(speaking french)

(speaking french)

Matias, thank you for joining.

Thank you for having me. Really happy to be here.

I'm really excited to chat because you have a very unique background.

You've crossed a bunch of worlds.

You work in consumer products.

You worked on developer platforms like at Facebook in the early days

and now leading product for Alchemy in the crypto space.

But what I find even more interesting is that you're not an engineer by training,

and yet you're building more than many engineers that I know.

How do you go about it?

I think it just comes down to being interested in a lot of things

and trying a lot of things over time

and getting better and better as models, I guess, get better as well.

That's awesome.

So today you're using AI a ton at work and outside of work.

But let's start with, like, Alchemy.

Like what was, aha moment you had maybe kind of bringing AI to work?

The first thing that I remember the first time we used AI,

I guess it was Codex in Slack.

Right.

It's probably a year ago or so at this point.

Yeah.

The first thing we used it for was to make small edits

to our documentation, our developer docs from Slack.

So we went from having to run the site locally and doing that whole process

that is fairly involved.

To just being able to add Codex on Slack in the docs channel in our company Slack

and to be able to just tell it to make changes.

We do that all the time at OpenAI too.

We still do it.

Now we have a bit more of a sophisticated setup

but that was really the first thing that we used

Codex for and AI in general at the company.

From there, the other big moment I remember

that was probably the bigger turning point, was using code review.

And so there was this instance, we had a small incident

that was related to a big migration that had happened months prior.

And in big migrations, some bugs will happen.

It was like a lot of code.

We had an internal postmortem about the incident

and we basically identify what the race condition was.

We fixed it and someone on the team had the idea to retroactively run code review, Codex,

to see if Codex would have caught the bug.

And it did.

Wow.

And so we did that a couple of times just to see.

People were starting to get curious and

you could tell that was a bit of a turning point in terms of model capability

and like a lot of the ingredients were right for for teams to start adopting those tools more.

From there, I remember a couple of days after seeing one of the engineers on the team

essentially using Codex review in pull request comments as his teammate.

He was just he submitted a PR that I saw go through Slack.

I click through it and I saw him do at Codex review,

address the comments that Codex would come up with, then review again.

Do that again and he was just doing this back and forth with Codex.

You could tell that people were getting over the hurdle of,

of like thinking that LLMs weren't good enough to help in a professional setting

with a really large complex code base,

with code that needs to be good enough to be served to a lot of people.

That’s really cool.

That was really a turning point.

Yeah, that's a very interesting point on code review.

I remember talking to many companies were just getting into like, you know, AI coding.

Code review was the first moment when they realized

that many incidents they had could be caught actually by Codex automatically.

And they replayed some of those similar to Alchemy.

I remember Datadog, for instance, back in January, said that

like more than one incident out of five would have been saved by Codex.

But I'm sure the model capabilities now are so good

with GPT 5.5, it could be more than half, easily.

Nine out of ten, maybe.

Maybe most of it.

Yeah. So code review was really a turning point for many companies

to get into Codex and really realize the power of AI.

Can you tell me a bit more about how you're using Codex at work today?

I use it a lot for a lot of different things.

A lot of those typical product management sort of tasks, writing documents.

We've created all these skills internally to help us

do the PM job more easily and better and faster.

That includes writing PRDs, analyzing customer feedback, like doing all those things.

I use Codex for that

and several members of the team reuse those skills as well.

We have this shared, repo of skills across the company,

across all the different functions, so that not only we can do the same

job and the same tasks faster and better

but also more people who are not necessarily PMs are able to do those things.

I'm sure, yeah.

And I think it's also telling a lot about, like, how we think about developers now,

not just as humans but also as agents consuming our own platforms and infrastructure.

We have to think about these agents and making sure that, like, tools like Codex

can actually integrate, like OpenAI’s API or Alchemy’s infrastructure very quickly.

Yeah, it's really several things that change at the same time.

First, the way we work has changed and Alchemy has a lot of very technical products.

It's an infrastructure company first.

And so all the developers internally have changed the way they work.

