An Agent Quietly Burned Through My Budget

An agent stuck in a loop does not announce itself. It retries, spawns follow-up work, and burns tokens overnight — and the first signal you get is an invoice. For a solo operator funding everything personally, that is not an annoyance. It is the reason to turn agents off entirely, which means the runaway loop costs you twice: once in money, and once in everything the agents would have done for you afterward.

Dashboards are not limits

Most tools answer this with a usage graph — informative at exactly the moment it is too late. What bounded spend actually requires is a hard cap the system enforces, not a chart you promise to check.

Task Machine budgets are hard limits. Set spending limits in money or tokens at any level — the whole workspace, a project, a goal, a single agent, one workflow. You get an alert at 80% and again at 100%. Agents pause when they hit the cap. Not "notify and continue" — pause.

Agents that need more have to ask

The part that makes caps livable is what happens at the limit. An agent that needs more budget does not silently fail, and does not silently continue: it asks. The request lands in your inbox like any other approval, with the context to decide — what it is working on, what it has spent, what it needs. You grant the increase in a click or you do not.

That turns the nightmare scenario into a Tuesday. The loop that would have run all night instead hits its cap at 11 p.m., pauses, and asks. You wake up to a request, not a bill.

Budgets are what make delegation cheap

The counterintuitive effect: hard caps let you delegate more, not less. When the worst case on any experiment is bounded — this agent can spend at most twenty dollars before anything needs me — trying things stops being a financial decision. You hand over more work precisely because the downside is fixed and visible, and the spending that does happen shows up attributed in your usage view, agent by agent, workflow by workflow.