Taken from Geoffrey Huntley's X Space discussion, titled "On hiring software engineers & identifying talent now that AI is here".
Transcribed by R J Hunter: https://privatebin.net/?4901e18323cbfaa3#3X9BgUN9yprVbB2NULmDexv9VBbcVH4QKFgtSk2nisy3
Main Topic: How to hire software engineers now that AI tools have transformed the field
Three Categories of Candidates:
-
Unacceptable - Engineers who have rejected or ignored LLMs, squandering the past year of learning opportunities. This bar is rising quickly in high-performance companies.
-
Acceptable - Engineers using tools like Cursor and experimenting with LLMs, but manually prompting and staying in the loop for testing/debugging.
-
Ideal - Engineers who:
- Understand the LLM inferencing stack deeply (request/response models, tool calling, array allocation)
- Think in automated workflows and loops rather than manual prompting
- Have built their own coding agents (even simple ones)
- Can explain technical details like how Cursor works under the hood
- Have developed custom Claude skills and prompt libraries with clear reasoning
Key Signal: The strongest indicator is whether a candidate has built their own agent and can explain how it works at a technical level - similar to how mechanics should be able to rebuild engines, not just drive cars.
Industry Shift: Founders envision running companies with 8-20 people using AI agents. The constraint in software development has fundamentally shifted.
Hiring Challenge: Existing hiring pools may need to be "wiped" because current interviewers lack the expertise to differentiate between these three categories.
Future Concerns:
- Dark agents (80% of enterprise agents run unmonitored)
- Need for agent governance and human ownership
- Organizations need internal "agent guilds" or centers of excellence
- High signal opportunities won't last - this knowledge will eventually become commoditized like leetcode