And so it's that's very clear to everyone now.

But that also means as builders of a developer platform, we have to change

the way we build tools so that other developers that we know

are also changing their workflows with AI are able to use Alchemy better.

So it's, the developer we're building for has evolved.

Now, we are very much assuming that 100% of developers are building

software with the help of AI.

Right.

And so it's evolving your platform to fit that new reality as one thing.

And then the other thing that's a bit more nascent

but that's also very interesting is, I guess,

this new type of developer that's an agent sometimes, autonomous agent.

And they don't always need the same tooling.

They need different things.

And if an autonomous agent shows up on Alchemy

and as far as implementation decisions go, they're fully autonomous.

They have to be able to sign up and integrate something

and use the blockchain to perform some kind of tasks.

We need to build the tools for that.

And it's building for those two.

And over time those two probably converge.

But for now they're still fairly different needs.

And my job is basically to try to figure out what we do for both.

Awesome.

Zooming out of Alchemy a little bit, I was reflecting recently

about my own startup.

And you were a founder before, as well.

How do you think now, about this time where you had to build without AI?

It was very different.

And it was hard, even if you were not an engineer.

Even more difficult, like the typical way you'd go about it 6 or 7 years ago

was to build some kind of prototype, to try to raise enough money

to hire a small team, to then build the real thing.

And that that's pretty much what I did with with my company at the time,

built the first prototype myself.

It was very difficult because it was essentially copy pasting code left and right until something worked.

Managed to raise a bit of money, hired a team of 3 or 4 engineers

to help build the first MVP and not only that took,

a decent amount of money and several people, but also took months.

I mean I haven't thought about this much

but I wonder how long it would take me to build the first version of the app

we’re a building at the time with AI today, just by myself.

My guess is that it would take less than a week.

I don't know what your experience was like.

You are building a much more complicated product than mine, I guess.

How long did I take it back then and how many people?

It took me a team of like 15 engineers for maybe a year and a half

to get to like the first like milestone of a V1 that we shipped

to like many customers.

It's pretty incredible how far we've come.

Like it's just never been easier to do something.

Yeah.

And to start something from scratch.

Yeah.

It's probably the best time to be a founder.

Like any idea you have, you can just build it.

Yeah.

Anyone who has an idea now has the tools to basically try it out.

Speaking of things that you do on the side, it's a great segue

because I wanted to ask you about how you're using AI outside of work.

It sounds like you're building so many projects these days

it's almost like one a day.

You know, it's amazing because it's never been easier to build things.

And yet I just kept feeling this sort of anxiety

that I, I wasn't doing enough pretty much all the time.

I think I'm a bit past that.

But the reason I am is because I try to work

on a setup that allows me to spend little time on my computer

and have essentially Codex work for me for hours while I'm doing something else.

It's like

I had gotten to a point a few months ago where I felt bad about going out

because I felt like I should be taking advantage of this

amazing technology we have and I should just be building things all the time.

I didn't want to do anything else.

I didn't want to feel like that either.

I didn't like that trade off.

And so I invested in a set of skills and a process and several things

that I set up for myself to basically be able to very quickly

either describe a new idea and have Codex come up

with a plan, implement it, test it, and let me know when it's done.

And it's very easy to to get Codex to work for hours at that point.

Or if I have an existing product and I want to explore new things,

I came up with this way of essentially having Codex

do research for me on what new features I should build.

It will look at other competitors.

It will look at other ideas.

It will try to figure out what my goals are

and what my own preferences are.

And I created this set of skills that basically gets Codex

to build all those things.

So it comes up with ideas.

It builds them in a modular way.

So it's going to feature flag them and build them

as experiments in the main app so that I can prompt Codex before going to bed

to go do some research about an app that I have.

Tell it to build the top ten features it can come up with,

build them as experiments in the product.

Wake up to ten feature flags that I can toggle on and off

automatically to basically

not be the bottleneck when it comes to generating new ideas

but still having the control of deciding what is good enough to stay in the project or not.

I love that. That's really cool.

