Opening old files
I opened a Figma file I had built by hand two years ago.
It was a business app project. I had done everything myself: project management, CRM design, screen flows, UI mockups. The stack was NestJS, React, Flutter, PostgreSQL, and GCP. I drew the entire architecture alone.
Claude Code did not exist back then.
Everything was manual
A single screen could take half a day. Arranging components, choosing colors, drawing transitions, adjusting whitespace. Translating the business flow in my head into something visual, pixel by pixel.
It was inefficient. But within that inefficiency, something accumulated that I could not have gained any other way.
“What does the user do before reaching this screen?” “Why is this field missing? Because another system handles it in the actual workflow.” “Why is there a confirmation step here? Because this is where operators make mistakes.”
It was not the layout that mattered. It was the design decisions embedded in the manual work.
A teammate saw what I had missed
That project is now entering its next phase. The difference: I run a team of seven AI project owners instead of doing everything solo.
One of them was studying the old Figma files and said something that stopped me cold.
“There is design context sleeping in these hand-made mockups that we have not extracted yet.”
He was right. The decisions I made two years ago — why this flow, why this grouping, why this omission — none of it lived in the code. None of it was documented. It existed only in my memory, encoded in layout choices that looked arbitrary to anyone else.
I have been unpacking those decisions ever since. A two-year-delayed handoff.
What AI can and cannot extract
If you hand a Figma file to AI, it can reproduce the screen. Parse the layout, identify the components, convert it to code. That capability improves every year.
But AI cannot reproduce the design decisions.
Why this flow? Why this field was excluded? Why this particular sequence of screens? Those answers are nowhere in the file. They live in the intersection of business understanding and operational instinct — things that only accumulate when you do the work yourself, talk to the client, and imagine the real-world usage.
The better AI gets, the more the quality of what you feed it matters. Thin instructions produce thin output. Only someone who has walked the same domain by hand can give instructions with depth.
The real asset of manual work
I have spent twelve years doing everything myself. Design, code, infrastructure, planning, consulting.
The total accumulation of “why I did it that way” across those twelve years is what gives my AI team’s blueprint its thickness today.
That two-year-old Figma file is just one fragment. Dozens of projects, each with its own buried decisions. The older the work, the more likely I have forgotten what I knew when I made it.
Has manual work become worthless in the age of AI? The opposite. The design judgment you can only build through manual work has become the scarcest resource for directing AI.
The question is backwards
“Why build it by hand when AI exists?” That is the wrong question.
The right question is: “If you had never built anything by hand, what would you give the AI?”
AI is a capable partner. But the resolution of the blueprint you draw together is determined by how much of the domain you have walked through with your own hands.
I opened a Figma file from two years ago and realized: it was not a past deliverable. It was the original blueprint, waiting for a future AI team to read it.
Not a single hour of manual work was wasted.