CORE GUIDE
Choose the automation boundary
Automate stable, observable work and keep humans where ambiguity, accountability or exception cost remains high.
Mental model
A task is a good automation candidate when its inputs and success criteria are clear enough to verify automatically and its failures are recoverable.
Why it matters
AI makes it easy to automate a process before understanding it. When hidden judgment, policy exceptions or responsibility boundaries are buried inside a model call, short-term leverage becomes long-term maintenance and operational risk.
01
Find the stable boundary
Decompose the workflow and score each stage by input stability, reversibility, verification cost and exception frequency. Automate stages with clear postconditions first. Keep high-impact or ambiguous decisions behind explicit policy and human accountability until evidence supports a wider boundary.
02
Example: support in a solo business
A solo founder can automate ticket classification, knowledge retrieval and first-draft replies early. Refunds, account access and unusual policy conflicts remain human decisions. Over time, measured exception rates can justify moving specific low-risk actions across the automation boundary.
Common failure modes
- Automating the entire workflow because one happy-path demo worked.
- Ignoring exception handling, ownership and maintenance cost.
- Adding human review everywhere without defining what the human must decide.
Engineering heuristics
- Automate tasks with clear machine-checkable postconditions first.
- Keep irreversible or high-impact actions behind stronger policy boundaries.
- Track exception rate and maintenance burden as automation metrics.
Takeaways
- 01Leverage comes from stable boundaries, not maximum autonomy.
- 02Verification cost helps determine what can be safely delegated.
- 03Every automation needs an owner, exception path and maintenance policy.
Related concepts from the Knowledge Graph
These relationships come from the canonical graph, not a separate Guide taxonomy.