Krystian Gadomski
Lead developer, architect, mentor. Still building.
Agents write code faster than I can review it. This channel is what I do about that.
For engineers and solo founders who build software with AI agents.
How I see it
The slow part is me now.
A lot of people think AI agents turned code review into a formality. The agent writes the code, the tests go green, you merge.
In one of my videos, I ran four agents in one repository at once. A viewer wrote under it: “Past that my review queue becomes the bottleneck.” I had the same problem. I caught myself reading their code too fast.
Agents get better every month. So every month I decide again how much of their code I still need to read, and how I check the rest. That's what my videos are about.
What you get
Each video tries one idea on real code.
My own numbers.
When I say something got faster, I show you how much, and compared with what. If I only tried it once, I say so.
The mistake comes first.
I show what went wrong before I show the fix. It's usually the mistake you'd make too.
It works on the tool you have.
Most of what I show is about how you work, not which tool you pay for. My first big jump happened on Copilot, before I switched tools.
My numbers
My best sprint used to be 28 story points. Then I hit 48.
These numbers are mine alone, from my sprints at work. Without AI, my velocity was around 20 story points. Then Copilot helped me in the editor. Later I changed how I work. I write a spec, split the work between agents, and review it in a separate session.
The first big jump came on the same Copilot, just used differently. Since then I've had sprints of 48, 57 and 68 points.
Story points are estimates, so they're not a perfect measure. But I compare myself with myself, on the same project, estimated the same way.
“The 3X development leverage is real without jeopardising our products quality and security.”
Episodes
Start here.
How many agents can one person review?
How I ship 3x more with parallel AI agents (and still review the code) on YouTube
Four agents worked in one repository at once, with no merge conflicts. Then I reviewed their code without opening an editor.
What I'm building
Still building, in public.
I build and review code at work and on my own apps.
When you build alone, nobody reviews your agents' code but you. On my apps, that's me.
I build them on camera because they're the only code I can show you in full.
Videos about building software with AI agents.
This site. AI agents built it from a spec I wrote.
About
Two decades in, still learning in public.
Hi, I'm Krystian.
At NEUCA, a pharmaceutical distributor, I designed their apteline.pl pharmacy shop. I defended it in front of their architecture and security departments. I also fixed their old order system instead of replacing it. It's still running.
I stayed long enough to live with my own design decisions. I know what a shortcut costs a year later.
After that I led development at Falck, on emergency services software. At Nicolaus Copernicus University I shared my experience with students in team programming labs. Some of the juniors I mentored have landed jobs.
Today AI agents write most of my code. I share what I learn, including what didn't work.
Work with me
Work with me.
I take on work through DevoJAM, my one-person company. Write to me there.
Questions
Fair questions.
Story points are just estimates. Why should I trust your numbers?
They're estimates, so they're not perfect. But I compare myself with myself, on the same project. The Product Owner I worked with confirmed the gain in public. It's my experience, not a study.
Do you read every line the agents write?
Less than I used to. Agents get better every month, so I keep checking how much I still need to read. But a separate session reviews the code, so the agent that wrote it never checks its own work.
My company only allows Copilot. Is this for me?
Yes. My first big jump happened on Copilot, before I switched tools.
Your demos run on your own app. My codebase is 15 years old.
I know. I can't show client code, so I show mine. At NEUCA I fixed an old order system, and it's still running. But I haven't used agents on a client's old code on camera, so I won't say I have.
Are you still current?
I build and review code at work and on my own apps. You'll see my mistakes in the videos.
Is this about building AI agents?
No. It's about using AI coding agents to build software. I don't cover AI news either.