Skip to content
Written summaryAnthropic · research reportAnthropic · appendix
Research Insights Made Simple logo
Research Insights Made Simple #20 · July 16, 2026Podcast · with Evgeny Sergeev

Why Coding Agents Do Not Eliminate Expertise

A review of Anthropic's research: delegation, verification, and recovery when using Claude Code

/ Research Insights Made Simple #20 · Coding agents & expertise

Slide contents

  1. 1. Why Coding Agents Do Not Eliminate Expertise

    A review of Anthropic's research: delegation, verification, and recovery when using Claude Code

  2. 2. Product telemetry is not a market law

    Original redraw of Anthropic's research frame

  3. 3. A large sample still has boundaries

    Redrawn from Appendix · Sample · p. 2

  4. 4. Claims pass through models and telemetry

    Original redraw of the measurement pipeline

  5. 5. Nine modes label the dominant work

    Redrawn from Figure 1 and Appendix · Work Mode

  6. 6. Most sessions are still about code

    Redrawn from Figure 1

  7. 7. People choose what; Claude chooses how

    Redrawn from Figure 2

  8. 8. One prompt triggers roughly ten actions

    Redrawn from Who decides what

  9. 9. Expertise is task-specific, not occupational

    Redrawn from Level of expertise

  10. 10. Three observable signals stand in for expertise

    Redrawn from Table 1 and Appendix · User expertise

  11. 11. Expertise and success are measured nearby

    Original diagram of possible construct overlap

  12. 12. Expert sessions contain over twice the actions

    Redrawn from Figure 3

  13. 13. Session mix shifts from fixing toward operating

    Redrawn from Figure 4

  14. 14. Task value is a relative estimate

    Redrawn from Figure 4 and Appendix · Task value

  15. 15. Success has three evidence thresholds here

    Redrawn from Table 2 and Appendix · Derived measures

  16. 16. The largest gap separates novices from competence

    Redrawn from Figure 5

  17. 17. Expertise shows more clearly under trouble

    Redrawn from Figure 5 · troubled sessions

  18. 18. Success tracks task knowledge more than occupation

    Redrawn from Figure 6

  19. 19. Validation supports the gradient, not causality

    Redrawn from Appendix · Validation and robustness

  20. 20. The study shows association, not cause

    Supported

    Expertise tracks outcomes

    Experts delegate more

    Gradient survives controls

    Not established

    Training causes the same effect

    More actions mean better work

    Result generalizes to all agents

  21. 21. Expertise may expand delegation bandwidth

    Original model: framing × verification × recovery

  22. 22. Enablement should teach judgment, not syntax

    Build in people

    Precise task framing

    Knowledge of invariants

    Diagnosis and recovery

    Provide in platform

    Domain context and vocabulary

    Checks by default

    Observable artifacts

  23. 23. Evaluate outcomes, delegation, and recovery

    Original scorecard derived from the report's limits

  24. 24. A pilot should separate input from outcome

    Measure expertise before the task

    Review artifacts blind

    Label trouble from telemetry

    Check outcomes after adoption

  25. 25. Leaders need a new work contract

    People own

    Goal and constraints

    Meaningful verification

    Decisions under trouble

    Platform owns

    Accessible context

    Provable results

    Boundaries and rollback

  26. 26. Track whether the expertise premium shrinks

    Verified-success gap narrows

    Novices recover more often

    Planning shifts to the agent

    Quality remains stable

  27. 27. Agents absorb execution, not judgment

    People still decide what

    Competence captures the main gain

    Expertise supports recovery

    Outcomes beat action volume

    Measure delegation bandwidth together with quality.

  28. 28. Expertise becomes delegation bandwidth

    Not more activity. More verified work per unit of human judgment.

    Book Cube

    The primary report and appendix are linked on the first slide. Continue the discussion in Book Cube.

    Alexander Polomodov, Technical Director & Fellow, T-Technologies

    @Book_Cube