were these tasks and what was the perceived value add that humans were
bringing to it before machines could do it?
Because we used to be able to take ourselves from A to B in cities by horse
(31:19):
and carriage, and we found it great.
And I think the skills of being able to tend to a horse while riding a carriage
are a little less valued today than they were back then, because you really
couldn't get very far if you didn't know how to manage a horse and carriage then.
And I think we have technology that just removes that problem and looks at it differently.
So I'm trying to not be naive and I'm trying to not be too gleefully optimistic
(31:42):
in the fact that there will be disruption.
And I think there will be situations where we'll have to ask hard questions
about the ethics of using automated systems and decision making.
And there's many cases where automating it has been either bad or a downright catastrophe.
I always think about Virginia Eubanks' book, Automating Inequality,
where I think in data analysis in the United States, certain minority populations
(32:06):
were at a systemic disadvantage because of the way credit reporting was poorly done.
And in some cases, when credit reporting was not available, people would put
in a zero instead of putting not available.
And that would damage averages as they were computed and have,
obviously, very bad effects.
So there are situations where automated systems will be in critical paths,
(32:28):
and it'll be important to really isolate those and understand that.
But by and large, we are just getting a new tool that is going to make bringing
up to the surface what is left for humans to decide, which can be in some cases
due to aesthetic issues.
Faith-based, principles-based, ethically-based decisions on a number of steps
(32:49):
that have been taken for us, which I don't know people were waking up in a particularly
excited way to go and do in the morning.
I don't think great people or the best team members wake up in the morning and
just like in such a hurry to go and type out meeting notes or just to collect
the file from the CRM database and understand which people we've emailed in the last three months.
I just don't, I don't believe that that's what people were excited to do.
(33:12):
You mentioned earlier on that this is not your first startup,
but I guess this is even so an interesting learning experience for you.
We're always learning, always modifying, always overcoming fresh challenges.
So what are some of the biggest challenges you faced this time around?
And what are some of your major learnings?
(33:35):
Well, I mean, I'm certainly not done. done i think new challenges every day
i think it goes back to you know the the framing and the mindset that you approach
problem with in the first company and by the way my first company was started
with the same business partner stan and i have known each other since undergrad
and we started our previous company together and the starting dust together again so.
(33:55):
I i'd say that by setting the bar
of your ambition higher you're going to see as
potential problems problems some things that you might not have even
considered as problems if if that wasn't if
that wasn't the mindset you were entering the problem with so
you're thinking about how to raise capital who to raise capital from how to
(34:16):
optimize the terms at which you raise that money 12 years ago we were just happy
to have investors return our calls and offer us a term sheet and I we weren't
particularly picky on the terms that came with that term sheet because we didn't
really have leverage to do any better.
But seeing firsthand from my experiences at Allen or at Stripe,
how very thoughtful founders were trying to play a few steps out what the consequences
(34:42):
of those decisions might be on themselves,
their independence of decision making with the board, the ability to attract
and retain top talent with aggressive compensation packages.
All these things come back to bite you if you haven't anticipated them.
And so sometimes I think one of the challenges is trying to also recognize that,
(35:04):
you know, it's still very hard to make perfect decisions.
And the time in which you make those decisions is still one of your assets.
Compared to larger companies, compared to incumbents, the one asset a startup
has is the velocity with which it can move.
And so you don't want to give that asset up by over-optimizing or over-calculating
on too many decisions. That's one challenge.
(35:26):
But other than that, I'd say, you know, it's broadly a lot easier the second time around.
There are certain things that you just don't spend time on or don't worry about
as much when you try and understand that it's all very simple.
You're trying to make something that people really want.
You're trying to get it in front of the kind of people that are likely to want it.
(35:46):
And you have to make sure that when they want it, they're willing to pay something
that is higher than what it costs you to make it.
That's basically what it is. And as soon as you depart apart from those considerations,
you're probably deluding yourself and wasting time on things that matter a lot less.
It's really focusing on those key things day in, day out and not getting too
distracted that is the major lesson, I think.
(36:07):
I love your ability to break complex things down into typically three components.
So now I'm going to put you on the spot. Who are the three people from the world
of tech and the world of the world of tech and the world of entrepreneurship
three people who've most inspired you to be who you are and why.
