Overview

Mitchell Hashimoto shares practical strategies for effectively adopting AI coding agents in development workflows. His three-stage approach focuses on deliberate practice and energy optimization rather than immediate productivity gains.

View Original

Key Arguments

  • Deliberate practice through duplication builds AI agent proficiency - developers should manually complete tasks, then recreate identical solutions using agents: Hashimoto literally did work twice: completing tasks manually first, then fighting agents to produce identical results without showing them the manual solution. This builds understanding of where agents excel and struggle.
  • End-of-day agent deployment maximizes low-energy productivity - schedule AI agents during natural energy dips: Block out the last 30 minutes of every day to kick off agents. The hypothesis is that agents can make positive progress during times when developers can’t work effectively anyway, turning dead time into productive time.
  • Selective delegation to ‘slam dunk’ tasks optimizes developer focus - once confident in agent capabilities, hand off routine work: After identifying tasks agents can reliably handle, developers should delegate those while focusing their energy on more complex, interesting problems that require human insight.

Implications

This systematic approach addresses the common problem of AI tools feeling more like obstacles than aids. Rather than expecting immediate productivity gains, developers can build genuine AI proficiency through structured practice and strategic timing, ultimately creating sustainable workflows where human creativity focuses on high-value problems while agents handle routine execution.

Counterpoints

  • The duplication approach may be inefficient in the short term: Doing work twice initially seems to slow down immediate productivity and could be seen as wasteful of development time.
  • Agent delegation requires significant upfront investment: The learning curve and trust-building process may not pay off for developers working on highly varied or one-off projects.