Specialist-model research desk

Hiring signals

What should a CTO ask before hiring internally?

Frontier Development research deskSection 05 · Page 5 of 12Last updated 26 October 2026
  1. Is the task stable enough to justify continuing ownership? A one-off feasibility question may need a bounded engagement; a core production capability may justify a durable internal role.
  2. Can we supervise the specialist? An expensive hire without access to domain reviewers, usable data and a product owner can still be blocked.
  3. Which work is recurring? Separate model experiments from operational incident response and customer integration.
  4. What must remain independent? The person optimising a model should not be the only authority deciding whether the final test passed.
  5. What happens during absence or turnover? Require documentation, reproducibility and access held by the company.

This is a make-or-buy decision about capabilities, not a contest between salary and a consultancy’s headline quote. Compare the work covered, duration, supervision, tooling, benefits, recruiting time and the opportunity cost of diverted staff.

TRACE THE EVIDENCE

Section source library

This page contains no direct external citations. These are the section’s references; they do not independently validate the desk’s proposed scoring or methods.

  1. Forward Deployed Engineer, Post-Training — Together AI, San Francisco →job-boards.greenhouse.io
  2. Machine Learning Research Scientist, Evaluations — Scale AI →job-boards.greenhouse.io
  3. Software Engineer — Training Product — Baseten →jobs.ashbyhq.com
  4. AI Field Engineer — Strategic Partnerships — Fireworks →jobs.ashbyhq.com
  5. A Field Guide to Rapidly Improving AI Products — Hamel Husain →hamel.dev
  6. Demystifying evals for AI agents — Anthropic, 9 January 2026 →www.anthropic.com