The Middle Step Nobody Wants to Do

Most of my personal projects generate more thinking than writing. I'll work through something, land on a real insight, and want to turn it into a LinkedIn post. Getting the raw thought out is fast — it's everything after that stalls me. Structuring it, formatting it, making it read like something worth publishing takes real time. And handing it to a generic AI tool doesn't actually save that time, because I end up spending just as long re-explaining my tone and rewriting whatever it hands back.

That's what I built for originally: a skill that takes an idea in whatever shape it shows up — a rambling voice memo, a messy note, a half-formed thought — and does the actual work of organizing it, structuring it, formatting it, and writing it back out in my voice instead of a generic one.

Once it worked for me, the same need kept showing up in other people's work. I built a version into a Slack-to-Notion agent for a marketing executive, who needed her raw voice memos turned into organized ideas that matched her team's format and her tone. Two other executives at similar companies wanted the same thing for their own notes and voice memos. Two individual contributors at other organizations wanted it for a simpler reason — they knew exactly what they wanted to say, they just didn't want to be the one formatting it. Same build, six different people using it.

I built a skill trained on a sample of a person's actual writing — real posts, emails, messages — so it learns their cadence, vocabulary, and structure straight from examples, not from a description of their tone.

The workflow after that is simple: brain-dump in natural language, whatever form is easiest — a voice memo, a rushed note, a rambling paragraph — and the skill picks it apart, organizes it, structures it, formats it for wherever it needs to go, and writes it back in that person's voice. Making it sound like them is the last step. Figuring out what they actually meant, deciding what matters, and formatting it correctly is the heavier lift.

It's less like building something new every time and more like tailoring a suit. The pattern — the process for training and wiring up the skill — is the same every time. What changes is the tailoring: whose writing it's trained on, how they actually think and structure things, and which tools it needs to plug into. That's also why generic, off-the-shelf versions of this never quite fit: they're sized for an average voice, not an actual one.

Built it for myself, then deployed the same use case for 5 others — a mix of executives, leadership, and ICs.

For the marketing executive's version, that meant cutting per-idea processing time from 30-35 minutes down to about 3 minutes — reclaiming more than 4 hours a week she'd been spending sitting down to structure ideas by hand.

Across the board, gets a first draft roughly 70% of the way there on the first pass — organized, formatted, and already in the right voice.

Removes the organizing and structuring work from the person's plate, leaving only the final edit pass.

This has turned into a repeatable build — not that every detail repeats, but the process behind it does. Once we have found a process that actually solves something, then we can apply that same process to similar problems elsewhere. The details change from person to person and industry to industry — the voice, the tools, the format — but the structure underneath is almost always the same.