I taught my AI agent a 1970s invention method, then graded it blind

Description and use

The TRIZ skill enabled the agent to cut inference costs to ~$15k/month while keeping wrong answer rate ≤3 %.

Problem

Reduce inference bill while keeping answer reliability below 3%

Outcome

Using the TRIZ skill, the team achieved a 65% reduction in inference cost while keeping wrong answer rate at or below 3% for 4 weeks, meeting the target within 8 weeks

Workflow

• Ask agent to restate problem and resources
• Identify contradiction (improving serving cost worsens reliability)
• Lookup TRIZ matrix for parameters 23 and 27
• Select inventive principles (#35 Parameter changes, #11 Beforehand cushioning, separation on condition)
• Generate actions: route specific ticket types to small model, fallback to large model when needed
• Validate with Socratic questions
• Create plan with resources, watch for, done when

Evidence

Improving serving cost (23, Loss of substance) worsens answer reliability (27, Reliability). Physically: the model must be small, to be cheap, and large, to be right.

Source

38 of 39 with the skill vs 1 of 39 without

Source

The plan ends like this: Watch for: wrong-answer rate on the small-model route above 4% for a week, then move that ticket tag back to the large model. Done when: monthly bill at or below $21k with wrong answers at or below 3% for 4 weeks in a row, within 8 weeks.

Source