A small team's operating system for coding agents
The promise of coding agents is not that a two-person startup becomes a two-hundred-person company overnight. It is simpler: a small team gets more attempts, more parallel exploration, and less time lost to mechanical work.
That only works when the team can still understand what is happening.
For solo developers and startups, the scarce resource is not raw compute. It is attention. Every extra terminal to watch, every hidden retry loop, and every decision that arrives without context spends that attention. A good agent workspace gives some of it back.
1. Make the work visible before you automate more of it
Do not start by asking how many agents can run at once. Start by asking whether you can answer three questions in a few seconds:
- What is being worked on?
- What is blocked?
- What changed?
In theboringoffice, live opencode events show up as a floor, an agent roster, work threads, activity, board entries, and mail. The point is not to create a game. It is to give the human a peripheral view of the work so they do not need to reconstruct it from scrollback later.
2. Queue decisions; do not queue the whole shift
A coding agent will eventually need a human to answer a question or grant a permission. The wrong response is to freeze everything until that person returns.
The Permission Queue keeps requests ordered and explicit: allow once, allow always, or reject. Questions can be re-opened with /question; dismissed permission prompts can be revisited with /perm. This makes human attention a deliberate input, rather than a background polling job.
3. Treat context as infrastructure
A productive system does not become blank every time a terminal quits. It remembers what happened, what was decided, and where the work stopped.
agentmemory provides that continuity for the office: sessions, lessons, semantic recall, board actions, and mail signals. The office restores the last chat by default, and /session lets you choose a past session when you need to return to a specific room.
4. Lower the cost of the first honest experiment
Adoption should not begin with an integration project. It should begin with a clear look at the thing you are considering.
theboringoffice --demo
Demo mode is intentionally labeled and uses simulated events. When the interaction makes sense, start live mode or attach an existing server. Small teams do not need another mandate. They need a reversible way to learn whether a tool earns a place in their day.
The best agent setup is not the one with the most moving parts. It is the one that gives a small team more leverage while preserving its ability to see, decide, and remember.