Every few weeks, something ships that would have read as science fiction in 2022.

A model clears an exam that used to gate an entire profession. An agent provisions its own cloud infrastructure, runs a job, and shuts it down without a human touching a terminal. A single founder pushes a product to production that would have needed nine engineers and two quarters.

The reaction usually splits into two camps, and both are lazy.

One camp says this changes everything and nothing will ever be the same. The other says it is a statistical parlour trick that will collapse under its own hype. Both camps get to feel clever without doing any work.

The useful question sits between them. Not whether AI is insane, because it plainly is. The question is what specifically changes for people who build products, sell them, and try to grow a business while the ground keeps moving.

I sell complex technical products for a living. Here is what I actually see.

What the craziness actually looks like right now

Strip out the discourse and three things are measurably true.

Capability keeps jumping in steps, not slopes. Progress does not arrive as a smooth line. It sits flat for months and then moves so far in one release that entire product categories stop making sense. Companies that raised money to solve a problem watch that problem become a feature in someone else's model.

The cost of intelligence is collapsing. The price to run a given level of capability has fallen by orders of magnitude in a very short window. Anything priced on the assumption that thinking is expensive is quietly being repriced. This is the part most business models have not absorbed yet.

Software started doing things instead of just answering. The shift from systems that respond to systems that act is the real threshold. A tool that answers a question needs a human in the loop. A tool that completes a multi step job needs a human only at the boundaries. That is a different economic object entirely.

None of that is speculation. It already happened.

Why the curve feels insane

Human beings are terrible at compounding. We reason in straight lines because straight lines described almost everything for most of history.

Exponential change breaks that instinct in a specific way. For a long stretch it looks like nothing is happening, and the sceptics look correct. Then it passes through the range where humans operate and appears to arrive all at once, and the sceptics look foolish overnight.

Nothing changed about the curve. Only the part of it we were standing on changed.

The mistake is not underestimating AI in the abstract. It is assuming next year will look roughly like this year, because every year before this one did.

This is why smart, experienced operators keep getting caught out. They are not stupid. They are extrapolating from a track record that no longer predicts anything.

What actually changes for businesses

Here is where the abstraction has to turn into decisions.

The cost of average work is collapsing

Competent, unremarkable output is approaching free. Average blog posts. Average landing pages. Average cold emails. Average code for a solved problem. If your business rests on producing acceptable work at volume, the floor is falling out.

Watch what that does to marketing. Content strategies built on publishing more than competitors are dead, because everyone can now publish infinitely. When supply goes to infinity, volume stops being a moat and starts being noise.

The same logic hits outbound. Personalisation at scale used to be a genuine advantage. Now every inbox receives a hundred messages that were personalised by a machine, which means the tactic that worked because it was rare has become the reason nobody replies.

Whenever a capability becomes universally available, it stops being an advantage and becomes table stakes. The advantage moves somewhere else.

Trust becomes the scarce asset

When anyone can generate anything, the constraint stops being production and becomes belief.

Buyers now assume a well written page might be machine generated. They assume the case study might be composed. They assume the review might be synthetic. That assumption is the single most underrated commercial fact of this decade.

What survives that suspicion is fairly narrow. Named people putting their reputation behind a claim. Specific numbers a company would not invent because they are checkable. Customers who will speak on record. Work you can point at and verify.

In other words, the things that are expensive precisely because they cannot be generated.

Where AI still falls apart

The honest section, because pretending otherwise helps nobody.

Models remain confidently wrong. They produce fluent, well structured, entirely fabricated answers, and fluency is exactly what makes the errors dangerous. A hesitant wrong answer gets checked. A polished wrong answer gets shipped.

They also lack the context that lives inside your company. A model does not know that the biggest account is quietly unhappy, that legal rejected that phrasing last year, or that the head of procurement responds badly to being rushed. That context is where most real decisions are actually made.

And they carry no accountability. When output goes wrong, a person absorbs the consequence. Responsibility has not been automated and will not be, which is why the last mile of anything serious still has a human name attached to it.

So the realistic picture is not a machine replacing a professional. It is a professional with unusual leverage, who still has to know what good looks like.

How to position yourself for the next three years

Practical, in order of how much they compound.

1. Build a distribution asset you own. An audience, an email list, a reputation in a specific niche. Production is commoditised, so attention is the bottleneck. Owned distribution is the one advantage that does not reset when the next model ships.

2. Get deliberately good at judgment. The scarce skill is no longer making the thing. It is knowing which version is right, which claim is defensible, and which idea should be killed. That comes from repetition and feedback, which means using these tools daily rather than reading about them.

3. Specialise until it feels uncomfortable. General knowledge is now free at high quality. Deep knowledge of a narrow domain, its politics, its buying process, and its failure modes is not, because it lives in practice rather than text.

4. Automate the boring middle, not the edges. Research, formatting, summarising, first drafts, data cleanup. Keep the first contact and the final judgment human. The edges are where trust is won or lost.

5. Shorten your planning horizon and lengthen your positioning. Three year tactical plans are now fiction. Three year positioning is not. Be clear about who you serve and what you are known for, then stay loose about the mechanics.

Questions people keep asking

Will AI replace sales and marketing jobs?

AI is replacing tasks rather than whole roles. The work that disappears first is the work that was already mechanical: list building, first draft copy, meeting notes, basic research. What survives is judgment, relationship building, and the ability to decide what is worth doing at all. The people at risk are not the ones who use AI badly. They are the ones whose entire contribution was the mechanical part.

What skills matter most in an AI driven market?

Three compound faster than everything else. Taste, meaning the ability to tell good output from plausible output. Distribution, meaning owned access to an audience that trusts you. And problem framing, meaning the ability to define what actually needs solving before anything gets built. All three are hard to automate because they depend on context the model does not have.

How fast should a small company adopt AI?

Faster than feels comfortable, but narrower than the hype suggests. Pick the two or three workflows where your team loses the most hours each week and automate those properly. Resist rebuilding the whole company around AI in one quarter. Companies that win here move quickly on a small surface area and keep a human accountable for anything a customer sees.

The part worth remembering

The craziness is not going to settle down. There is no version of the next few years where this becomes calm and predictable, and waiting for clarity is itself a decision with a cost.

But the fundamentals underneath are steadier than the headlines suggest. People still buy from people they trust. Attention is still earned rather than generated. Judgment still belongs to whoever is accountable for the outcome.

The tools got wildly more powerful. What makes them worth anything did not change.

If you are working out how to position a complex product while all of this moves, that is the work I do. Tell me what you are building.