kwj.ai · acquisition inquiries from >$999view prospectus →
The Domesday Book ofKWJ · AI

Engineering · 9 min

Prompt-engineering vs fine-tuning: the breakeven

Most prompt-engineering problems are not fine-tuning problems. The reverse is also true.

By C.W. Jameson · Published 19 May 2026 · Last reviewed 19 May 2026

Operators reflexively reach for fine-tuning when a prompt isn't working. Almost always, the prompt was the wrong tool for the wrong job, not the wrong size of the same tool.

When prompt-engineering is the answer

The task is well-described, the model has the capability, the failures are about specification rather than capability. Add examples, tighten the schema, write the constraints explicitly.

When fine-tuning is the answer

The task is repetitive, the volume is high, the prompt is approaching context limits, and the capability is just beyond reach. Fine-tuning amortises the prompt over training and pays back over millions of calls.

When both fail

Pick a different model. Sometimes the capability isn't there.

Frequently asked

Is RAG a substitute for fine-tuning?

Different tool. RAG retrieves facts; fine-tuning shapes behaviour. Both can coexist.

Partner offer

Anthropic's Claude family is the model lineage most operators end up on for serious agent work. The free tier remains useful.

Try Claude →

Affiliate link — see disclosure.

From the Almanac shop

The Operator's Compendium

Every agent harness, every routing pattern, every cost trick. 90-page PDF.

$29Coming soon

All guides