It's like without spending a ton of time on a computer

have these moments, like during my lunch break at work,

I have my personal laptop and I try to send things to Codex

so that it can work while I go back to work

before going out on weekends to dinner, or to have drinks before going to bed.

I have these Codex moments, I guess, where I just try to dispatch a bunch of work,

use all these processes and these workflows I have with skills

and other actions so that Codex can work while I'm not paying attention.

And it's amazing how much you can do if you kind of invest in it that way.

That's so cool.

I mean, honestly, I would love to see your setup and how you're using the Codex app

and kind of bringing all of these tools together.

Yeah. Let's check it out.

Let's check out linear to start.

So what you see here is basically a collection of projects.

You see there's this 12 of them in progress. At the same time.

I'm not working on all of them

in one go, but maybe we can pick one and I can show you how this works.

This one, for example is a mac app that since then

I've also built it as an iOS app that uses Codex App server

to essentially help me write better while I'm at my computer.

It's something that I built for myself

for work first, and that, I kept building out since.

The way it works is that you have a global shortcut.

In my case, it's like command shift spacebar,

and it brings up this little window that you see there.

That's cool.

And this is using Codex App Server.

So in the settings it's using my GPT subscription behind the scenes.

And if you were to write something in my case, I'm not even going to write.

I usually talk to my computer just because it's so much faster for everything.

Like people at work a year ago would make fun of my little headset.

I basically looked like a call center.

Call center operator.

And now a lot of people have those similar headsets, so it's okay.

But if you say you're on slack and you want to write something,

Yeah.

but you need to clean it up

and you want AI to make it sound more professional or fix typos or things like that

like I'm going to record something that would be like,

“Hey, I'm not totally sure I understand what you're saying.

Maybe we should just get together and figure it out” or something like that.

Obviously, this is not a great sentence.

I just dictated that to my computer.

And then if you hit command return in this case,

it's going to rewrite it for me in a way that sounds more professional

because the professional mode was selected.

This is basically using Codex Server behind the scenes and GPT 5.5.

You have a model picker.

I just added 5.5 since it just came out.

And this is just an example of a little app that I made for myself for Mac.

I put the same setup in an iOS app after to have

like a keyboard extension and help you rewrite things quickly and so on.

I don't know if a lot of people use their,

their Codex subscription for things that are not just coding.

I know that people use the Codex app server to build other

coding interfaces, but in this case I basically used it

because I wanted the inference to be done, through my ChatGPT account.

And if we go back to to the process and how to find time,

what you see here is this linear project.

I didn't create this project.

There are like 159 issues that are done, a bunch that are in review

and some that are in progress.

I did not create or write a single one of these.

Wow.

Codex did.

What I did is to talk to Codex about what I wanted

through these skills that I wrote for myself, to basically help

me create a plan that turns into milestones

that you see here, like the foundation and the preferences and so on, a bunch of the features.

Codex will create that plan,

break it down into milestones, create the tasks,

work on them in a way that makes sense based on this agent’s .md file that I created,

that essentially summarizes the way I like to work.

And I put in a bunch of time and, putting that in a single file.

So when I create a new project or I have a new idea, I can basically just

initialize that project with that agent’s .md file.

From there, I use this skill I have to create a plan

and after that it's just telling Codex, build the plan.

That's really cool because you're using linear as the interface

as like your backlog and your release milestones and things like this.

But like Codex is not just writing

all the issues, it's also picking them up to close them and work on them.

Yeah.

It's of both delegating the, I guess, the project management

and the setting up of the project and the milestones and the tasks as well as the

the execution of the tasks too.

I’m initiating the project, sharing the initial idea,

giving Codex enough information about my preferences

and what I want the product to be so that it builds

the version of the product that I want.

I think the biggest gap when you code with LLMs, when

if the LLM produces something that's surprising to you, in a negative way,

it didn't do what you wanted, usually it's because it had to make one or

multiple assumptions and it just didn't make them in the same way you would have.

And so in order to avoid that, I have this process

that requires me to clarify all those things upfront.

And once Codex has all of that, it can just go and build.

