Peter Steinberger ("the claw father," now at OpenAI) argues that the era of juggling 10+ terminal windows was a false peak of productivity where the human was really acting as scheduler, router, and memory. He describes a shift toward managing a long-lived manager agent that delegates to a team of workers, enabled by three changes: server-side compaction, coordination, and automated triggers. As agents get better, the human bottleneck moves from tokens to compute to attention — so the real work is deciding where to spend it. His closing thesis: "The future is not 20 terminals, it's better loops."
Steinberger opens on how much has changed in a few months. In January, juggling 10 or more terminal windows — always waiting for one to finish so he could steer and queue new work — felt like peak productivity. In hindsight it feels silly: he wasn't orchestrating, he was polling, acting as the scheduler, the router, and the memory.
Pairing with one agent across 10 terminals wasn't pairing at all — it was managing 10 direct reports. His new default is talking mostly to a long-running manager that delegates work to a team, dropping down to pair with a worker only for tricky work. He now manages the manager of a small company of agents.
Three changes made the shift possible. First, server-side compaction made long-running tasks reliable enough that he stopped optimizing around fresh sessions. Second, coordination lets one thread create and steer the right projects. Third, automation can wake the same manager when something happens.
Together these give persistent context, delegation, and triggers — "There's your loop."
Once the loop works, the bottleneck keeps moving. Last year the constraint was tokens — "fixed" by joining OpenAI (a strategy he admits does not scale). Then the constraint became compute, as parallel threads made his MacBook sound like a jet engine; that's mostly solved by using separate test boxes so agents can run tests off-machine.
Now the primary constraint is attention, and unlike tokens or compute you can't simply add more. So the most important skill today is deciding where to spend it. Watching code fly by felt necessary with earlier models — you'd hit escape and steer — but the latest generation understands intent well enough that watching is largely a waste.
He sketches the flow: someone files an issue on an open-source project, the manager wakes, checks it against the project's goals, notes, and vision, and decides if it fits. If so, it spins up a worker that investigates, implements, runs tests, and another agent reviews. When the manager needs him, it returns a PR, the original issue, the proposed diff, maybe a video or a running build he can VNC into. He reviews once, leaves a note, maybe approves, and the loop continues and can land after checks pass. The agent runs the inner execution loop; the human sets direction and makes decisions in the outer loop.
He notes Paul Salt already runs a version of this — a pinned "chief of staff" that wakes every 10 minutes to coordinate GitHub work, creating sidebar threads to jump into. Once the manager is long-lived, tying it to a laptop feels wrong: Codex can already move work between hosts, and OpenClaw has a gateway and nodes, but neither feels like the final form. The manager shouldn't be a session trapped inside an app — it should be an agent you can text, steer from Slack, or reach from anywhere. The closing thesis: the future is not 20 terminals, it's better loops.
"I thought I was orchestrating. Really I was polling. I was the scheduler, the router and the memory."
"I manage the manager of a small company of agents."
"So we have persistent context, delegation, and triggers. There's your loop."
"Now I'm primarily constrained by attention. And unlike tokens or compute, I can't simply add more of it. So the most important skill today is deciding where to spend it."
"The manager shouldn't be a session trapped inside your app. It should be an agent that I can text, steer from Slack, or hear from wherever I am."
"The future is not 20 terminals, it's better loops. Let's build them."
| Time | Topic |
|---|---|
| 00:00 | Intro: the claw father takes the stage |
| 00:31 | Juggling 10 terminals wasn't orchestration |
| 01:33 | Three changes: compaction, coordination, triggers |
| 02:05 | The moving bottleneck: tokens, compute, attention |
| 03:39 | Inner vs outer loop: manager delegates to workers |
| 05:12 | Un-trapping the manager; better loops, not more terminals |