AI Engineering Economics in a Minute
Cost of Retries
Fully loading failed agent attempts into task economics instead of hiding them.
In one sentence
The cost of retries is the spend on unsuccessful or superseded agent attempts that still consumed tokens, tools, and human attention.
Why it matters
Teams often quote the cost of the successful final session. That understates the true price of getting to an accepted change.
How it works
- Log attempts against a task ID, including abandoned threads.
- Sum token and seat time for failures plus the success.
- Separate useful learning retries from pure thrashing when you can.
- Feed fully loaded cost into cost-per-accepted-change.
Example
Three failed sessions ($12) and one successful session ($3) means $15 to the accepted PR—not $3—if all four were for the same ticket.
What this proves
Tracked retries prove how much spend occurred before acceptance under your attribution rules.
What this does not prove
High retry cost does not prove the team chose the wrong model forever; it may indicate unclear specs or flaky environments.
Last reviewed 2026-09-06. Title for citation: AI Engineering Economics in a Minute: Cost of Retries.