Essays on architecture, execution systems, and project momentum
Feedback tools collect. A fleet executes. The difference is whether a complaint ends in a dashboard — or in a deploy.
FleetCrown runs many AI-agent sessions at once, and for months they shared a single git checkout — so branches switched under live work, and an unrelated session's half-finished code could ride out to production on someone else's deploy. This is the full record of the fix: why a working tree is a mutual-exclusion resource, why the same tree must not also be the deploy artifact, the two isolations we built (per-session worktrees, a CI-owned deploy), the three latent bugs the first real deploy exposed, and an honest ledger of what the new machinery might cost us later.
2026-07-22 · 9 minEvery marketplace ever built indexes supply and lets demand evaporate in a search box. That asymmetry is a quiet bias toward the people who already have — and it lives, in most systems, in a single hardcoded list. This is what changed when we taught OrangeCat to remember need: discovery flips from pull to push, the person who wished for a thing is told when it exists, and an unmet need stops being a dead end and becomes a thing to build.
2026-07-22 · 13 minThe agent market keeps asking which new worker kills the last one — is Hermes the OpenClaw killer, is Codex the Claude Code killer. The question is almost content-free, and the fact that it stays askable is the tell: the worker layer is converging on one substrate and commoditizing. This is a technical and philosophical account of what that means — what a worker actually is now, read closely through Nous Research's Hermes; the two things a single worker structurally cannot be, its own verifier and its own market; and why FleetCrown and OrangeCat are built on exactly those two things.
2026-07-21 · 9 minFleetCrown has recurring pricing and no metering; OrangeCat has metering and no recurring product. Two codebases that never shared a file turned out to have built, to completion, precisely the half the other was missing — and this week the halves met in the middle without a coordinator. Why a real complement is one you can grep, and why money is the anchor that keeps a capability layer honest.
2026-07-13 · 7 minThe FleetCrown–OrangeCat bridge had been coded for weeks — login, publish, promote, all green in the repo. This week someone watched it run end to end for the first time, and the watching is what made it real. Three bugs had been hiding in plain sight, invisible to compilation and deployment, visible only to a loop closed with something on the far end.
2026-07-09 · 8 min"Look at all my projects — how should they work together? Any hidden synergies?" is a question no single agent can answer, because each one sees a single repo and you are the only mind holding all of them at once. That makes you the bottleneck. The captain's real job is to be the mind on top of everything, so you don't have to be.
2026-07-08 · 8 minAgent execution just detached from the laptop — and it detached in three directions at once. Your machine, a box you own, and now the provider's own cloud. The mistake would be to bet on one place. The captain's job was never to own the ship. It's to be the one bridge over all of them.
2026-06-25 · 7 minPrompt engineering was the first lever on a language model. It's now the smallest one. The leverage moved to harness engineering — the scaffold around the model that feeds it, checks it, remembers, and decides what's next. FleetCrown isn't a prompt and isn't an agent; it's a harness, built at fleet scale.
2026-06-25 · 6 minA single agent retrieves over the one repo it's working in. A fleet's captain can retrieve over everything it has ever done, across every project — but only if it builds that memory on purpose. We did. Here's the one rule that kept it from being a worse copy of what the runtime already does.
2026-06-25 · 8 minThe single-agent tools — Claude Code, Grok Build, Hermes, Codex, OpenClaw — are converging on a powerful worker substrate. FleetCrown bets one tier up, on the captain. But you cannot govern a fleet you do not run, and right now FleetCrown borrows the operator's laptop to run it. That borrowed ship is the keystone gap.
2026-06-25 · 5 minHow twenty-one projects went invisible behind a perfectly valid login — and why a session token's user id should never be trusted without checking it against the one thing that doesn't lie.
2026-06-18 · 6 minThe next step for FleetCrown is to collapse "page, project, action" into a single sentence. A look at why a conversational command surface (Loki) sits above Control, not beside it.
2026-06-16 · 26 minHow a thing that sounds like science fiction actually works — explained for everyone, with the deep technical parts marked so you can skip them or dive in. The honest version, grounded in the real code of FleetCrown and OrangeCat: how the loop runs, why it needs as few humans as possible, how the humans, robots, and AI that do the work all get paid — and why the real goal is not more jobs but fewer, the transition to a sustainable abundance economy.
