• Jeremy Scheff's essay describes a loss of craft satisfaction, not a loss of employment; he reports his own business is having one of its best years.
• Scheff says he now reaches for an agent before his text editor, even though he calls the habit 'addictive and meaningless'.
• Simon Willison frames agents as a gift for programmers who mainly want working systems and a real loss for those who enjoyed solving the problems themselves.
• Neither source offers productivity, quality or headcount data, so neither can settle whether agent-assisted teams ship better software.
• Engineering leaders should ask which engineers are being asked to give up the work they valued, and plan for that before it shows up as attrition.

The strongest claim in Jeremy Scheff's essay 'My obituary for programming' is about meaning, not money. Scheff, who runs zengm.com, says coding agents remove the puzzle-solving that made the job satisfying, while his own business is having one of its best years ever. Simon Willison linked the essay on 19 October 2026 and drew a split: a gift for people who want working systems, a loss for people who loved solving the problems themselves. Neither offers data, so this is a morale signal rather than a finding.

What Scheff says was lost

Scheff's essay, dated October 8, 2026, describes the loss as craft, status and judgment. He says programming was a game with real-world value, that being one of the few people who could instruct computers precisely carried standing, and that his senses now feel 'hijacked'. He argues that if hard things become easy, it is fair to ask whether they matter, and he says vibecoding 'is not programming'. These are his own reported experiences.

He still uses the agents

Scheff does not argue that agents fail. He credits them with diagnosing and fixing bugs from a report or error log, with speedups he describes as 2x and 3x in optimization work, and with large refactors he had long postponed. He says he now reaches for an agent before his text editor and calls the habit 'addictive and meaningless'. The tension he describes is between measurable usefulness and a felt loss, and the essay leaves it unresolved.

Willison's reading: a split verdict

Willison's post is a short link note, and its point is a division by motivation. In his reading, agents are a gift for programmers who mainly enjoy building working systems, and a real loss for those who enjoyed the puzzle itself. He quotes Scheff's lines about tricky logic puzzles and about setting the strategy and the timeline. The post adds no data of its own, so it frames the debate rather than settling it.

What the evidence does not show

Neither source measures anything a team could act on as proof. There are no controlled comparisons of agent-assisted and unassisted engineers, no defect rates, no hiring or headcount figures. Scheff's best-year business is one anecdote, and the speedups come from his own projects, which we have not independently reviewed. Anyone citing this essay as evidence that agents shrink programming work, or improve it, is reading more into it than it contains.

Verdict: a retention risk to plan for

The practical lesson is narrow and defensible: adoption can succeed on output while quietly costing the people doing it something they valued. Engineers who joined for the puzzle may disengage even as throughput rises, and a tool that works can still be resented. Leaders rolling out coding agents should ask which engineers are giving up the work they liked, and test that with their own surveys and attrition data rather than this essay.