To be honest, I doubt the current state of things will hold much longer. Current AI can automate virtually all of my work as a pre-AI senior and is rapidly closing in on being able to automate the types of work that have emerged post-AI too. I have a hard time imagining a world where a drop-in end-to-end replacement for a full software team is more than a year away. I'm sure I'll have a job for a year or two past that, but there'll be some small stock market bump eventually and a bloodbath will begin the second the C-suite takes a closer look at what their engineers are even doing anymore.

https://nitter.cf/shaharglazner/article/2119796029256032278

A readable link for those of us without Twitter/X account

If you can’t make things right when your agent goes wrong, then yeah you’re not adding value. If you can’t tell the difference between the agent hallucinations and the useful stuff, then yeah you’re not adding value. If you can’t steer the agent in a useful direction instead of just standing back, then yeah…

At first I thought tacit knowledge would be a differentiator: LLMs are trained on writing, so they can’t learn things that have never been written down. But tacit knowledge is exactly what reinforcement learning post-training is about. So (just as with humans) the limitation is in providing a rich enough environment, with good enough feedback, to train in the tacit knowledge.

I'm a fan of the 'sandwich' model: someone needs to say what to do, someone needs to do it, and someone needs to say if it's been done [well enough]. Even if AI can do that middle bit entirely, we still need humans at the beginning and the end. Replace the humans at the start and you have a paperclip-optimizer; replace them at the end and you have slop.

The problem is that software in corporations has typically been product managers at the front, software engineers in the middle, and QA at the end. Software developers who also have good product sense and do diligent QA are still needed. Programmers who treated their job as leetcode exercises are not needed anymore.

I don't buy the reductio ad absurdum, which hinges on the assertion that some developers are still 10x more productive than others due to differences in their engineering potential. I think that A) however you're measuring productivity is probably not capturing the full story, and B) one of the most important reasons people have differences in productivity is because not everyone puts the same amount of effort in. It's not necessarily because they don't have the ability to stay off of reddit and work longer hours and prompt more furiously, it's because not everyone is equally motivated to do so (hence the entire field of management).

I also don't buy the idea that there is special human stardust that gives us permanent irreplaceable economic value. Even if the tacit knowledge that the current generation of engineers were able to build up in the years prior to ~2025 does give us value that won't be completely replaced for a long time, why would you assume that the delta between what you can do and what claude can do will still be worth $200K in 5 years?

On Being an [Software] Engineer (i.e. AI agent manager) in 2026

Oh he means programmers. I thought he meant real engineers.

The trick is to remain intellectually stimulated and challenged.

> Maybe one day agents will have their own experiences, too. Maybe they'll develop intuitions they can't quite explain, just like we do.

Machine / program / software cannot have these things in my opinion because this stuff is produced by human consciousness which we cannot even fully explain yet. Someone has "to teach" the agent to have this stuff, but we don't have that knowledge. And also "intuition" is a feeling (like love, hate, sadness, joy) so no agent can have that until they have the limbic system, prefrontal cortex and other stuff.