I love a confident AI prediction as much as the next person doom-scrolling LinkedIn. "Every business will run on agents by 2026." "Nobody will write code again." "Enterprises will never adopt this." Pick your flavor. They can't all be right, and most are going to age like milk.

What actually bugs me: people are making macro predictions about AI using the same mental models they used for every other tech wave. SaaS adoption curves. Cloud migration timelines. Mobile rollout. None of that applies when the tools themselves change weekly. We're now saying "remember six months ago?" the way people used to talk about the early internet. That should tell you something about how broken the old framework is.

The Only Skill That Matters Right Now

I started using these tools early, back when the basic models felt like magic. Six months later I would not have guessed they'd be this capable. Planning cycles that used to take a quarter now take an afternoon. Dev work that used to take weeks happens in hours. If your prediction framework was built on the old cadence, throw it out.

The best career advice I ever got, way before any of this: stay curious and keep learning. That's the whole strategy. What you know today has a shelf life measured in months. What matters is whether you'll actually open the tool tomorrow and try something.

Stop Asking "How Do I Use AI"

The worst way to learn these tools is to sit down and try to "learn AI." I've watched people drop money on intensives, follow influencers, take notes on prompt frameworks, and ship nothing. Then they tell me the tools are overhyped. Sure.

Wrong question. Try this: what's a real problem you actually want to solve, and can AI help? That reframes everything. You stop trying to build a demo and start trying to fix your own life. Organize your inbox. Draft the thing you've been putting off. Build the small internal tool nobody at your old job would have prioritized.

The other thing nobody admits: fail fast, and use the tool to debug itself. I spent hours one weekend reworking a system because I'd accepted the first solution the model gave me. It did exactly what I asked. It also didn't fit the framework I was building. That was my fault, not the model's, and the only way I figured it out was by feeding the mess back in and asking what was wrong. The learning was in the breakage.

The Confident Futurists Are The Tell

Here's my contrarian take: the more confident someone sounds about where AI is going in two years, the less you should trust them. Nobody knows. Anyone selling you a definitive future is selling you something else underneath it (usually a course, a consulting package, or a personal brand).

What we can say honestly: the tools have real limits today, and they will change. The people who lose their jobs to AI won't lose them to AI in the abstract. They'll lose them to the coworker two desks over who actually opened the thing on a Tuesday night and figured out what it could and couldn't do.

My real complaint: I'm tired of the prediction industrial complex. Everyone's certainty has gotten weirder, and meanwhile the actually useful move is so unsexy it doesn't trend. Pick a problem you care about. Open the tool. Break it. Ask it why. That's the curriculum.

If someone tells you they know what work looks like in 2027, ask them what they got wrong six months ago. If they can't answer, you have your answer.