Compute Budget

Framing

The problem

Agent runs can spend tokens, time, and money invisibly — users only learn the cost after the bill arrives.

The pattern

Show pre-flight cost estimates with mode selection and a hard cap; meter the live spend and pause at threshold.

Why chat breaks here

Chat hides compute as a black box — there is no surface to set a ceiling, see live spend, or decide mid-run whether the answer is worth more.

Risks

Cost meters can become anxiety theatre if they are imprecise, distracting, or punish exploration.

Avoid when

The task is cheap enough end-to-end that a budget interface costs more attention than it saves money.

Use when

Agent runs spend tokens, time, or money in ways users cannot see or limit at the start.

DOPE evaluation

Directability

Raise the cap, switch mode, or stop and pocket partial output any time

Observability

Live meter shows tokens, time, and dollars consumed against the budget

Predictability

Mode and cap are explicit upfront — no runaway spend hidden in a long run

Explainability

Estimate breakdown shows which steps and tools drive the cost

In the wild

• Replit Effort-Based Pricing (Replit) — Effort-based pricing since July 2025: a checkpoint costs what the work cost, and hovering the usage icon in the Agent tab shows the figure for each one. Account → Billing adds threshold alerts and a hard budget cap. The $0.25 in the old note was the previous flat rate — under effort-based pricing simple changes typically land under it.
• Devin ACUs (Cognition) — Session Insights reports ACU consumption once a session closes, and Settings → Plans carries the month's usage, remaining quota and on-demand credit balance. Devin no longer defines an ACU as roughly 15 minutes — it is now "the amount of agent effort required to complete a given task," scaling with inference and model, at $2.25 per ACU self-serve. Post-hoc, not pre-flight.
• Claude Code /usage (Anthropic) — /usage prints the session's token counts and a dollar figure, plan usage bars, and a 24h/7d breakdown attributing spend to skills, subagents, plugins and individual MCP servers — the closest thing in this set to an explainable cost breakdown. /cost survives only as an alias. The dollar figure is computed locally at list rates, so it may not match the bill.
• Cursor Spend Limits (Cursor) — The dashboard's Spending tab shows real-time usage across both model pools, remaining allowance, and request-level cost; a monthly spend limit hard-stops AI features until the next billing cycle, and May 2026 added dollar-threshold alerts to Slack or email. All of it sits outside the editor — in there you get a notification only once the limit is already hit.
• ElevenLabs Studio (ElevenLabs) — The one row that carries the pre-flight half: Studio tells you how many credits generating the selected text will cost before you commit, and warns what an export will cost when paragraphs still need converting. Credits rather than dollars, and there is no ceiling that stops a job once it is running.

FAQ

When should I use the Compute Budget pattern?

Agent runs spend tokens, time, or money in ways users cannot see or limit at the start.

When should I avoid the Compute Budget pattern?

The task is cheap enough end-to-end that a budget interface costs more attention than it saves money.

What problem does Compute Budget solve?

Agent runs can spend tokens, time, and money invisibly — users only learn the cost after the bill arrives.

Why is chat the wrong fit for this?

Chat hides compute as a black box — there is no surface to set a ceiling, see live spend, or decide mid-run whether the answer is worth more.

Related patterns

• Often paired with: Draft Mode — Budget caps the run; Draft Mode caps the model. Together they make exploration cheap and commitment explicit.
• Alternative to: Approval Gates — Stop on cost vs stop on step. Both are pause points — Budget is the economic gate, Approval is the human one.
• Often paired with: Autonomy Dial — Set how much autonomy the agent has, then bound how much it can spend exercising it.