Theodoros Ampas

Theodoros AmpasSep 13, 2026

I've stopped trusting the phrase "we don't reject candidates, we just rank them." That's the standard defense every screening vendor gives: a recruiter still clicks reject, so the tool never officially decides anything. Technically true. Operationally hollow. When hundreds of applications arrive for one role and a recruiter works down a ranked list until the interview calendar is full, the ranking already decided who gets seen. The click just confirms a decision made earlier, by a system nobody has to name in the rejection email. An adjacent regulated decision looks nothing like this. An insurance claim denial comes with a written reason attached to a named reviewer. A candidate who lands below the fold of a ranked list gets nothing — no reason, no reviewer, no record that anyone actually looked. If a company is going to let a model decide who gets an interview in practice, the fix isn't switching the tool off. It's writing down, per rejected candidate, which criterion the ranking weighted against them — the same accountability a denial letter already carries in a different industry. The question worth asking anyone running one of these pipelines: if a regulator asked for that reason on your last ten rejections, could you produce it?

Theodoros AmpasSep 17, 2026

Last week one API call took out every metric I track, because I'd bundled three of them into a single request. The platform I pull analytics from had quietly retired two of the three field names I was asking for. The whole request came back with a 400 — and took the one metric that was still alive down with it. The easy fix is updating the field list. I did that. But this vendor has deprecated fields three times in eighteen months, so the list will just rot again, and I'll be back here in a few months with the same outage. The real fix was smaller: stop asking for everything in one shot. Request each metric on its own. When one fails, keep whatever the others returned, and mark the failed one "unreadable" — carrying the vendor's actual error — instead of quietly writing it down as zero. That distinction is the whole point. "I can't see this right now" and "this measured at nothing" are different facts. A system that can't tell them apart will happily learn from numbers that were never real. Proof before claims: I'd rather show you the outage and the fix than tell you my scoring is reliable.

Theodoros AmpasSep 17, 2026

A market can be moving in your direction and still not be ready to change its behavior I had the chance to ask Alli Webb, founder of Drybar, a question during a School of Mentors call Drybar was built around the polished blowout Her new brand, Messy By Alli Webb is almost the opposite idea: embracing natural texture and moving away from perfection. So I asked her: How did you know this was a real market shift and not just your own preferences changing? Her answer stuck with me It started personally during COVID. She stopped using as much heat, noticed her hair becoming healthier and started embracing its natural texture But she did not immediately turn that into a business She kept watching She noticed the same shift appearing in beauty, advertising, red carpets and the way women were talking about their hair Only then did the personal observation start looking like a market signal. But the part I found most interesting came after. Even when the trend is real, getting people to actually change an established behavior is still difficult That distinction applies far beyond beauty. I see the same thing with AI. A company can believe AI is important. A team can understand the opportunity. The technology can work. And adoption can still fail because changing the way people actually work is a completely different problem. The lesson I took away: Spotting the shift is only the beginning. Understanding whether people are actually ready to change their behavior is the harder part Big thanks to The School of Hard Knocks for putting these conversations together and to Athan Garcia for hosting and creating the space to ask real questions