I remember some time in 2023, going out for dinner with some coworkers (a mixture of software developers, PMs, business analysts), and speaking to the devs, telling them that in five years their job would look completely different, and that they’d no longer write most of their code by hand. They smirked and gave me a “yeah right” look. I’ve been largely correct in my prediction (though much earlier than 5 years). Sadly, I still see people today who are clinging on to a world that no longer exists.
Many of the observations and predictions below are fairly obvious (to me at least), with some having become more obvious since the release of Opus 4.5 in November 2025.
Observations
- Your technical blog is largely dead. Prepackaged answers are obsolete. I still see people writing tutorials in 2026. Competent exposition of a topic does not compete with an LLM that can give a highly customized explanation that is fitted to the individual. Knowledge is commoditized.
- Human output isn’t worthless, but certain forms are definitely substitutable, so a human merely supplying some (e.g. technical) output commands less of a premium (and less of your attention).
- Implementation is now separate from engineering, and knowledge is separate from capability.
- The important skill is communication and orchestration.
- For economic purposes, the question becomes “what is the output of a capable person working with models versus a person using the old approaches and workflows?”
- One that has been repeated ad nauseam but is still very true: taste is a differentiator when production becomes abundant.
- Problem formulation has become disproportionately important.
- Verification has become a larger part of the job. I don’t believe that AI code is necessarily bad (and I also think that views on what good code vs bad code looks like needs to change), but if production time collapses, more time should be spent verifying systems work as specified.
- There are psychological injuries taking place among software developers (and more recently, mathematicians): status loss, and epistemic specialness. The market around their intelligence has changed, and their ability to see what others can’t (e.g. some obscure compiler behavior) is no longer special to them—LLMs do this just as well and better.
Predictions
- As GPUs and memory became expensive and scarce, so will robotics parts.
- LLMs will become good enough to drive/pilot a robot (everything goes into the transformer hole).