That's really cool.

And so that's the first part and if you go back to

to this writing app as an example, the next step after

setting up a new project, building a V1 of a product

and having Codex essentially do all the work for me, usually while I'm not even on my computer.

I wanted to find a way to generate

new features and keep working on a

on an existing product without being the bottleneck.

I didn't want all the ideas to have to come from me

but I still wanted some level of control

when it comes to deciding whether or not I like that feature.

Right.

So I came up with this process that is basically a way

to build experiments with Codex on an existing project.

And it's a skill, again, that

essentially has me tell Codex I'm interested

in building features in this direction, or this is the high level goal that I have.

Go do some research about what types of features you could build.

Go figure out what the competitors to my products are building, things like that.

And it will come up with a lot of features based on what I'm saying.

And it's going to set them up and build them with feature flags

as experiments in the code base.

And what that means is that you're able to do things like what you see here.

You have

You can flip any switch of any feature.

That's right.

So you have a long list of of experimental feature flags.

These were all built in one go.

One evening while I was sleeping.

So before going to bed I said,

“Hey Codex, I'm building this AI writing assistant for myself.

Go figure out who else is building that.

What kind of features do they have?

Score the ones that are most interesting, most impactful, something like that.

Build all of them but build them in a modular way so that when I wake up,

I can toggle each one on or off independently to see if I like them.

And so you basically see a list of all of those there.

We can pick a few if you want to see.

Like we're going to output language is basically going to translate

Sure.

just going to

It’s fascinating that you did not use Codex just to make the plan and implement

but also use Codex

even before to make the research online about what the feature should even be.

Yeah.

And so Codex is great.

It's it's way better than me at doing research.

It doesn't it doesn't get bored of searching for things.

And so as long as you tell it what to research

and how you expect the findings to be structured

and in my case, you also tell me what to do with those findings.

It's basically just a sequence of fairly small steps.

but it will do all of them in one go, because it's all encoded in a skill or a set of skills.

And so it's able to come up with all of this build all of it,

build in a way that makes it easy for me to test

and makes it easy for me to decide whether or not I want it to stay in the product.

And so we selected a few here.

If we go back to the rewriting.

They just showed up instantly.

They just show up there.

And so in this case you see that you have a translation feature.

In this case you can change the

the output format of the message as an example.

It's like Slack style.

Yeah.

And so those are all small features that I got Codex to ideate

and build in a way that,

makes it really easy for me to essentially decide

if I want that to stay in the product or not.

So, those are all the things that I ended up coming up with to try not to spend

too much time building in a way but still end up building quite a few things.

It's really cool.

Instead of starting from scratch, it's basically

trying to think about the things you do repeatedly, or even trying to think

about the ways you like to do things, putting that into a skills,

putting some of that in an agent’s .md file.

Sometimes it's additional tooling and other things that make you go even beyond that.

And then having AI do most of the work.

That's awesome.

Have you built any other like projects on top of the Codex Observer?

Maybe more like on the coding side?

I built a few things like, this is an AI writing assistant.

It's a mac app. It's a consumer.

I use OpenClaw and I've built a couple of other things on my phone, even an Apple Watch app.

That are

It was just me tinkering, trying to see if I could have

different or new entry points to start a coding job, essentially.

So, in OpenClaw, for example, I set up my OpenClaw that my assistant's

name is Lou has got a pretty funny profile picture if you want to see it.

Yeah.

It’s

look at this guy.

[Laughs]

So what do you do with your OpenClaw here?

What do you talk about?

This little guy, Lou, it basically, it mostly has two jobs.

The first one is to make jokes in group chats with friends.

It's not very useful but it's quite funny.

The second one is to just help me code.

And so it's connected at home.

It's running on a dedicated machine.

It's got access to Codex and it's got integrations

that make it easier for me to do certain things on the go.

For example, I have an OpenClaw discord and Codex integration,

so that I have a bunch of discord channels.

Each one is bound to a specific repo.

And so if I want to do something from my phone on the go and I wanted to

to be able to run at home and do things that maybe aren't so easy

to do yet on the cloud or things like that.

