AI should elevate humanity, not the other way around.
We envision small teams of centered, rested people directing large teams of AI colleagues. Shipping at the speed of imagination.
Bridging the almost-there gap.
AI is staggering right now. It writes, it codes, it reasons.
AI is maddening right now. It lives in a thousand open tabs. It forgets between sessions. It works one-on-one when the work is many-to-many.
Bridging the gap between almost-there and actually there takes a substrate. An operating system for AI-native companies.
The substrate.
AI-native organizations need primitives that can answer three questions, 24/7: What's true? What to do? What happened?
A library of what is true.
Structured, linked, queryable. Strategy, brand, customers, products, research — every dimension every agent and human needs, in one canonical place.
A playbook of what to do when.
Triggers, rules, plays. The decisions you've already made, encoded so an agent team can act on them at any hour without paging you. NASA has done this for decades.
A ledger of what happened.
An append-only record of every action, decision, handoff, approval. New agents pick up work by reading it. So do humans. The Navy has done this for centuries.
The near future, as we're living it.
We've been running an AI software factory for most of 2026.
Agents are becoming colleagues. Clear roles, decision authority, just-in-time context. A senior PM agent holds our business in her head and catches inconsistencies a human director wouldn't catch for months.
Execution is moving to AI at factory speed in the domains that compile. Software ships overnight, set up during the day. Creative and design work still takes the time it takes. Project and task work across the whole company starts to run like a factory. Marketing, ops, research, all of it.
The bad news: our old processes and tools are collapsing under the strain. Sprints, tickets, dashboards, inbox-driven coordination, all built for a different tempo, suck now. We often go home feeling like we've been in a car accident, making ten times more decisions per hour than pre-factory.
Where the work is.
Two projects in alpha, ready for first hands.
Proof of concept.
One agent, one job, the first proof of an AI teammate. Raven is your senior product manager. She holds the canonical library — your code, specs, decision threads, all queryable from one place — and uses it to make you look good as a project owner: sharpening what you ship so downstream agents aren't guessing, surfacing the questions only you can answer before you open the day.
She lives on Claude. Setup is from a terminal — book a call if that's not your thing.
Honestly: she's good at the job and rough in real ways. We're hardening her against software factory load and need projects that aren't ours pushing on her. If you've got a real codebase and the patience to file what breaks, that's the trade.
Become an Alexandrian →
Field manual.
The 7 Turn Work Week. The first version of running our company on AI broke us in ten days — concussion-like symptoms by Friday, ten times more decisions than pre-factory. The second version was different: shorter weeks, structured turns, AI doing execution, humans holding judgment. We're rested enough to direct again.
We're writing what we learned, one chapter at a time, while we live it. The book is the field manual for surviving the gap between AI you can almost trust and the substrate that makes it trustworthy. Years, not months.
What we need: people reading along and pushing back. Tell us where the framework fails for your shape of work. Alpha Readers shape the chapters before they're locked.
Become an Alpha Reader →
The arc.
One agent, real load.
Software factories are hitting the context wall first — that's where Raven goes in. We're hardening her against real codebases this year. If you're running multi-agent development, or you're an operator with edge using AI to do what used to take a team, this is where you come in.
The whole company runs on Alexandria.
The same primitives spread out of engineering. Design, marketing, ops — every function that's now operating at factory pace needs the same canonical context, the same encoded judgment, the same record. Alexandria becomes the nervous system. The cockpit ships in here.
Federated.
Company to division to team, all reading from one substrate, all writing to one record. Model-agnostic. The senior PM agent from the top of this page isn't a tool anymore — she's a layer of organizational intelligence that gets sharper every action it takes. Alexandria becomes as foundational as cloud compute.
The Team
Danvers Fleury
Architect of Alexandria
Started at Monster.com weeks after its IPO, fell in love with the problem of how teams scale and never stopped working it. 25 years across startups, turnarounds, and organizational design.
Jess Martin
Builder of Alexandria
Recruited out of school to build The Sims. Two decades building knowledge representation and real-time collaboration systems across multiple startups.
Come build with us.
If you've read this far, you're probably one of two kinds of builder.
The first runs a software factory and has hit the wall of being the coordination layer. The second is a business operator with an edge — domain knowledge, market access, a strategic read — who wants to build an AI team, go forth, and prosper.
If that's you, become an Alexandrian. Forty-five minutes on a video call with one or both of us. The Alexandrian Agreement is here, so you know exactly how we work before you book.
If you want to follow without installing yet:
- Read along. Labnotes at sociotechnica.org, weekly.
- Read the book as it's written. Become an Alpha Reader.
- Watch the substrate evolve. Alexandria on GitHub.
Stay in the loop.
Labnotes weekly. Book chapters as they ship. Product updates when they matter. One list, one email.