Knob

Effort

You might say: reasoning effort, thinking budget

A dial for how much reasoning the model does before it answers.

What it does
More effort spends more output tokens and time for a better shot at hard problems. Set it per task or subagent so each job gets the reasoning it needs.
Ask
How much reasoning should this task or subagent spend before it answers?
Typical
Low for mechanical edits and lookups; Medium for everyday coding; High for tricky bugs and multi-step plans; Max for the hardest problems.
Trouble
Low effort on a hard problem produces a confident, shallow answer; max effort on a one-line rename spends time and cost without useful gain.

What to look for

  1. An effort level beside the model in a run record
  2. Higher effort with more output tokens and longer latency
  3. Different subagents carrying different effort settings

Connections

Sources

  1. Matt Pocock, AI Coding Dictionary, EffortA dial for how much reasoning the model does before it answers. More effort spends more output tokens for a better shot at hard problems.
  2. Matt Pocock, AI Coding Dictionary, EffortHigher effort costs more and runs slower. The reasoning is emitted as tokens, billed as output tokens even when you never see them, and produced one token at a time — so turning effort up lengthens the wait before the answer arrives and adds to the bill.
  3. Matt Pocock, AI Coding Dictionary, EffortLow | Mechanical edits, lookups, well-specified changes with one clear path.
  4. Matt Pocock, AI Coding Dictionary, EffortMedium | Everyday coding — the usual default.
  5. Matt Pocock, AI Coding Dictionary, EffortHigh | Tricky bugs, design decisions, multi-step plans.
  6. Matt Pocock, AI Coding Dictionary, EffortMax | The hardest problems, where a wrong answer is expensive to unwind.
  7. Matt Pocock, AI Coding Dictionary, EffortSet effort too low on a hard problem and you get a confident, shallow answer that skipped the reasoning the problem needed — it reads fine and is wrong in a way that costs you later. Set it to max for a one-line rename and you sit through a long think that produces nothing the lowest setting wouldn't have.
  8. Lauren Tan, pstack, Phase A: FramePick each subagent's reasoning level and fast or standard speed for the job, per the Subagents section of the poteto-mode skill.

Use it on your project

teardown-apply: hold "Effort" (effort) from the teardown at https://teardownatlas.com/software-factory/ against my project

Paste it into a coding agent that has the teardown skills.

An AI agent wrote this map with the teardown skill. It can contain errors: check the sources before you rely on a part. About and legal