I can just do it while I'm walking around somewhere.

I talk through discord by talking to Lou and it's going to work on Codex.

That's an example.

Another thing that I did recently was try to see if I could

build an Apple Watch complication on the home screen

so that you could record a short voice memo and have that trigger a Codex job.

So, like, imagine you found a typo on a website or something like that.

Obviously it mostly works for one shot things but you would just

very briefly like you would record a voice memo to say, “Hey,

fix this typo on the landing page of this website.”

And you have the watch and you have the companion iPhone app to do that.

Yeah. So, let me show you this, this thing that I built.

This is my Apple Watch.

What I essentially built here is a way to just send a voice memo

like create a test.md file in my skills repo.

And that basically sent that memo to this iPhone app.

That is now, it transcribed the message, it understood the intent.

Based on that message is going to route it to the correct place.

In this case, it's going to understand that the skills repo means my own personal skills repo

because it has context on my GitHub account

and then it's going to run the execution of that task.

So in a couple of minutes I'm basically going to get a notification that says it created a

That second one is now successful?

Yeah.

So now it's running the execution here.

And this is all running with like Codex app server behind the scenes?

Yeah.

So if you if you go to the settings here,

you basically see that, there's two different ways it can run.

There's the typical you put an openAI API key and you can do it that way.

And the other one, what is here called back in session

is basically Codex app server running somewhere else.

That's so cool.

Roaming off of my account, connected to my GitHub with all my context.

I love this.

I can basically just start Codex jobs on the go.

It's like

it's not useful for everything but it's one of the ways

that I kind of like to play around with what becomes possible.

It's like one step closer to making

I guess, the entire implementation and implementation detail.

It's like I just talked to my watch for 10 seconds and it did absolutely everything the right way.

Incredible.

I don't think we're always there yet, depending on, you know, how complex

the actual ask is, but for small things it's kind of convenient.

It was kind of fun to play around with just to see how to work with Codex App Server.

And yeah, I just kind of try to do that.

And by the way, this is exactly what I love to see, right?

Like we open sourced the Codex CLI, the Codex harness,

and we also open sourced the Codex app server just for like fostering this like open ecosystem.

Like a builder, like you can extend the way

you use Codex in this case from your watch or your phone.

Even though, you know, we have not shipped that yet.

So this is really, really cool how you can use your coding agent the way you want.

I think you guys should build that

I think we should.

That complication on Apple Watch and on my iPhone home screen so I don't have to do it myself.

That'd be great.

Sounds like a great idea.

Let's see.

Look at that.

The whole request just arrived.

The request is right there.

Alright, out of this test.md file obviously not the most useful task but

but it just shows that it works. It's kind of fun to build as well.

Pretty cool.

That's really cool that you build this.

And by the way this is exactly why we open source not just to the Codex CLI

and the Codex harness but also the Codex App Server.

We want builders like you

to be able to kind of poke at it and also integrate that in their own ways.

It's really cool that you can have a coding agent

from your watch, which we don't have it just yet, you know.

You better build it.

We we probably should build it.

Yeah.

It's a great idea.

I'm curious,

like, we launched a ton of things in the past, like, few weeks,

both on the Codex side and GPT 5.5 our latest model.

Have you had some, like, surprising moments with these two?

Yeah, there's a couple of things.

The first one I use computer use recently that was,

it was a lot better than what I remembered or I guess I was surprised by how good it was.

I was at home last weekend doing something on a Raspberry Pi

and I needed to to do something pretty tedious.

Like copy paste a bunch of different domain URLs and go put them in some admin panel somewhere.

And I didn't really want to do it by hand.

So I told Codex to basically SSH into the Raspberry Pi.

It was like dangling from a TV to go do it for me.

And it just did.

Wow.

It found a way to just connect to the Raspberry Pi through my browser.

It basically found all these URLs that I had in a file that I gave it.

It just did the whole thing.

What was funny is that I was trying to use computer use, I guess, to save time.

And I spent the entire time just watching it.

That’s so funny.

