Specialist-model research desk

Next steps

What should we do first?

Frontier Development research deskSection 12 · Page 1 of 12Last updated 26 October 2026

Choose one costly, repeatable workflow and appoint a buyer-side decision owner. The immediate output should be a short decision brief with a credible no-build option, not a production purchase order.

For document extraction, name the document family, critical fields and downstream action. For support software, distinguish a draft response from an autonomous action; they have different acceptance and permission requirements. Write the actual problem in operational terms: rejected records, review minutes, incorrect routing, excessive latency or cost per accepted task.

This is a proposed decision process. It has not been tested against the reader’s workload, and no universal saving, timetable or project price is implied. Hamel Husain’s error-analysis guide supports beginning with concrete failures rather than a preferred model architecture.

TRACE THE EVIDENCE

Sources cited on this page

  1. Hamel Husain’s error-analysis guide →hamel.dev