Gergely Orosz posted recently about a trend he’s watching:
CTOs, VPEs and heads of engineering at startups and mid-sized companies are leaving, burning out, taking career breaks.
His list of reasons is accurate. I want to add the one I lived, because I don’t think it’s burnout in the way people mean it.
A couple of years ago I took a senior operating seat at a company with real transaction volume and three platforms that had been bolted together during a growth sprint. The mandate on paper was stabilize and scale. The mandate in the room was simpler: use AI to do more with less.
So I did.
We integrated AI into the core development workflow and cut average development time by roughly half. Velocity increased during my time. We:
Rapidly refactored a monolith into modular services
Migrated workloads across clouds
And absorbed a large-scale DDoS attack in hours with no repeat incidents
On the scorecard we were handed, we delivered. That is when the actual problem became visible.
Hitting an AI efficiency number does not close the number. It resets it.
Fifty percent faster becomes the new floor. Twenty-five percent smaller becomes proof the org can be smaller still.
And underneath the arithmetic, something worse was happening: the definition of done was quietly eroding. Speed was measurable. Quality was not. Security was not. Whether we were building what customers actually needed was not.
The trigger for me was watching this pattern show up everywhere, not just in one company. Nontechnical founders read about AI coding gains, hire a technical leader, and expect a production-worthy product on a timeline that assumes judgment is now optional. What ships instead is a feature race. Security becomes a phase two item.
Nobody runs a real product management cycle, because discovering what people genuinely want takes time and AI does not shortcut that. It lets you build the wrong thing faster.
Even an MVP has a quality floor. If people put their information, their money or their family’s wellbeing on the other side of your software, you own that. Nobody in the room is asking about it, and the leader who raises it starts to sound like the obstacle.
Founders are now taking strategic advice from the same tools.
I use AI heavily. When I asked it for go-to-market thinking for PerkyPet, I got a clean, confident, entirely generic answer. If I had walked into an VC meeting with it, I would have been laughed out of the room, and I know that because the investors tested me on exactly that. I had the right answers because I did the reading, talked to veterinarians and learned the industry.
AI cannot do your time in the field for you. It will happily generate something that sounds like it did.
I have spent most of this period in a hospital with my father. I am not going to write about the medical details. What I will write about is what I saw around the care, because it changed how I think about this entire subject.
The failures were not technological. They were:
Miscommunication
Incomplete handoffs
And information that existed somewhere in the building and never reached the person who needed it in time
No tool deployed into that environment would have caught them, because the breakdown was in attention and judgment, not in software.
The exception was Dara.
She used AI throughout, constantly, to validate the medical data in front of her and to advocate for him with the people making decisions. She was right, repeatedly. The recommendations were sound and they were not acted on in time.
She could use AI that way because she is a double board-certified physician who knows when the output is wrong and can stand behind the reading.
Put the same tool in the hands of someone who cannot evaluate it and you do not get her judgment. You get confident answers that nobody in the room is qualified to challenge, arriving faster than anyone can catch them.
That is the whole argument, in the highest-stakes setting I have ever stood in. The tool was not the limiting factor. Whether anyone listened to the person holding it was.
It is also why I am building PerkyPet the way I am.
A pet parent will not have Dara in the room. They will have an app and a worry about their dog at 11pm, and whatever we hand them is what they will act on. If we get that wrong, an animal gets hurt and a family carries it.
So we developed 35 veterinary-validated prognostic indicators before writing a single line of marketing copy.
We run more than 65 AI agents, and Dara owns the clinical and scientific rigor behind all of it. PerkyPet is built to complement veterinary care between visits, never to replace it.
My job is the operating discipline: move slowly on the things that are critical and quickly on the things that are not. A consumer app playbook would have shipped first and validated later. That sequencing is not negotiable here.
Thanks for reading my Substack! This post is public so feel free to share it.
The CTO role is not broken.
Being handed AI as a substitute for strategy, sequencing and senior judgment is what makes it unsurvivable. Those two jobs look identical in a job description and feel nothing alike by month four.
So be honest about which one you’re offering.
If AI is how you plan to hold headcount flat and skip the unglamorous work, you’re not hiring a technology leader. You’re hiring someone to absorb a strategy gap. The good ones figure that out fast, and a lot of them already have.
Check out the Gergely Orosz’s Substack if you’d interested in Engineering Management advice.
Sincerely,
Tim Maliyil, founder of PerkyPet
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.