CZ Podcast #359: PostHog
In this episode of CZ Podcast #359, we welcomed Vojta Bartoš and Tomík Belada.
I met Vojta a few years ago when I interviewed him for a job at Ataccama. I think he ended up turning us down and trying his luck with his own startup.
Then we crossed paths at the IRONMAN triathlon in Hradec Králové and again at Czechman this year. That’s where he told me he’d been at PostHog for a few months and that the work was incredibly interesting and just completely different. Then came the story about an offsite in Barbados. One thing led to another, and we finally got Vojta on the podcast.
I use PostHog exclusively through Claude Desktop, so I thought we should invite Tomík, who looks after PostHog for us at Ataccama.
What did we get into?
The job interview
Vojta talked us through his PostHog interview. One of the less usual bits caught my attention: a paid, one-day trial PostHog calls SuperDay.
Candidates get a product brief and have a day to show how they tackle it, flesh it out, and get something working. By the way, they get paid $1,000 for the day.
What’s interesting is that the brief is deliberately too big to finish in a day. The point is to see what you can code, how you think and prioritize, and what you can actually deliver.
The ideal AI team size
How big should a team be when it builds software with LLM-powered tools and AI agents?
I think it should be quite a bit smaller than before. A lot of the cognitive load of explaining things, getting people to agree, and persuading them goes away. The more you can do on your own with agents, the less coordination you need.
We even touched on one-person teams. But don’t worry, one size doesn’t fit all. Whether you have three people or four is hardly the deciding factor.
PostHog itself runs on small, autonomous teams of two to six people, each meant to work almost like its own startup.
The question, of course, is what that autonomy costs. Depending on individuals, sharing knowledge, keeping the product going, covering for someone who’s away. None of that goes away just because we have AI agents.
Bottlenecks
AI makes generating code dramatically faster and shifts the bottlenecks elsewhere.
Pull/merge requests and code review. Both Vojta and Tomík said that after evaluating the available tools, they ended up using their own code review bots anyway. PostHog even has its own ReviewHog and StampHog.
I find that an interesting paradox. We can generate a lot more code than we used to, but someone still has to understand it, review it, and take responsibility for it. Writing code got faster. Understanding it may still take just as long.
AI slop grenades. A pile of generated code the author doesn’t understand, leaving the next person in the chain to make sense of it.
Vojta said he doesn’t write any code by hand anymore. That makes it all the more important that he reads it and understands what he sends downstream. Otherwise he’s just lobbing an AI slop grenade at somebody else.
Platform engineering and DevEx. As more people use AI agents, I think platform teams will become even more important. Developers and their agents need an environment where they can work on their own, run tests, and ship changes to production safely.
I was surprised when Vojta mentioned that their Developer Experience team has just three people.
They handle CI, the merge queue, code review bots, code quality, and a whole bunch of other things that keep the rest of the teams going. Developer Experience is growing to include Agent Experience, too.
Product engineers, or developers at the feeding trough
For me, the bottom line of the whole episode is the big shift toward product engineers.
To me, that’s a clear paradigm shift. You go from the developer who, as Filemon put it so well on the podcast, sits at the feeding trough waiting for a Jira ticket to appear, to the developer who talks to customers regularly, understands their needs and problems, and thinks for themselves about what to build and why.
As AI gradually takes over writing the code, I think understanding the problem and deciding what’s worth building will matter more than ever.
UI is dead?
On the product side, we were pretty much in agreement about UI is dead and what might take its place: MCP tools and public APIs.
Vojta said PostHog is working from the premise that by 2030, the UI won’t be the main way to interact with the product.
I find that especially interesting. If AI agents become the main users of our products, what the product lets you do through APIs and MCP will suddenly matter a lot more than how many screens and buttons the UI has.
By the way, Vojta mentioned more than 1,200 MCP tools on the podcast. To give you an idea, here’s their public list.
Change is the only constant
My favorite topic to end on.
PostHog seems incredibly fluid when it comes to organizational change. Vojta said he was already on his fourth team after just a few months.
I always say change is the only constant. But this kind of setup only works if you have the culture to support it.
And that takes me right back to the beginning. How PostHog chooses people, how it works with them during onboarding, what feedback they get, and how decisively it deals with things when the fit isn’t right.
Without that, small autonomous teams are just another org chart.
Related articles
Related CZ Podcast episodes
- #357 · Převážně nevážně
Dagi and Filemon discuss what has caught their attention and what has been bothering them lately. · Czech audio
- #355 · Efektivnější AI modely s BottleCap AI
BottleCap AI researchers discuss more efficient models. · Czech audio
- #261 · Leadership
On leading people and teams. · Czech audio