Last month, Take-Two CEO Strauss Zelnick called the idea that AI could generate the next Grand Theft Auto "laughable", "not even the littlest bit." He also pointed out something quieter and more interesting: generative AI didn't actually lower GTA 6's budget. The tools helped here and there. The savings never showed up. The hit is still coming from the humans.
A lot of founders are writing a different memo right now.
The delusion is well-funded
In February, Block went from 10,000 employees to under 6,000, with Jack Dorsey explicitly blaming AI in a letter to shareholders. Meta begins its own 8,000-role layoff today, citing reallocation toward AI infrastructure. According to Nikkei Asia, nearly half of all tech layoffs in Q1 2026, about 37,000 of 78,000 cuts, were directly attributed to AI and automation. The shareholder story is clean: fewer engineers, same output, model picks up the slack.
Then the METR randomized controlled trial walked into the room.
Experienced developers were given AI coding tools to work on their own codebases. They predicted a 24% speedup. After the study, they felt 20% faster. They were actually 19% slower. A 43-point gap between the vibe and the stopwatch. That, with respect, is the most expensive feeling in modern tech.
What AI is actually for
Strip away the hype and AI does one thing genuinely well: it collapses grunt work. Boilerplate, scaffolding, the first draft of anything. Used properly, it reduces the cognitive load of execution so your team can spend more hours in the 20% where senior judgment lives, architecture, taste, what to build, what to kill.
That is what AI is for.
What it is increasingly being used for is something else entirely: to reduce the emotional load of leadership. Hard conversations are tiring. Performance management is tiring. Telling a team you over-hired is tiring. So executives reach for the algorithm as cover, and Fortune reports a similar pattern, with HR leaders admitting many "AI layoffs" lacked any real strategic intent.
That isn't strategy. It's cowardice with a deck.
How we actually use AI at Propello
At Propello, AI is everywhere in our stack. Claude writes alongside our engineers. Our content team ships drafts faster than ever. Research that used to eat a week now eats an afternoon. We genuinely love the tools.
And yet, across Define, Design, Develop, and Launch, every single thing a client pays us for is decided by a human. Which problem matters. Which architecture survives the next two years. Why the brand should feel this way and not that. A model can give you ten options. It cannot tell you which one to bet the company on.
We're turning that experience into a series. Everything we publish here on the pragmatic use of AI, for engineers, designers, marketers, and every other function that keeps a real business running, will come from our own work with these tools across client projects. Stay tuned.
The reframe worth stealing
AI is now a commodity. Your competitor can rent the same model you can, at the same price, with the same prompts on the same Reddit thread.
What is not a commodity is an engineer who can interrogate a problem, push back on a bad spec, talk to a customer without sounding like a stack trace, and decide what is worth building. IBM tripled entry-level hiring this year. Same AI, opposite conclusion. They figured out that if you fire your juniors, you have no seniors in five years. Math is undefeated.
So stop optimizing your bottomline by 12%. Use AI to expand your topline by 300%. Ship the things that were too expensive last year. Build the products you wrote off as fantasy. Hand your team better tools, not pink slips.
Your engineers aren't competing with the model. They're the reason it ships something anyone wants.
P.S. This week, we'll be announcing our first original product, something built specifically for founders and business decision-makers. If you're on this list, you'll see it first.