The AI-pilled engineer: most in demand, most exhausted

Season 2, Episode 2 • 38:50

The AI-pilled engineer: most in demand, most exhausted

Welcome back to I'm a Software Engineer ~ What Next? The podcast for devs figuring out their next move. Hosted by James Wilson and Matt Sinclair, we talk honestly about what it takes to build a meaningful career in tech.

This is part one of a two-parter on going all in on AI, and part one is the engineer. James opens with a hunch: the most AI-pilled engineers he speaks to are the most in-demand people in the market today, and also the most exhausted, and nobody is pricing it. Matt, who runs six to twelve coding agents at once, knows exactly what he means.

In this episode, we cover:

  • What "AI-pilled" officially means: Ramp's AI spend data, where the top 1% of firms spend over $7,000 a month per employee, and why token spend is about as good a metric as lines of code
  • Simon Willison on using coding agents well taking "every inch" of his 25 years as a software engineer, and being wiped out for the day by 11am
  • Agentic coding as broad work done at the intensity of deep work, and the low-level anxiety of six or eight agents running at once
  • Create and edit becomes instruct, ingest and edit: the new cognitive load of taking in code you didn't write
  • Read every line when you're starting cold; after that, the work is the guardrails and the tests more than any single prompt
  • Skill or phase? What the first month of back-to-back Covid Zoom calls suggests about whether this settles down
  • "Dunning-Kruger as a service", and why both extremes ("it's useless" and "we don't need engineers") are insane
  • Overhiring dressed up as AI layoffs, a US tax change on R&D costs, and why more small companies are hiring engineers
  • On-demand software: tools that were never worth building before, the new version of the office Excel macro
  • Rule one of security: don't write your own security software, and certainly don't vibe code it
  • AI-assisted coding interviews, where candidates can't explain code they didn't write, and why interviewers should rethink what they're testing
  • Ask about the requirements before you build: you're not Google
  • Why token maxing is nonsense: give a junior and a senior engineer the same subscription and both generate 300,000 lines; the senior throws 250,000 of them away

⏱️ Chapters

  • 0:00 The AI-pilled engineer: most in demand, most exhausted
  • 0:54 What "AI-pilled" officially means: Ramp's AI spend data
  • 5:59 Simon Willison: every inch of 25 years, wiped out by 11am
  • 6:33 Broad work at deep-work intensity: six or eight agents at once
  • 9:47 Create and edit becomes instruct, ingest and edit
  • 10:26 Read every line when you start cold; then it's guardrails and tests
  • 14:31 Skill or phase? The Covid Zoom-call parallel
  • 15:41 Dunning-Kruger as a service
  • 18:28 Overhiring dressed up as AI layoffs
  • 22:42 On-demand software: the new Excel macro
  • 24:24 Rule one of security: don't vibe code it
  • 29:02 AI-assisted coding interviews
  • 33:32 Ask about the requirements: you're not Google
  • 36:51 Why token maxing is nonsense
  • 38:35 Next week: the AI-pilled company

🔗 Links

🗒️ Show Notes

🎙 Hosts

🌐 More from us

🎬