Application: Founding Engineer, Jack & Jill.

I am choosing to stop contracting

I left Pfizer to contract by choice. It pays well and it is the wrong shape for what I want next: a decade on one thing that matters, with people faster than me. Your JD says the work is connecting the candidate's agent to the company's agent. I have shipped four MCP servers that do exactly that kind of connecting, so the sentence is not abstract to me.

I am 42, in Folkestone, an hour from King's Cross. Shoreditch in person, five days a week: yes.

Two things I checked before applying. The £360k+ figure is your published floor for every engineer, not a special seat, so I am not asking for one; the conversation I want is about the cash-to-equity split and the grant. And the founders have done this before: Matt built Omnipresent to a $600m valuation and an exit to Deel; Saaras took Kular from zero to $3m ARR in a year through YC. I would rather work for people who have carried a P&L than people who have read about one.

Product notes from a candidate who builds agents

Jack held an hour's conversation, kept my profile straight, cleared a flag on my brief unasked, and gave a useful pass on my CV. These are the things I would want to own.

1. Groundedness on facts about the company itself. Jack told me the stack was Python on AWS and Azure with Kubernetes and Terraform, and that Sequoia led the round. The JD says Next.js, FastAPI, Postgres and Creandum. A candidate who repeats Jack in an interview embarrasses both of them. The fix is a versioned fact store for first-party claims, and an eval that replays real conversations and fails the build on a contradiction.
2. Conversation memory across turns. Every reply re-pitched the same two contract leads, including the one after I said I wanted the founding role. A turn-level record of what has been offered and declined would make Jack sound like he is listening.
3. The handoff between Jack and Jill. Jack is right not to back-channel. But when a candidate says "I want to work for you", a typed, consented intent event that Jill can read should exist. After I applied, Jack told me the application was "outside the web-sourced tracker": he cannot see his own company's pipeline. That is an agent-to-agent protocol problem, and it is the kind I have shipped.
4. Observability you can argue from. All three need a trace: which retrieval fed which claim, which memory was read, what the handoff carried. I would build this first.

The numbers, with sources