[Me & X] AI predictions
X : Can you give me an interesting prediction for AI?
Me : No. Most predictions are either already happening, surprisingly dull or wild guesses.
X : Examples?
Me : Already happening or dull? Try rapid growth (punctuated equilibrium), no choice over it (Red Queen), prompt engineering (co-evolution of practice), vibe coding (conversational programming), endless unforced errors such as getting rid of software engineers just as demand for them will go through the roof (Jevons paradox and the new activities created), the inevitable march to the one person $1 billion company, failure to adapt (inertia), machine psychologists (new jobs and new activities).
X : What about an exciting prediction?
Me : Unpredictable, that's in the wild guess area. We won't clearly know who the winners / losers are until we hit about 5-8 years after industrialisation kicks in.
X : No predictions at all?
Me : How about this, I suspect there will be some very high profile failures in the next few years as companies use AI to try and migrate their legacy. CIOs have been proclaiming they don't need software engineers and their AI will re-write their legacy estate. Those projects will start failing spectacularly when they hit production environments. Those CIOs will be prime candidates for public failure.
X : Any interesting things?
Me : I like the agentic AI stuff, the building applications with swarms of agents. We are going to see 30M LoC systems converted by an agentic swarm (with coding, testing, architecture) which no-one understands and then placed into production.
X : You think that's dangerous?
Me : I think it is statistically impossible that these systems build correctly and people need to think of imperfection as your friend. Think biology and antibodies. Success requires embracing imperfection, not pretending it doesn't exist.
Originally published on LinkedIn.