Dylan Colon · Product, AI, and the unglamorous middle of HCM

Musings

I make theboring parts ofworkdisappear.

I watch how people actually work — the sighs, the workarounds, the ticket filed at 4:58 on a Friday — and turn what they teach me into AI that earns its keep inside HR software.

Stick around. This gets fun.

Product management is a translation problem.

Executives speak in outcomes. Engineers speak in classes and methods. Customers speak in sighs.

Most PMs pick a language and get very good at it. I got greedy. I wanted to hold all three in my head at once — and then, once the tools caught up, to write them down in the same document.

These days my user stories open with the outcome and close with the architecture. Engineers stopped translating and started shipping. Nobody misses the meeting in between.

“A frontline worker should be able to ask ‘where’s my paycheck?’ and get the real answer — theirs, today.”

same sentence, in the story I hand engineering:

As an hourly worker on my phone, I ask “where’s my paycheck?” and get my answer — not a help article.Agents one reads my timecard and pay calendar; one checks whether payroll has run; one writes the reply the way a person would.Hands off to a human the moment the numbers don’t add up.Done when I know the amount, the day, and why.

A mountain of feedback, and the people hiding inside it.

Every product has a pile like this. Interviews, survey comments, feature requests, the thing someone said in a demo — thousands of them, each one a person trying to get their day done.

So I stopped reading one at a time and started listening at scale: embeddings, clustering, a little stubbornness. The pile sorted itself into a handful of people — personas the product had been serving unevenly for years.

The roadmap more or less wrote itself after that. The hard part was believing it.

Synthetic feedback, real technique — the actual pile stays at work.

twenty percent

the rest fills in as you scroll — that’s the whole trick

Twenty‑percent slices.

Big strategic goals are terrifying, and terrified organizations don’t fund things. So I’ve made a habit of building the first twenty percent — a real, clickable, slightly-held-together prototype of the scary long-term thing — and putting it in front of the people holding the budget.

That’s the whole method for agentic systems, too. Build the smallest agent that does one real job. Release it. Watch where it stumbles — the feedback arrives in hours, not quarters. Rebuild. Release. Add the next job. You don’t stop; the thing just gets more capable every week it’s in someone’s hands. Occasionally you stand on a stage and say “watch this.”

Then I teach it.

Whatever I figure out, I try to hand off — as skills and plugins other PMs can run, as sessions in front of a few hundred people, as the annoying colleague who says “here, let me show you” one more time.

The best thing I’ve built at work isn’t a feature. It’s a product org that writes with the machines instead of around them.

here, let me show youhere, let me show youhere, let me show youhere, let me show youhere, let me show youhere, let me show you

Now —about you.

You’ve read this far, which means one of two things: you’re building at the edge of AI and/or HCM — I speak both, fluently, at parties — or you’re procrastinating beautifully. Either way, hi.

If it’s the first, here’s what I’m usually thinking about:

  • The gap between your demo and your invoice — and who’s going to close it.
  • What your users already know about your roadmap — the workers, the managers, the admins — if you listen to them at scale.
  • Whether your PMs can read the code yet. (They can. They just don’t know it.)
  • The one boring workflow your customers would pay to never see again.

I collect hard, unglamorous problems the way some people collect records. If you’ve got one — and you’d rather ship a slice than a deck — let’s talk shop.

Say hidylancolon@outlook.com  ·  LinkedIn ↗

What I do when nobody’s paying me to.

Mostly: learn things I have no business learning yet. Right now that’s robotics and computer vision — small motors, cheap cameras, and the very specific joy of a machine that can see the mug it’s about to knock over.

currently poking at → Hello, world (again)

it can see you. it cannot yet see the mug.

Dylan’s Musings

Notes on AI, product, and whatever’s on the bench this month. Unpolished thinking welcome — each note is labeled seedling, growing, or evergreen by how baked it is.

All musings

  1. Design Systems & Claudeseedling · Aug 17, 2026
  2. Hello, world (again)seedling · Aug 16, 2026