# MidnightDev

> Alex Bouchard — forward-deployed AI lead and founder of MidnightDev. Embedded AI product builds for operators.

This file lists Markdown versions of the site's public pages for agent and LLM readers. Each page is also served as Markdown at its normal URL when the request sends `Accept: text/markdown`. The twins are generated from the rendered pages at build time. Everything in one file: https://midnightdev.dev/llms-full.txt. Updated: 2026-10-22.

## Pages

- [Alex Bouchard — Forward-Deployed AI Lead · MidnightDev | ReadableByAI, Yapword | $400M CRE](https://midnightdev.dev/index.md): Embedded AI product builds for operators. Founder @ MidnightDev. Shipped ReadableByAI and Yapword. $400M closed in commercial real estate.
- [Alex Bouchard | Forward-Deployed AI Lead](https://midnightdev.dev/alex-bouchard.md): Forward-deployed AI engineering evidence across shipped products, evaluation systems, PostHog telemetry, SEO/GEO, distribution, and end-to-end operations.
- [The Build Room — Case File 001 | MidnightDev](https://midnightdev.dev/build-room.md): How one person shipped two AI products end to end: authority boundaries, the Yapoleon voice ledger, a 14-day App Store release, and the telemetry behind it.
- [Résumé — Alex Bouchard | Forward-Deployed AI Lead](https://midnightdev.dev/resume.md): Forward-deployed AI lead. Two consumer AI products live on the App Store, built solo end to end — model behavior, safety gates, reliability, cost controls, native release, and telemetry.
- [Forward-Deployed AI & Growth Systems Consulting — MidnightDev](https://midnightdev.dev/services.md): Forward-deployed product strategy, AI engineering, technical diligence, SEO/GEO, PostHog, and organic growth systems consulting from Alex Bouchard at MidnightDev.
- [Midnight GEO Pro Pack v1.2 — $99 | MidnightDev](https://midnightdev.dev/geo-pack.md): 15 Claude Code skills for GEO/AEO, named staff (eight specialists + a chief), a worked 18-domain evidence pack, ChatGPT probe harness, 12 months of updates, and email support. $99 one-time.
- [Ship Log 2026 — 8,784 commits through October 13 | MidnightDev](https://midnightdev.dev/ship-log.md): A local Git recount through October 13: 8,784 non-merge commits across 103 repository identities, on 258 of 286 days. Project table retains its dated August snapshot.
- [Privacy Policy — MidnightDev](https://midnightdev.dev/privacy.md): Privacy policy for midnightdev.dev — what data is collected via the contact form, how it's used, and your rights.
- [How I put AI agents to work | MidnightDev](https://midnightdev.dev/work/agent-workflows.md): I set the goal. AI agents make the changes. Another agent and automated tests check the work. Only permitted changes can be released automatically.
- [AI Boardroom Forecast Audit — MidnightDev](https://midnightdev.dev/work/boardroom.md): The AI boardroom that killed a $65.9M forecast.
- [Caduceus — MidnightDev](https://midnightdev.dev/work/caduceus.md): Agents cast against measured gaps, not generic roles — and the strongest dissent ships intact.
- [Caliber — MidnightDev](https://midnightdev.dev/work/caliber.md): A team of agents, or a team of agents fitted to you. Concept stage, not a product.
- [Dozen — MidnightDev](https://midnightdev.dev/work/dozen.md): The model writes captions. It does not own the irreversible edges.
- [That's My Best — MidnightDev](https://midnightdev.dev/work/thatsmybest.md): Screenshots in. A playable social memory test out.
- [Yapoleon's Court — MidnightDev](https://midnightdev.dev/work/yapoleonscourt.md): An AI game the LLM can't game.
- [Yapword — MidnightDev](https://midnightdev.dev/work/yapword.md): A daily word game where the character remembers how you played.
- [Writing | MidnightDev](https://midnightdev.dev/writing.md): Notes on shipping AI products: evaluation, calibration, and the parts of the work that cannot be checked automatically.
- [A very crowded Tuesday | MidnightDev](https://midnightdev.dev/writing/a-very-crowded-tuesday.md): We made 520 distinct Git commits in a day. The interesting story is the harness, the agent team, the red checks, and the one human still deciding what matters.
- [Your firewall is your AI policy | MidnightDev](https://midnightdev.dev/writing/ai-licensing-deals-in-status-codes.md): I probed 18 major sites with the user-agent of every AI crawler that matters. Who gets a 200 and who gets a 403 lines up with who signed deals and who is in litigation — and two of my own findings did not survive the data.
- [Benchmarking a generative character when there is nothing to diff against | MidnightDev](https://midnightdev.dev/writing/benchmarking-a-generative-character.md): How a whole-game evaluation harness tested whether Yapword's AI character kept its voice after hidden reasoning was removed from production inference.
- [Calibrating an LLM judge for a game people are trying to beat | MidnightDev](https://midnightdev.dev/writing/calibrating-an-llm-judge.md): A scoring model that grades live players has no reference implementation to check itself against. What I measured on Yapoleon's Court, what broke, and the calibration metric I had to change.
- [Let the agents work | MidnightDev](https://midnightdev.dev/writing/learning-by-building.md): I set the goal. AI agents make the changes. Another agent and automated tests check the work. Only permitted changes can be released automatically.
- [Sibling contrast | MidnightDev](https://midnightdev.dev/writing/sibling-contrast.md): A gray tile, 6,202 words of research, and the contrast check I had never run. The full boards, equations, and rabbit hole into human vision.
- [The model cannot own the money | MidnightDev](https://midnightdev.dev/writing/the-model-cannot-own-the-money.md): Dozen matches brands with small TikTok and Instagram creators. The AI writes the captions. It never accepts a deal, posts, or pays anyone. Here is why, and where the build stands.
- [Unsigned rules | MidnightDev](https://midnightdev.dev/writing/unsigned-rules.md): No line in an agent instruction file says who wrote it. Four ways that goes wrong, measured on one machine: 193 rule files, 41 of 51 pairs drifted, 17 of 22 rules with nothing enforcing them. The record that fixes it, and a script to count your own.

## Optional

Hand-written companion notes for the essays above. They have no HTML page of their own.

- [How I put AI agents to work: evidence](https://midnightdev.dev/writing/learning-by-building-evidence.md): Dated counts behind the agent workflow: auto-merged pull requests, the merge gate before and after (934 merged PRs, 792 with a placeholder review), one day's throughput, the council ledger, and the report format.
- [How I put AI agents to work: technical notes](https://midnightdev.dev/writing/learning-by-building-technical-notes.md): Technical notes for the agent workflow essay.
- [How I put AI agents to work: threat model](https://midnightdev.dev/writing/learning-by-building-threat-model.md): Eight threats to an agent workflow, with what enforces each today, what is policy only, and what is open.
- [Agent failure catalog](https://midnightdev.dev/writing/agent-failure-catalog.md): Ten named agent failures from Alex's records since August 2026, each with the control that caught it and what changed.
- [A very crowded Tuesday: technical notes](https://midnightdev.dev/writing/a-very-crowded-tuesday-technical-notes.md): Method and limits behind the 520-commit day: what the local Git and GitHub evidence shows, and where it stops.