Where the Time Goes

Fictional characters in this episode: Jonas (junior developer), Daniel (senior developer; in episode 1 a stand-in for the study participants, from now on part of the team), Miriam (team lead for Daniel and Jonas), Hanna (wedding photographer, this episode's image).

Small glossary

• Pull request: a proposed change that someone reviews before it goes into the shared code.
• Diff: the side by side view of old and new code.
• Merge: combining changes into the shared code; on many teams only after a review.
• Commit: a saved snapshot in version control.

Tool tip: read your lead times instead of guessing

A coding agent (for example Claude Code) can run Git and the GitHub CLI on its own. Example prompt: "For my last 10 pull requests, show me the elapsed time between the first commit, opening, first submitted review and merge." These are lead-time intervals, not pure waiting times: nights, questions and rework are mixed in. You need additional records to separate active work from actual waiting.

What's behind it:

• gh pr list --author @me --state merged --limit 10 --json number,title,createdAt,mergedAt,reviews,commits returns creation and merge time, reviews and commits for each pull request. https://cli.github.com/manual/gh_pr_list
• git log --format='%h %aI %s' lists commits with their author date (ISO 8601); a first commit is not necessarily when work started. https://git-scm.com/docs/git-log
• Claude Code works with gh (for example gh pr create) and can be used in scripts with git log | claude -p "...". https://code.claude.com/docs/en/common-workflows

Against duplicate functions: have the agent search for existing functions before it starts, and write your project's key building blocks into the notes file it reads at every start (Claude Code: CLAUDE.md, loaded in every session; Claude Code can also read AGENTS.md; AGENTS.md is supported by Codex, Cursor and the GitHub Copilot coding agent, among others). https://code.claude.com/docs/en/memory · https://agents.md

Office version: a document's version history shows who changed what and when (for example in Google Docs). https://support.google.com/docs/answer/190843

Sources

Named in the episode

• Atlassian (2025): State of Developer Experience Report 2025. Survey of 3,500 developers and managers in six countries; 16% of time spent coding; 68% save more than 10 hours a week with AI; 50% lose more than 10 hours a week to organizational inefficiencies. Vendor survey, self-reported. https://www.atlassian.com/blog/developer/developer-experience-report-2025
• Lee, Sarkar, Tankelevitch, Drosos, Rintel, Banks, Wilson (Microsoft Research and Carnegie Mellon, CHI 2025): The Impact of Generative AI on Critical Thinking. 319 knowledge workers, 936 examples. Self-reported. https://doi.org/10.1145/3706598.3713778 (PDF: https://advait.org/files/lee_2025_ai_critical_thinking_survey.pdf)
• Anthropic (December 2, 2025): How AI is transforming work at Anthropic. Survey of 132 engineers and researchers, 53 interviews, Claude Code usage data (August 2025). Self-estimated share of work using Claude rose from 28% to 59%; more than half can fully delegate only 0-20% of their work; one engineer estimates over 70% of his work is code review. Selective environment, no control comparison. https://www.anthropic.com/research/how-ai-is-transforming-work-at-anthropic
• Faros AI (July 23, 2025): The AI Productivity Paradox Report 2025. Telemetry from more than 10,000 developers across 1,255 teams; with high AI adoption 98% more merged pull requests, 91% longer review time, 154% larger pull requests; no significant correlation at the company level. Vendor data, correlation. https://www.faros.ai/blog/ai-software-engineering

Mentioned without names

• Alphabet (October 29, 2024): Q3 earnings call, remarks by CEO Sundar Pichai ("more than a quarter of all new code at Google is generated by AI, then reviewed and accepted by engineers"). https://blog.google/inside-google/message-ceo/alphabet-earnings-q3-2024/
• GitClear (2025): AI Copilot Code Quality: 2025 Data Suggests 4x Growth in Code Clones. 211 million changed lines, 2020-2024; in 2024, copied lines exceeded moved lines for the first time. Vendor data, correlation. https://www.gitclear.com/ai_assistant_code_quality_2025_research
• Goodhart's law (Goodhart 1975; the commonly quoted wording is by Marilyn Strathern, 1997). Covered in depth in episode 8. https://en.wikipedia.org/wiki/Goodhart%27s_law
• Goldratt, E. M., Cox, J. (1984): The Goal. Theory of constraints. https://en.wikipedia.org/wiki/Theory_of_constraints
• Stack Overflow (2025): Developer Survey 2025, AI section. 66% of respondents to the question about frustrations name "AI solutions that are almost right, but not quite". https://survey.stackoverflow.co/2025/ai
• Ebert, Castor, Novielli, Serebrenik (SANER 2019): Confusion in Code Reviews: Reasons, Impacts, and Coping Strategies. https://doi.org/10.1109/SANER.2019.8668024
• Warm, Parasuraman, Matthews (2008): Vigilance Requires Hard Mental Work and Is Stressful. Human Factors 50(3). https://doi.org/10.1518/001872008X312152
• DORA: Software delivery performance metrics (including change lead time). https://dora.dev/guides/dora-metrics-four-keys/

More on this (not in the episode)

• Baolin, Harvey (DORA, March 10, 2026): Balancing AI tensions. Thematic analysis of 1,110 open-ended responses from Google software engineers; time saved is redistributed into review and prompting effort; recommendation "Measure impact, not output". Qualitative, one company. https://dora.dev/insights/balancing-ai-tensions/
• DORA / Google Cloud (2024): Highlights from the 10th DORA report. For every 25% increase in AI adoption, an estimated 1.5% drop in delivery throughput and 7.2% drop in delivery stability. Survey, correlation. https://cloud.google.com/blog/products/devops-sre/announcing-the-2024-dora-report
• Dijkstra, E. W. (1988): EWD 1036 ("lines spent" rather than "lines produced"). https://www.cs.utexas.edu/~EWD/transcriptions/EWD10xx/EWD1036.html

Looking back at episode 1

• Becker, Rush, Barnes, Rein (METR, 2025): Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity. https://arxiv.org/abs/2507.09089

Foundation of the series

• Healthy AI First Software Development, version 2 (October 1, 2026), chapter 3 "What AI changes about work".