10 May 2026
Hand-coding is the slow path — agentic orchestration workflow
The last time I wrote code by hand was over a year ago.
Not because I can't. I still read code fluently and understand exactly what's happening under the hood.
But hand-coding is now the slow path.
Here's what my workflow actually looks like today:
I architect first. Draw out the system, define the boundaries, think through failure modes. That part is still 100% human.
Then I hand it to an agent — Claude Code, usually — with the spec and let it implement. Meanwhile I spin up another agent on a different task. Sometimes I have 10 running at once.
Then I gate the output. Review it against the architecture. Manually test the critical paths. Catch the drift before it compounds.
The biggest lesson from 8 months of building this way: AI doesn't replace engineering judgment. It amplifies it. Feed it a bad architecture and it builds the wrong thing very fast. Feed it a clear spec with well-defined boundaries and it ships what would've taken a team days — in hours.
The skill that matters now isn't typing speed. It's orchestration. Knowing when to let the agent run, when to pull it back, and how to design systems that are worth building in the first place.
Engineers who figure that out early are going to have a significant advantage.
The ones waiting for AI to "prove itself" are already behind.