– context management: Modern harnesses and models have sufficient mechanisms to auto-compact
– Prompt engineering: the former „engineer’s skill of the future“ is deprecated already. Nowadays you can give a model a one-liner and it will recreate Photoshop for you, with all features and more complete than you could have ever described it.
– sub-agent harnesses: Codex, Claude Code, etc. are automatically spawning sub-agents without the need for extensive ecosystem
– loops/goals: seem not to be necessary anymore
– Agents.md: waste of tokens? See the paper https://arxiv.org/pdf/2602.11988
Your opinion?
Edit (Oct 16th 2026): The same post on LinkedIn gathered a lot of good comments: https://www.linkedin.com/feed/update/urn:li:activity:7517578946054447104/. (sidenote: It was very interesting, the post went viral for a few hours with many impressions and people commenting in like every other minute. And suddenly it completely stopped.)
Edit (Nov 1st 2026): Stumbled over a comment by Andrej Karpathy, where he describes similar thing:
My prompts are getting shorter and dumber because the model „gets“ what I’m doing even with fewer words. It’s almost to the point where if you give it too much, you’re constraining the model with too many requirements, when the thing it would have done otherwise was better/smarter. So I feel like the prompting meta is moving more towards fewer bits administered just right. I go back and forth on whether I should make a video on it because on the one hand people have these questions but on the other my best practices look so dumb it’s almost embarassing.