Amp adds a “rush” agent mode for faster, cheaper runs on small tasks
November 7, 2025 — Amp introduced a new agent mode called rush, intended for quicker, lower-cost edits on small, well-scoped developer tasks. The mode trades some capability for speed and cost savings: token-by-token, rush is about 67% cheaper and 50% faster than smart, though its overall prompt-to-result behaviour depends on task complexity.
What rush changes under the hood
Rush is a deliberate combination of system prompt, tools, and model selection — currently running on Haiku 4.5. It supports all of Amp’s subagents, including both the oracle and the librarian. To gain execution speed, the mode omits visible TODO lists in its workflow.
Best-use cases
Rush is designed for compact, well-defined work streams where quick iteration matters more than maximal reasoning. Typical fits:
- Simple bug fixes
- Small UI tweaks
- Minor features
For best results, mention the specific files that require changes. When applied to these kinds of tasks, rush can deliver substantial savings in latency and cost.
When not to pick rush
For broader or ambiguous work, rush is not recommended. Tasks that tend to degrade rush’s efficiency include:
- End-to-end features
- Bugs without a clear diagnosis
- Architecture refactors
In those scenarios, rush may expend extra tokens and time correcting mistakes and can even be slower than smart while offering only modest cost savings. The team’s examples show a small bug completed in 37 seconds at $0.12 (noted as 44% faster and 77% cheaper than smart), while a complex refactor saw rush take twice as long and only 19% cheaper.
How to enable rush
Rush can be activated from the Amp CLI with the command: mode: use rush
It is also selectable from the Amp editor extension’s prompt field.
Original source: https://ampcode.com/news/rush-mode
