Field Manual · Ten Failure Modes to Watch
Ten ways using AI quietly goes wrong
None of these look like failure from the inside. They look like productivity. That is what makes them worth naming.
The failure modes below come from the field manual at the back of AI Won't Save You. Each one has a symptom you can spot, a reason it is so tempting, and a correction you can apply this week.
1. Prompt collecting
Symptom: you own hundreds of prompts and have few finished projects.
Why it happens: prompts create a cheap feeling of future capability.
Correction: delete most of them. Keep prompts attached to real, recurring workflows. A prompt without a trigger, a standard and a use case is clutter.
2. Research as avoidance
Symptom: every uncertainty starts another research cycle.
Why it happens: AI makes research feel like action without requiring any exposure.
Correction: compare the information you consumed with the evidence you created. Once new information is unlikely to change your next action, take the action. More on this in the 10-10-1 rule.
3. Model hopping
Symptom: you keep switching tools because another model might be slightly better.
Why it happens: the technology moves fast, and novelty feels strategic.
Correction: pick a primary tool for a commitment window. Switch when a specific limitation matters, not because a benchmark moved.
4. Fluent acceptance
Symptom: polished output gets less scrutiny.
Why it happens: professional language signals competence.
Correction: strip the output down to claims, assumptions, sources and consequences. Match verification to the stakes, as in this chapter on critical thinking.
5. Premature automation
Symptom: you automate a process whose exceptions you do not understand.
Why it happens: automation is visible leverage, and manual work feels primitive.
Correction: experience, extract, systemize, then automate. Require strong examples and an escalation line before giving a system autonomy. See the Systemization Ladder.
6. Permanent apprenticeship
Symptom: you keep doing low-value work by hand because "I need to stay close to the craft."
Why it happens: identity gets attached to tasks that once built your expertise.
Correction: set graduation criteria. Once the lesson is internalized and you can keep your judgment sharp by sampling, offload the repetition.
7. Agent theater
Symptom: a complicated multi-agent setup produces impressive logs and little useful output.
Why it happens: agent count feels like capability.
Correction: measure closed loops. One reliable workflow that reaches an acceptable result beats ten agents generating activity. More in delegating to AI agents.
8. Attention bankruptcy
Symptom: AI saves production time but floods you with more to review.
Why it happens: every efficiency gets reinvested in more output.
Correction: compress before you escalate. Some saved time must stay saved.
9. Identity outsourcing
Symptom: you increasingly ask AI who you are, what you want and who you should become.
Why it happens: a coherent outside interpretation feels like self-knowledge.
Correction: treat AI's interpretations as hypotheses. Test them against your behavior, trusted people, your commitments, prayer and reality.
10. Authority drift
Symptom: because AI is often useful, you start treating usefulness as legitimacy in every area.
Why it happens: repeated correctness builds trust, and that trust migrates upward.
Correction: separate execution, judgment and ultimate authority. Do not let competence in one category silently grant sovereignty in another. See why intelligence was never the point.