Specify, do not type
Spend the effort on the prompt, the context and the acceptance criteria. A vague instruction produces a plausible diff you then have to debug, which is slower than having written it yourself.
Zero Manual Coding
What the people here have achieved by directing an AI agent instead of typing every line themselves — in their own words, with what it saved them, and with enough detail that you can do the same thing tomorrow.
Your work email is enough to read and post. A Cursor API key is optional, and only adds your own agent cost and delivery metrics.
The wall
What people here have achieved
Every card was written by the person who did the work. Some are features, some are migrations nobody noticed, and some are an afternoon of tedium that turned into twenty minutes. All of them count.
Nothing on the wall yet
Nobody has shared an achievement yet. The first one is the hardest, and it sets the shape everybody else copies — so it may as well be yours.
The practice
What Zero Manual Coding actually means
It is not a ban on typing. It is a default: before writing a change by hand, try to get there by describing it well enough that an agent can. The point of the exercise is that a task you can specify precisely is a task you understand, and the specification outlives the diff.
Zero is a direction rather than a score. Some work will always be quicker by hand, and nobody is measured on how little they typed. What is worth measuring is whether the same problem gets solved faster the second time somebody meets it — which only happens if the first person wrote down how they solved it.
Review like it is a colleague
Agent output goes through the same review, tests and standards as anything else. Zero Manual Coding changes who writes the first draft, not who is accountable for what merges.
Turn a good prompt into a shared one
A prompt that worked twice belongs in the shared commands and rules library, not in one person’s history. That is the difference between an individual getting faster and a team getting faster.
Write it up, honestly
Every card on the wall says what somebody achieved, how the agent helped, and what it saved against doing it by hand. Modest numbers people believe are worth more than impressive ones they do not.
Take part
Three steps, about five minutes
- 1
Pick something you achieved
Anything you got done with an agent. It does not need to be a feature: a migration nobody noticed, a flaky test finally fixed, an afternoon of tedium compressed into twenty minutes.
- 2
Say how you did it
How you directed the agent, and what it did rather than you. Add your own estimate of the effort and the days saved. Two fields are required and it takes about five minutes.
- 3
Share it to the wall
It appears here, it is searchable by tag and team, and it counts towards the totals above. Sign in with your work email; the Cursor API key is optional.
What did you get done this week?
Sign in with your work email. The wall, the shared commands and rules, the resources and the AI news are all open to you from the first click; add a Cursor API key whenever you want your own agent metrics alongside them.