Elliot Little
Operating Report · London · Est. MMXXVI31 JUL 2026
operational · interviewing

Shipping got easy.
Judgement didn't.

I'm Elliot. Builder-operator in London, 4x founding hire. I shipped a production AI tutor the UK government selected for its national programme on safe AI tutoring, and I build the layer that decides whether output is good enough to ship. This site is one of those products: every number on it is computed from real activity, not claimed.

Elliot Little
live since 22 Julagent identity verifiedcookies 0analytics 0source public

the through-line

Models write plausible things now. Checking them didn't get easier, and that is where most of my work has gone. At Zero Gravity I built a tutor that coaches a student to the answer and will not hand it over, with an evaluator grading every session before a student saw the mark. ward separates a real safeguarding disclosure from a child having a bad day, which is the distinction keyword filters get wrong. crux turns the same question on me: it logs what I rejected in my own sessions, so I can tell whether I am still thinking or just approving.

The work, and what each piece decides.

what I am doing now

  • Looking for my next senior product or AI role. Between roles and using every day of the gap to ship. Email works: elliotjlittle@gmail.com.
  • Built this site over two days in public. v1 was a portfolio; what it is now, and the agents that keep it current, are on /loops.
  • Running argus daily: a private fleet of agents that reads the AI news each morning and writes me a brief before I sit down.
  • boulot, my open-source career-ops system, got its public release this month. My partner and my sister both use it.

telemetry

commits · 7d120Sat: 0 commitsSun: 18 commitsMon: 7 commitsTue: 20 commitsWed: 7 commitsThu: 23 commitsFri: 45 commits
repos touched · 7d5public repos
agent runs · all time8scheduled + on-demand
agent tokens · all time08 runs metered

sources: GitHub API, data/spend.json, git history. rebuilt 2026-07-31 09:23 UTC @ 9132c56. the count on /loops is lower because it is frozen at the agent's last run.

fit engine

Fit Engine — Interactiveruns in your browser
static site, zero backend: nothing you type leaves your browser. the composed briefing only travels to the agent you choose. hosted inference is on the roadmap.

the report

01

built

A production AI tutor the UK government selected for its national programme on safe AI tutoring, a safeguarding layer for apps serving under-18s, and the agent tooling underneath both.

02

writing

Essays on shipping AI to users who can't absorb a wrong answer, on trust and adoption, and on where responsible-AI-by-checklist breaks.

03

loops

The agents that maintain this site: cadence, cost, the human approval gate, the stopping rule on each, and the shipping log the inner loop last wrote.

04

changelog

Every change is a commit, badged by who made it: me or the agent. The receipt trail for every number above.