Learn AI now

I’ve been doing this for twenty-three years, and for most of them “new technology” meant the same ideas in a new coat. The same thing, now in the cloud. The same thing, now managed by someone else. The same scheduled job you set up and forgot about, now with a nicer dashboard. Useful now and then. Genuinely new, almost never. AI is the first thing in my whole career that made me stop and say wow.

I don’t mean machine learning, which has been running in the background for years. I mean what we have now. A model does most of my personal technical work today: it drives my browser, writes firewall rules, stands up EC2 instances, and spins up small working apps when I want one for myself. No agents to wire up, no prompts to engineer. I tell it to do something, and it digs in, learns what it needs, and gets it done. Not a demo, not a someday, today, on my own machines. The first time you watch it do in two minutes what would have taken you weeks, or never happened at all, the penny drops.

Step back from my own desk and it only gets bigger. These systems are already taking on work that used to need people. Governments are writing reports about frontier models as an extinction-level risk and locking down the chips behind them like munitions. There are demos where a model hallucinates an entire working operating system on the fly, windows and menus and files, none of it real and all of it functioning. Whatever you think the ceiling is, it is higher than that.

The easy out is to wave it off over the costs. It is power-hungry, it drinks water by the datacenter, and it is driving up the price of everything from enterprise hard drives to the card in a gaming PC. All true. All solvable, and the fixes are already being drawn up: small modular reactors for power, datacenters headed for orbit where the cooling is the cold of space, the results beamed back over satellite constellations. These are engineering problems, and engineering problems get solved. Betting against that is how you end up left behind.

The shape of the future rhymes with the past. We spent two decades moving computing off our own machines onto someone else’s, and this is that same move taken to its conclusion. Instead of a console streaming a game to your television, a phone or a thin client on the wall streams you a hallucinated version of whatever you want, generated in real time, as much of it as you ask for. The heavy lifting happens elsewhere. You just get the result. The screen itself may be the next to go. Neuralink already has more than twenty people controlling computers with their thoughts, and it named its first product Telepathy with a straight face. For now that’s reading intent out; piping a signal back in is the harder half, and it’s what their Blindsight project is chasing. Wire two of them together and telepathy stops being a brand name and becomes the literal thing. Give it ten years.

That still sounds like science fiction, but so did most of this a few years ago. The good science fiction has been mapping these directions for a while: Daniel H. Wilson’s neural implants in Amped, his machine uprising in Robopocalypse, the runaway AIs of Dan Simmons’s Hyperion Cantos. The wonderful and the alarming are both on the table, which is the real reason to understand this rather than watch from the stands.

That is why sitting it out is not really an option, whoever you are. If you are hands-on in security or IT, this is about to sit under everything you defend and run, and you cannot protect or operate what you do not understand. If you lead, you will be the one deciding what to adopt, what to spend, and what to be wary of, and you cannot make those calls from the outside. The people who learn it now will spend the next decade ahead. The ones who wait may spend it wondering why the work keeps going to someone else. The time to learn AI is now. It really is the technology of the future.