(36:32):
Good question i think i think one of them is and i'll start with him although
uh saturn adela i think is an inspirational character.
Satya Nadella has added more market cap to Microsoft than Bill Gates ever did.
He's made a few very ambitious decisions or has let very ambitious decisions
(36:57):
be made under his stewardship of Microsoft and has turned a very large ship around,
quote unquote, or at least has made it into one of the most exciting technology
companies to talk about in 2024, which I think if we were having this conversation
in 2010 10 would not seem like a like a likely scenario and and i think that speaks to a tremendous,
(37:20):
yeah an inspirational ability to not be overwhelmed by something that seems
impossible to do and just getting on and doing it i think.
Yeah, actually, I'd say that my co-founder is a source of inspiration.
My co-founder started coding at a very young age and trying to find business
(37:42):
applications for technological hacks and ideas that he would have.
And he started his first company in high school, which was a completely stupid idea.
It was to sell storage online and allow people to have a hard drive in the clouds.
And who on earth would ever want that? and and and
i think understanding that there are many different ways
(38:04):
to see the same reality and that that's okay and then
being able to talk about the differences in those viewpoints and
really break down with no to low to
no ego why you believe the things
you do is one of the most constructive working relationships you can have and
i think that and that you know 20 plus years on the fact that we're working
(38:25):
together again is is is in part because of that i think because of the continuous
desire to continue a relationship that I think is very positive in that way.
And then you asked for three, so I'm going to have to come up with a third one.
And I'd say, I'll cheat a bit, but I'd say that what I'm increasingly seeing
(38:49):
is that people are complex, people are nuanced to every complicated problem.
There's a solution that's simple, elegant and false.
And so rather than take people wholesale and accept the things that they might
do very differently from what you prefer, I'd like to spot what people were
(39:10):
great at and see that as a source of inspiration.
So, you know, I think Steve Jobs was a horrendous manager close up,
but I do think that he was brilliant in pushing for product quality and bringing
in inspiration from fields that,
that had previously not really communicated to each other to maintain a bar
of simplicity and beauty in some of the products that he built. I think that.
(39:36):
I think, you know, John and Patrick Collison,
the founders of Stripe, were always quite impressive in their ability to elevate
and bring back to first principles a certain number of situations that they were faced with.
And, you know, they were obviously very well surrounded and had a lot of people
(39:58):
to go and get that inspiration from, but to try and, you know,
challenge pretty much everything that needed to be challenged rather than just
go ahead with what had been done previously.
And so, you know, the reverse auction for IPO price of Google is another example of that.
I think, you know, it's always cited as one of those moments where the legal
and banking industry is like, what are you actually going to say?
(40:19):
It's like, well, why not?
You know, if people really want to bid for this, they should give their best number.
And more recently, I think I've found myself,
really appreciating some of the ways in which Xavier Niel here in France is
happy to put his reputation and focus in the balance on big ambitious projects.
(40:44):
I think 42 was, you know, people remember him for free. People know him for
some of his more recent projects, obviously, Scaleway and Qtai in artificial intelligence.
But I think 42 was a brilliant move in actually going back to the roots of why
the French educational system is relatively well known for being very elitist
(41:06):
and having a hard selection at many steps of the process, especially after high school. all.
And that comes with pros and cons. And I think what you get when you see people
from that particular track is people who are able to resist a certain amount of stress and pressure.
It doesn't mean that they're great at the things you're hiring them for,
but that's what they were able to do when they were between the ages of 18 and 20.
(41:27):
And I think using that principle to bring people into an environment where you
could actually bring back that high bar of competitive competitive mindset in
a much shorter timeframe,
focused on software development, build a brand around that, that people in less
than a decade now reference as a solid brand for software engineering is amazing.
(41:49):
I think it's hard to, I've always got a lot of respect for ideas that are hard
to imagine the world without once they've been put in the world,
but that nobody had thought of in that obvious a way before,
at least in that ecosystem.
So yeah, there's realizing that I've only mentioned men and so of course my
wife's voice is ringing in the back of my head right now,
(42:13):
but yeah I think we're actually at Dust excited to bring back some of the creative aspects of.