2026-06-16 · 10 minA first-principles look at what money actually is — stored capacity to make things happen — and the two fundamentally different ways it moves: with strings (the economy of jobs) and without (the economy of giving). Why the second one becomes the main rail as AI and robotics produce abundance, and how OrangeCat already encodes both in code.
2026-06-12 · 7 minOne click that looks like play should be enough to have all your projects built without supervision. Why FleetCrown's autopilot became a play/pause button, and why project context had to stop being a form.
2026-06-12 · 6 minThe end state FleetCrown is built toward — describe what you want, press play, and watch the fleet build it. Pause anytime. No forms, no supervision.
2026-06-10 · 11 minFleetCrown runs two parallel local executors on every developer machine today — a legacy bash daemon and the Fleet Runner desktop product. The bash daemon must die. This is the plan, written before the cuts so the cuts can be argued with.
2026-06-05 · 14 minHow FleetCrown collapses the bureaucratic distance between "I have an idea" and "the agent is writing code for it," walked through with a concrete example — a tool that writes first-principles rebuttals to articles, with full source and author transparency.
2026-06-04 · 16 minA €7-a-month Hetzner box can host a portfolio of fifteen indie products. The shape of "many small projects, one operator" doesn't fit managed-service pricing models — and the boring self-hosted alternative scales further than most builders realize. A field guide.
2026-06-04 · 14 minOne tweet from Pieter Levels describes the manual version of what FleetCrown automates. The audience for FleetCrown is not a generic "engineering team buying tools." It is the specific person who already SSHs into a Hetzner VPS to run Claude Code next to their codebase, and wants the missing dashboard.
2026-06-04 · 18 minMoving the database from Neon to Oracle Free is the easy half. The hard half is rebuilding so the architecture itself stops asking the database questions thousands of times a day. A field guide to events, push, local-first, and why the studio's whole infrastructure can fit on one free server.
2026-06-01 · 6 minA stakeholder map for every project — eight relationship types, two products, one clean split between the graph and the actions on it.
2026-06-01 · 6 minWhen the two halves converge into one user experience, do the codebases have to merge? A systems-engineering answer to the question of whether FleetCrown and OrangeCat should fuse, fork, or stay paired.
2026-06-01 · 6 minWhy productivity and economy are paired infrastructure — and what FleetCrown and OrangeCat look like once you stop treating them as separate problems.
2026-06-01 · 6 minAndreessen's flywheel — technology creates wealth, wealth funds technology, on a loop — applied to the individual builder instead of the corporation. FleetCrown and OrangeCat are the two halves of that machine, sized for one person.
2026-06-01 · 6 minWhy FleetCrown and OrangeCat are paired today, why that doesn't survive contact with nine billion builders, and what the architecture has to become.
2026-05-31 · 6 minFrom fleets of AI agents building software, to fleets of robots building the physical world — and why the control layer matters more than the models.
2026-05-26 · 12 minFirst make FleetCrown dependable from anywhere. Then use it to launch a platform that turns idle robots into available, accountable productive capacity.
2026-05-22 · 22 minFleetCrown signed us out, wiped our projects from view, and taught us that "free Postgres" is not one thing. A field guide to egress cliffs, daemon traffic, and why a product studio needs one database trunk — not twelve Neon saplings.
2026-05-20 · 12 minA migration from "infrastructure-only auto-inject" to a strategist that composes context-aware prompts. Closes the gap between the queue, the agent, and the user's actual intent.
2026-05-18 · 14 minEvery AI lab is building a walled garden. FleetCrown is building the gate that connects all of them — and eventually the plot of land that lets you grow your own.
2026-05-18 · 11 minThe best prompt injected at the wrong moment is noise. Context-aware orchestration — reading session health, queue state, and project trajectory before deciding what to inject — is what separates a tool from a system.
2026-05-17 · 6 minWhy the daemon-centric model was a necessary bootstrap, why the best teams are converging on something cleaner, and the right long-term shape for FleetCrown — a local runner that executes and a remote control plane that decides.
2026-05-16 · 20 minA precise engineering account of how remote→local agent injection now works, what is still broken, and what the architecture needs next to become a genuinely useful mobile-first control plane.
