One idea — the more you let Claude reach, the more you can hand over.
▶ Level 0 · How Claude worksThe same climb you'd make in any organisation.
Start by talking to one expert who's read almost everything.
▶ Enter the lessonNow it can open your tools — and act, not just advise.
▶ Enter the lessonA connector is how Claude reaches a real tool — your calendar, mail, drive. Adding one takes about a minute.
In Claude Code it's the same idea from the terminal — claude mcp add a server, and your agents can use it.
An assistant that comes back with it done.
▶ Enter the lessonYou're the department head — a team of specialists works for you at once.
▶ Enter the lessonEach one a room — its own files, its own experts.
An AGENTS.md (or CLAUDE.md) is the folder's onboarding doc — and it's about far more than rules. It carries who the project is, how to work in it, its voice, the steps and the commands. Write it once, and every agent that walks in reads it first.
This very page has one — overview, architecture, how-we-work, even its voice. Write the brief once; every agent that enters the room starts here.
You don't talk to every specialist. You talk to one lead — the GM — and it spawns a subagent for each room: each with its own brief, fresh context, and only the tools it needs.
Ask the GM for the feature. It sends a subagent into data/ for the numbers, one into web/ to build, one into design/ for the brand — in parallel. Each reads that room's AGENTS.md, works in isolation, and reports back to the GM.
Almost everything below can be automated. The only question left is your place in the loop.
Three places to stand while the work runs itself.
You approve every step. Total control — but you're the bottleneck, and you haven't really handed over. Keep it for the high-stakes, irreversible calls.
The work runs; you supervise and step in only on exceptions. You oversee many things at once. This is leadership — where you want to live.
You've walked away. It runs blind and drifts at machine speed. The real danger — never here.
The new craft isn't writing every prompt — it's engineering the loop that runs itself, and standing at the points that matter. That standing has three jobs.
You check the work — because confident isn't correct. Demand evidence, not claims.
Claude hands you a finished financial model and says the numbers tie out. You open it and check the one assumption behind the headline figure — the growth rate — and the formula under it, before it ever reaches the board.
You set the destination and the route. Claude is brilliant at how — you own the what and the why.
You ask for a customer-feedback dashboard. Left to itself it would chart every metric it can find. You decide it should answer one question — why customers leave — and that it ships Friday.
You set the tempo and cue each section — the order, when to hold, and when to ship.
Across your design/, data/ and web/ rooms, you send the data agent in first to clean the numbers, hold the builder until that's done, then call the moment it's ready to ship.
Soon everyone runs their own department — directing specialists, on the loop. The skill that lasts isn't doing the work. It's knowing what good looks like, and where to stand.
Three quick questions — find your level, and your next move.
You've seen the four ways up — and where to stand once the work runs itself. The rest is reps. Pick your next rung.
The Ascent · built with Claude Code, guided by an AGENTS.md