Making yourself redundant
Language models can absorb and propagate team culture. Turns out that’s more interesting than asking them to write tests.
Onboarding is indoctrination. Not pejoratively — you’re exposing a human to social contracts until they stop guessing how things are done here. Takes weeks. Sometimes months.
I’ve been encoding that process into skills for a language model. Not the trust built over lunch. The mechanical stuff. Product processes, engineering conventions, the things that live in a wiki nobody reads or worse, someone’s head.
The bottleneck is you
New hire needs environment setup. You walk them through it. Again. Fourth time this quarter. Same liturgy, different acolyte.
I built a skill for this. Checks tools are installed, walks through the codebase, makes recommendations based on what’s actually running — not what we assumed everyone uses. The new starter isn’t blocked on your calendar. You’re not reciting scripture.
Product management as YAML
Two skills I’m publishing handle work I’d typically do in a fractional CTO role: sizing and forecasting.
The sizing skill encodes our estimation criteria. Fibonacci scale, reference examples, decomposition rules. Reads an issue from Linear via MCP, proposes a number. Parent issues carry zero points; work lives on leaves. It catches the orphaned “write tests for X” tickets that shouldn’t exist.
The forecasting skill sums points, calculates dates, offers to update Linear. Handles team-size multipliers for the sub-linear scaling of adding people.
This used to consume time disproportionate to its complexity. Now it’s minutes.
Full lifecycle
I have skills for starting work on a ticket. Creating branches with naming conventions. Running test suites before PRs. Checking commit signatures. Maintaining linear history.
Each encodes behaviours an experienced person carries implicitly. The conventions. The checks. The things you piece together from observation and code review comments.
The bit most people aren’t seeing
Conversations I’m having focus on the obvious: generate a file, scaffold a component. Useful. Limited.
The interesting application is encoding the social contracts of collaboration. How work moves through a system. How we size it. How we ship it. These have historically been transmitted through slow, lossy human observation.
A language model with the right skills doesn’t replace that. It makes implicit things explicit. Information on demand, in context.
That’s the transformation. Not generating code. Propagating culture.