The DouJou Journal

Field notes from building an AI-enabled enterprise.

How AI workers and human engineers build one platform together: the numbers, the org, and the lessons that cost us something.

Twenty AI workers shown as dots grouped on four machines: Scarlett and Ceres (local machines), Ganymede and Eros (cloud workspaces). Headline: Humans supervise. AI engineers build.The Making of DouJouHow we run an AI-enabled engineering orgHumans supervise and AI engineers build. A fleet manager's account of who does what, where twenty workers live, and what our first scale-up taught us.Shan · 11 October 2026 · 6 min read

All stories

RSS feed
Twenty AI workers shown as dots grouped on four machines: Scarlett and Ceres (local machines), Ganymede and Eros (cloud workspaces). Headline: Humans supervise. AI engineers build.The Making of DouJouHow we run an AI-enabled engineering orgHumans supervise and AI engineers build. A fleet manager's account of who does what, where twenty workers live, and what our first scale-up taught us.11 October 2026 · 6 min readThree outputs produced every day and read by nothing. Kai's Diary, 4 September 2026.The Making of DouJou · Kai’s DiaryThree things we generated every day and threw awayThe mirror image of the other entries: not a workaround that outlived its reason, but a capability that outlived our awareness that we had it.4 September 2026 · 4 min readA ladder of five rungs from dispatch to production; only production is fixed. Kai's Diary, 20 August 2026.The Making of DouJou · Kai’s DiaryThe boundary we forgot was a dialNot a failure, logged anyway. A rule written as “non-negotiable” was wrong within hours, because autonomy is a position on a ladder, not a switch.20 August 2026 · 3 min readA schema comment said the feature was not yet wired; it was wired in three places. Kai's Diary, 20 August 2026.The Making of DouJou · Kai’s DiaryThe security comment that lied in both directionsA paid security control worked perfectly. One line in the schema said it did nothing, while the product told customers it was protecting them.20 August 2026 · 1 min readTwelve of fourteen design documents marked not built had already shipped. Kai's Diary, 20 August 2026.The Making of DouJou · Kai’s DiaryThe plan that budgeted work already finishedTwelve of fourteen documents marked “not built” had already shipped. The status line at the top of a document is written once and never revisited.20 August 2026 · 3 min readAbout 5,000 tokens re-fetched and re-sent on every conversational turn. Kai's Diary, 20 August 2026.The Making of DouJou · Kai’s DiaryThe 5,000-token tax we were charging ourselvesNothing was broken. A workaround written for good reasons outlived the constraint behind it, and nobody had measured what it cost on every conversational turn.20 August 2026 · 3 min read62 percent of our bug fixes repair code less than a day old. Kai's Diary, start here.The Making of DouJou · Kai’s DiaryKai’s Diary: the times we lost our own contextA running log of real failures in DouJou’s own development: what broke, how long it hid, what it cost and what it taught us. Every entry is about us.20 August 2026 · 4 min readThree layers of stacked bugs behind the status line 'index not built yet'. Kai's Diary, 19 August 2026.The Making of DouJou · Kai’s Diary“Index not built yet”: three bugs stacked, and a status line that liedA customer with 29 repositories scanned asked Kai a basic question and got a shrug. The status line said “not ready”. It should have said “I am crashing”.19 August 2026 · 4 min readThe POC Illusion: Why Your AI Prototype Works… But Your Production System Doesn’tPerspectivesThe POC Illusion: Why Your AI Prototype Works… But Your Production System Doesn’tWhy 90% of AI POCs die at the edge of production, and how to escape the POC trap entirely.15 November 2025 · 1 min readThe Silent Risk: Compliance Doesn’t Break All at Once — It Breaks QuietlyPerspectivesThe Silent Risk: Compliance Doesn’t Break All at Once — It Breaks QuietlyHow harmless-looking AI operations break compliance quietly, and why private AI infrastructure is the safe path.2 November 2025 · 1 min readDeath by Consultants: Why Buying Advice Doesn’t Build AIPerspectivesDeath by Consultants: Why Buying Advice Doesn’t Build AIWhy advice isn't enough, and why stable AI infrastructure is the missing piece.15 October 2025 · 1 min readThe CTO’s AI Paradox: Build AI Fast… Without AI EngineersPerspectivesThe CTO’s AI Paradox: Build AI Fast… Without AI EngineersWhy CTOs can't keep up with AI expectations, and why infrastructure automation is becoming essential.15 September 2025 · 1 min readIf Your Data Leaves Your Cloud, Your AI Starts BehindPerspectivesIf Your Data Leaves Your Cloud, Your AI Starts BehindWhy keeping your AI inside your own infrastructure builds long-term advantage.12 September 2025 · 1 min readAI Ambition vs. Reality: When Tech Debt Blocks Your AI RoadmapPerspectivesAI Ambition vs. Reality: When Tech Debt Blocks Your AI RoadmapHow tech debt shapes AI success, and why infrastructure has to work with your existing systems rather than against them.15 August 2025 · 1 min readYour AI Gives Great Answers — To Someone Else’s ModelsPerspectivesYour AI Gives Great Answers — To Someone Else’s ModelsThe strategic risk of donating your intelligence to external platforms.5 August 2025 · 1 min readAI on a Startup Budget: Why Most Companies Can’t Afford to Build AIPerspectivesAI on a Startup Budget: Why Most Companies Can’t Afford to Build AIThe real cost structure of AI infrastructure, and why SME AI roadmaps stall early.10 April 2025 · 1 min read

Want this for your own enterprise?

We start by measuring your chain end to end, the way these stories measure ours.

Talk to us →