How we run

The goal is to never be the bottleneck, so the work runs on fleets of agents. An agent is a software assistant set up to do real work on its own: it reads files, runs commands, writes drafts and code, and checks its own work before showing a person. A fleet is a team of them assigned to one job under a person’s direction. A typical week has somewhere between 10+ fleets running 2-50 agents each. Some fleets are working on and for companies, some on research, some on a build, a review, a memo, or a problem still being thought through. Some are working; some are waiting.

To make all this functional, agents get a place to work, the memory they need, the repository or source system they are responsible for, and a task list. All that context is part of the agent. An agent or fleet might sit idle for a day; the context is still there when the work resumes.

Currently, Herdr with Pi as the harness is the go-to for fleet operations. Everything stays visible: the agents, the workspaces, what is blocked or moving. The agents have different roles and different model tiers; different models get picked depending on the work. There are also layered and sequenced skills that call different agents and different models depending on the work. The newest development is a “Collective” agent framework that lets a fleet operate on its own, relatively unsupervised. Wild times.

Behind the workspaces are the services that keep things working: memory and state backends, repository and task graphs, timed jobs, webhooks, harness hooks, health checks, and launch-and-reconcile processes. Some work starts when a person assigns it. Some starts when a timer, event, or service says it is time. With the new Collective, the fleet can also start work on its own; what it starts still comes to a person for review.

More work used to mean more hours, or more people, or both. This way, the expensive hours go to the framing, the thinking, and the decisions. The cheap hours run as standing work in the background. That gives a person much more scale than they used to have.

The fleet also changes what can be taken on at once. The old way was one thing at a time, with everything interrupted when something urgent arrived. Now several operations stay warm at once.

A task can begin in a repository, pick up decisions from memory, use a service, and leave a result in a task graph or review queue, or just in a document.

A person sets the context, assigns or redirects the work, and inspects the results. To be fair, there is a lot of steering along the way.

A person decides what work gets done, whether a result is good enough to use, and whether a system keeps running.

No autonomous inbox replies here. Tried it; not good. AI generates many drafts from materials, but a person always edits and sends.

The agents also produce too much stuff: drafts, artifacts, digressions. It is a lot to wade through. There is intuition in what to pay attention to.

The system also stays separated by company and purpose. Memory keeps context between sessions. It is not one giant pool. Operational state keeps current facts, sources, freshness, and proposed changes.

It’s nowhere near perfect. Memory drifts, agents need their boundaries reset, and a service update breaks a connection. A task graph needs repair, a timed job fails. Nothing breaks silently, though: the agents report their own failures, so a person sees what broke and decides the fix. The system has to be watched, updated, and occasionally simplified.

Ten years ago, this was one company, a handful of direct reports, and a calendar. Work moved serially: a person in every meeting, on every thread, and often the bottleneck.

Now several operations stay warm at once: a client engagement, a build, research, a talk. Memory keeps the context between sessions, and a new standing workspace with the right role and review loop takes an afternoon to spin up. Braydon still makes the judgment calls, now at review checkpoints instead of at the front of every task.

Four companies and this consulting practice now run on AI, and a week that used to end with everything unfinished now ends with most things in a decent shape. The system itself is infrastructure: watched, simplified, and repaired. That trade is worth it.