I have a theory. Most founders talking about AI on podcasts are either selling you fear or selling you a tool. Rarely do you get someone just describing what actually happened inside their company over the last year, mistakes included. That is exactly what I stumbled into while researching how services companies are adapting to AI, and it quietly wrecked an excuse I have been using for months.
The excuse was simple. “We are still figuring out our AI workflow, so we are not behind yet.” Turns out, according to this conversation, that window closed a while ago.
The podcast came from GeekyAnts, an app studio, and featured a casual chat between two people who clearly go way back. There was a whole tangent about a caravan road trip and old studio setups before AI even came up. I almost closed the tab. I am glad I did not, because what followed was less of a pitch and more of a confession.
The line that got me
At one point, the guest said something like this: what used to take months to build now takes hours, but nothing about reviewing, testing, or validating that output has gotten any faster. Read that twice. We have spent two years celebrating how fast AI writes code, and almost no time solving the fact that a human still has to decide if that code is actually right. That gap is quietly becoming the most expensive problem in software right now, and almost nobody is pricing it into their roadmap.
He also mentioned that a well known company reportedly blew through a full year of AI budget in a matter of weeks. I cannot verify the number, but I do not need to. Every founder I know has some version of that story now. The spend moves faster than the strategy.
Where I stopped nodding along
I want to be clear that this podcast is not gospel, and a few claims deserve real pushback before anyone acts on them.
- The idea that two or three people can outperform a fifty person team sounds inspiring at a conference and reckless in an actual hiring decision. It might be true for a scrappy prototype. It falls apart the moment you need long term maintenance, domain memory, or someone to own a decision six months later.
- The push for everyone to become a generalist “builder” ignores that depth still wins in complex systems. Speed without expertise just moves the mistake earlier in the process.
- The claim that junior developers can now perform at a senior level with AI assistance quietly assumes they already know what good output looks like. Reviewing AI generated work well is, ironically, a senior skill.
None of that makes the core message wrong. It just means the honest parts of the podcast are more valuable than the exciting parts, and most viewers will remember it backward.
Why I think this matters more than another AI tool list
What actually stayed with me is not a tool recommendation or a productivity hack. It is the reframing that AI did not remove the hard parts of building software, it just relocated them. Building got easy. Deciding what to build, and confirming it actually solves the problem, got harder to ignore. That is not a comfortable message, but it is a useful one, especially if your team has been measuring AI success purely by how fast something ships.
If you run a company right now and you have not had an honest internal conversation about where your actual bottleneck sits, building or validating, this podcast is worth twenty minutes, even with its rough edges. I am not affiliated with GeekyAnts and cannot speak to what it is like working with them directly, but the way they described the mismatch between client expectations and reality felt earned rather than rehearsed.
So here is the question I am sitting with, and genuinely curious how others would answer. If building is no longer the bottleneck, what is your team actually optimizing for right now, speed, quality, or just looking busy while the ground shifts underneath everyone. I would love to hear how other founders are answering that, because I do not think there is one right answer yet.