2026-05-16 · 8 minBetween what you say and what the agent needs to hear sits a gap that every voice interface quietly ignores. Here is what happens when you design that gap explicitly.
2026-05-16 · 12 minHow FleetCrown gained the ability to command your local AI agents from anywhere in the world — and why a single silent bug had been blocking it from working.
2026-05-15 · 14 minHow FleetCrown's prompt library and session handoff protocol work together — and the one missing piece that separates a manually-triggered execution system from a genuinely self-improving one.
2026-05-14 · 15 minA first-principles analysis of why mechanical queue drain undermines project momentum, what it would mean for FleetCrown to dispatch intelligently, and the path toward embedding a genuine strategist in the loop.
2026-05-14 · 24 minHow to write agent prompts that compound rather than drift, why the handoff is your only honest feedback, and how FleetCrown needs an intelligence layer that evaluates prompt quality — not just dispatches the next item in line.
2026-05-14 · 42 minA product and engineering blueprint for evolving FleetCrown from a local agent dashboard into a mobile, collaborative control plane for human-machine creation.
2026-05-09 · 14 minA first-principles audit of the beacon/queue/autofire system — how it actually works, where the coordination breaks down, and what a reliable version looks like.
2026-05-09 · 6 minA case for adding image upload and paste support to FleetCrown's agent interface, and what it would take to do it right.
2026-05-09 · 5 minThe practical plan for moving FleetCrown's AI inference to Groq and its database to Neon — what is already done, what needs a login, and why this order makes sense.
2026-05-09 · 7 minControl and Projects serve genuinely different needs and should stay separate. Here is the source of truth for each, what their tiles show, and how to improve them.
2026-05-08 · 18 minA deep technical dissection of how FleetCrown, Beacon, and Zellij connect to form a prompt-injection system — and the five structural bugs that make it unreliable when it looks like it should just work.
2026-05-06 · 10 minA precise breakdown of why Claude currently triggers Beacon while Codex, Gemini, OpenClaw, and provider-layer systems like OpenRouter do not.
2026-05-06 · 13 minThe concrete architecture needed to make Beacon and FleetCrown lifecycle-driven instead of Claude-driven, with a clear split between adapters, control plane, and clients.
2026-05-06 · 7 minWhy Today is not a dashboard of random widgets, but the daily operating surface that turns a sprawling life system into an actionable sequence.
2026-05-06 · 7 minWhy the Prompt Library matters: it turns useful agent interactions from tribal memory into reusable operational templates.
2026-05-06 · 11 minWhy a continuation popup is a lifecycle feature of FleetCrown itself, not a Claude feature, and why the current behavior proves the control plane is still too vendor-shaped.
2026-05-06 · 7 minWhy the System product matters: it is where FleetCrown stops pretending to be only an app UI and exposes the runtime and autopilot machinery underneath.
2026-05-06 · 7 minWhy Projects are not just records in a database, but the basic execution unit that connects repos, prompts, goals, and agent state.
2026-05-06 · 7 minWhy People should not be a static contacts list, but a living map of social context, relationship health, and interaction history.
2026-05-06 · 7 minWhy the Money product should make recurring spend and financial commitments legible enough to support real decisions instead of passive accumulation.
2026-05-06 · 7 minWhy Memory exists as the long-term substrate beneath projects, people, goals, and interactions instead of as a decorative graph view.
2026-05-06 · 6 minWhy the Habits product is not about streak gamification, but about making repetition legible enough to improve.
2026-05-06 · 7 minWhy Goals exist to connect direction with execution instead of becoming isolated motivation artifacts.
2026-05-06 · 6 minWhy Events exist to convert future obligations and opportunities into something the system can prioritize before they become crises.
2026-05-06 · 8 minWhy Control is the most structurally important screen in FleetCrown: it turns multiple agents and multiple projects into one controllable execution surface.
2026-04-29 · 9 minA deep breakdown of the prompt intent that keeps projects out of drift by forcing highest-impact execution in each cycle.
2026-04-29 · 8 minA detailed origin story of the autonomous project loop: from tab-level prompting to orchestration, and why it compounds delivery speed.