Doing the thing for me because I was impressed by how I was going to the right place,

opening the right thing, copy pasting stuff for me.

It was just the funny thing.

I love how the cursor moves also in that you can kind of follow along the animations.

It's pretty but you don't want to touch it but you kind of do at the same time.

Because you can actually let it disappear in the background too.

Yeah, that was pretty cool and recent.

The other thing I guess is 5.5 just launched

and I have this old hackathon project that I've had for over ten years

that I use almost like my personal eval in a way, that is,

I know that models nowadays can one shot this project

but every time a new model comes out I try to do it again

just to see how much better it gets.

How many more details it gets right. How the design improves.

It's a bit of a silly project but I can tell you about it.

It was, it's a thing that was called SnapCat, which is a bit absurd.

It was basically this app for cats to take selfies.

And the way that it works is that, I mean, I guess I could just show you.

But it's essentially

a black screen with a red laser dot that moves around.

And

For the cat to catch?

Right. And so if you are a cat, we can pretend you're a cat for a second.

And you try to catch that. Every time you did

and if you don't mind trying to tap to catch the dot.

I guess I'm not as fast as a cat.

You're really not a very good

I just did.

I think I got it.

There you go.

So as you do that, every time you hit the viewport

you’re basically taking a photo with the front facing camera.

Woah

And so you can exit this cat mode with the volume buttons,

which cats don't really have easy access to.

At that point you go to the gallery.

You see the upside down pictures you just took of yourself as a cat.

You can basically select one, share it and that kind of stuff.

And what's cool here is two things.

The first one, it easily one shot this project.

It's something that over ten years ago we did at a hackathon.

It took us over a full day to build.

We had a team of five people.

It was quite difficult to do really quickly.

Now this is one really well crafted prompt and a couple of skills

and it can build the whole thing for you in one go.

How did you build the UI?

Was it like 5.5 building the UI? Was it image gen?

That's the second thing, so the first thing, it can one shot the skeleton of this app really easily.

What it did in this case after that was

I described the style of an app I wanted.

I said I wanted something light and colorful and playful and that kind of stuff given the theme.

I generated an image of that UI, and then I told Codex to build that.

And once it did that with the camera roll view that you're looking at right now,

I like that style.

And I told it “Go redesign the settings page based on the same style.”

And it generated another image and then implemented it.

Then it did the same for the home screen.

So it's a pretty different and new way to go about building UI.

So that's your eval, every time there's a new model you're just rebuilding SnapCat?

I basically just look at how easy it is for cats to take selfies.

Is it the best you've seen so far?

It's the best by far.

It is, it’s a lot better than the last time I tried this a few months ago.

It is infinitely better than what we managed to build in a full 24 hours.

Wow.

And I basically made this in one shot last night in preparation for today.

Incredible, I love this.

Well, I guess now you have selfies of me among your cats.

I'll text you one of them.

That's so cool.

Thanks so much for showing me your setup.

I love watching users building with the Codex app.

Maybe to to close out,

what would be your advice for other founders or builders, who would like to master the codex the way you do?

You know, I think it boils down to a few assumptions

that I've made early on that I think have been really, really helpful for me.

The first one is assume it's possible.

Like if you have an idea, more often than not, it can probably be done.

The second one is assume you can do it.

That one's a bit more personal

but a lot of people could be building these things.

And I think the blockers that they think they can't.

I think assume you can.

In this case, maybe be in denial even that you can't do it.

It's usually true.

And the third one is when you try to get something out of an LLM and it doesn't work,

this one's a bit difficult,

you have to put your ego aside, assume it's your fault.

Don't assume that the tool isn't capable.

Assume you haven't found a way to communicate it or convey what you want just yet.

And I guess try again. Like I've done a lot of trying again.

And I think that's what's gotten me quite a long way into building things.

So that's probably the advice I give to people.

I love that and it feels like the perfect note to end.

Anyone can just build anything.

It's true.

It sounds cheesy but it's true.

Thank you so much Matias. This was a great conversation.

Merci.

Thank you.

(speaking french)