Why I avoid prognosticating on the future of AI

I have a rule that I avoid prognosticating on the future of technology and AI. But let me break that rule briefly. 😁

The pace of improvement in LLMs — for real-world application in wealth firms — has slowed. The leap to GPT-3.5 was seismic. Since then, the progress has been incremental.

GPT-4o was a big step in agentic ability. 4.1 added useful gains in voice interaction. But for AI-powered workflows and insights in advice, is GPT-5 really better? Not in any meaningful way.

You can already create a modular AI workflow — say a client-file compliance check — with 4o or 4.1 and get powerful, practical results. And 4o is 18 months old now. 😳

Yes, Anthropic, Google and others have progressed too. But the same pattern holds: for everyday use, big leaps are rarer, refinements more common.

The point is, don’t assume a miraculous new LLM is just around the corner. More likely, we’ll see gradual evolution of the models we already have.

And that’s actually good news. It should encourage firms to innovate now — by deploying proven AI models in nimble, modular ways. Because for discrete tasks in wealth management, today’s models are powerful, accessible and affordable.

So I don’t waste time obsessing over what’s next — I’d rather take advantage of the opportunity sitting in front of us today. 🙂