How to keep thinking while working with AI agents
A useful companion to the recent human-in-the-loop notes: agent-heavy work can bias engineers toward fast skimming and judgment, so maintaining slow thinking may require deliberate writing, book reading, and non-agent reflection time.
Links: Original source · Shared link · Related link
Logged at IST: 2026-08-13 08:27 IST
What it is: Tom Critchlow sharing Sean Goedecke’s essay on preserving slow thinking while agent-heavy software work becomes more frenetic.
Gist: Goedecke argues that AI-agent workflows increasingly turn software work into rapid triage: spin off agent sessions, skim outputs, decide whether schema changes, tests, architecture, or implementations look right, then switch again. That is skilled work, but it uses the fast “skim and judge” part of the mind more than the slow “hammock time” required for deep thought and creativity.
His proposed counterweight is deliberately old-fashioned: write in your own words and read actual books, especially dense non-fiction. Writing with an LLM does not work the same way because the value is not only the finished prose; forcing yourself to articulate the idea is how the idea becomes real. Reading slowly and then writing about what you read keeps the slow-thinking habit alive.
He is not arguing against agents. The footnote says he routinely uses six or seven agent sessions on one task: exploration, multiple implementations, review, manual testing. The warning is that some engineering problems, especially large refactors in complicated codebases, still require taste and sustained thought. Current models can often make changes without many errors, but they cannot reliably do them tastefully.
Newsletter angle: Strong companion to the “human is the loop” item: if agent work pushes us toward fast triage, engineers need explicit practices for keeping slow thought alive outside the agent loop.