Research and not just development and allowing people
to not just follow a gradient descent in optimizing everything but
allowing them to have the space and time to find
(42:34):
new solutions to problems that we we
were just confronted with and i think marie curie
in france of course i'll go and wave the french
flag again here a bit it's one of those you know one of
those pioneers that proved everybody around her
with her husband of course but okay things that have been previously mentioned
as being impossible which is a question of grit
(42:55):
and perseverance and and
just following on that point you you made reference to females and made reference
to your your wife there how challenging is it for you to build a truly diverse
team at at dust is that something that you've paid much attention to?
(43:16):
Yes, so if I'm not mistaken, it's the 1999 Stanford paper by O'Reilly,
and I'm forgetting a number of the other authors on diversity and team performance.
And I think it's not always quoted the way I think it was intended.
I get a lot of inspiration from the way in which Commando teams were designed
and invented around World War II.
(43:37):
And as a startup, you're dead until proven otherwise.
You're really a nonprofit that's got a few months wants to make a point.
And so what you need is to increase your chances of survival.
And I think NASA or the SAS have a lot in that toolbox.
And what you need is diversity of skills. You need diversity of skills and mindsets
to approach problems in a way that is novel and often superior.
(44:01):
And so that's something that we really try and do.
Pragmatic and modest way in which we do it is that we don't pay too much attention to people's.
Cvs when they come in for interviews and it's not
even recorded in the hiring process right now we
try and have chats a few chats and a screening
on on skills and what we call technical screening before really
spending too much time thinking about what people did
(44:24):
when they were just after high school when it comes
to diversity gender i i think
it just you know you just have to have a
hiring bar and try and increase the top of
the funnel to increase the chances of you meeting a
certain number of diversity goals in in scaling engineering
teams at stripe one of the engineering leaders there as a woman i have a lot
(44:46):
of respect for pretty just pragmatically said you know if if we want to have
more women in seats we just need to see a lot more women at the interview stage
not to adapt our interview bar but force ourselves at the very first steps of recruiting.
So the second, if I'm not mistaken, the third engineer to join Dust is a woman.
(45:08):
She was a fantastic software developer that I'd met in a previous part of my
career and that I was excited to have join us.
The first person to lead the business operations team, also a woman who had
had a very successful career scaling the French branch of an international company.
And I think that makes it easier. when there's seven people in a room and two
(45:30):
of them are women, then it just makes it easier for anybody walking in the door
to find the environment maybe a little less tough to be the first in.
It's a longer topic, and I don't want to skim over the other forms of diversity
that need to be contemplated.
I think in software development, we're fortunate in that it's by and large a new industry.
I think if you're very determined, hardworking, and curious,
(45:54):
there's not a lot that you can't learn and practice online.
And so some of the more traditional ways in which,
elitist groups have favoured themselves and their own are a little challenged by that.
I think we've met some incredible candidates who came to software development
or the tech world more broadly by very surprising ways.
(46:16):
And one of them was always curious to, you know, their ability to just be in
wonder and or at something new that they just wanted to get their hands dirty with.
So I think as we grow, we'll also have to continue to understand what for the
brand, the employer brand that we build, all the role that we want to play in
the ecosystem that we're in, we want to do. I think there's a time and place for everything.
(46:38):
I don't believe that a 12- or 13-person company can do everything,
but I think we can always challenge ourselves to do more.
So we incorporated the company in France. That wasn't something that a lot of
people told us to do. We decided to build a team here.
That wasn't necessarily something that a lot of people encouraged us to do.
I think we're trying to challenge ourselves to be positive contributors in the
(47:01):
ecosystems that we want to live in.
And when I look at some of the ways in which this industry has virtualized and.
Digitized almost everything, I am left with the question of,
you know, how does it positively contribute to the streets that you live on
and the parks that you walk by and the places that you actually want your kids to go to school in?
And so building teams more locally
(47:24):
in places that you want to live in seems like
a pretty sure way to redistribute uh obviously
hopefully positive outcomes where uh where you're going
to to spend your time that's uh one of the ways in which
i think it's a lovely lovely expressive way of getting your points across i
wish you and your diverse team a dust huge never diverse enough but you know
(47:50):
diverse in skills for sure and and and and hopefully more diverse in many more ways over time.
Well, in your increasingly diverse team, wish you huge success cracking team productivity with AI.
And thank you so much for joining me on today's show.
Thank you very much for having me, Gary. It was a pleasure to chat.
Popular Podcasts
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.
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
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.