It speaks, and it has to know the rules

We are at the start of a new kind of industrial revolution. Not the steam and steel kind. This one changes how we deal with the industries we already live inside: how we buy insurance, see a doctor, book a flight, open a bank account.

Two things are happening at the same time, and I think the overlap is what matters. Software is getting cheap to make. And the way we use software is moving from typing and tapping to simply talking.

Let me take them one at a time.

Software is getting cheap, in a good way

When I say cheap I don't mean worse. I mean the cost of trying something has dropped. A team can put an idea in front of real people in days instead of a quarter, learn it was wrong, and try the next one before the old budget meeting would even have happened.

Developers have noticed. In Stack Overflow's 2025 survey, 84% of developers said they use or plan to use AI tools in their work.

It is not fully there yet, and the honest evidence says so. In early 2025 the research group METR ran a careful trial with experienced open source developers working on their own projects. With AI tools they were 19% slower, even though they believed they had been faster. The same Stack Overflow survey found that 66% of developers were frustrated by answers that are "almost right, but not quite."

But look at the direction. By February 2026, METR said it believes developers are likely sped up by AI tools now, and that its new numbers probably understate how much. We are getting there, and quickly.

And now it speaks

If you haven't watched Halt and Catch Fire, you should. Late in the first season, Joe MacMillan, who has bet everything on building a PC, wanders into a room at COMDEX where Apple is showing off the Macintosh. The machine says hello in its flat, mechanical voice, and Joe, stunned, says two words: "It speaks."

The show took a small liberty with the calendar. The real Macintosh first spoke on 24 January 1984, when Steve Jobs pulled it out of a bag in front of a few thousand people and let it introduce itself: "Hello, I'm Macintosh. It sure is great to get out of that bag." The room went wild over a computer reading a script.

Forty years later it doesn't read a script. It listens, it answers, it keeps up. When OpenAI introduced GPT-4o in 2024, it could already respond to speech in about 320 milliseconds on average, roughly the pause between two people in conversation. You don't hunt for the right menu. You say what you want and it does it.

All those Tony Stark moments with Jarvis that looked like pure fantasy the first time I watched them? They are turning into product roadmaps.

So what does this have to do with industries?

Software has been creeping into every industry for decades. Forms moved online, paper became PDFs, call centres got phone trees. Put cheap software and natural conversation together and it stops creeping. It walks straight into the middle of how the work gets done.

That is where it gets risky, because an industry is not a website. It is workflows, processes, laws, rules, regulations and a long list of variables nobody ever wrote down.

The trick is knowing the industry in and out.

Here is what happens when you skip that part. In 2022 a passenger asked Air Canada's website chatbot about bereavement fares after a death in the family. The bot told them they could buy a regular ticket and claim the discount afterwards. The airline's actual policy said otherwise. In February 2024, British Columbia's Civil Resolution Tribunal held Air Canada responsible for what its chatbot said, and rejected the idea that the bot was somehow responsible for itself.

The money was small, about $812. The lesson wasn't. The technology did everything we ask of it. It talked, it was friendly, it answered instantly. It just didn't know the rules of the business it was speaking for.

Now the other side. The Permanente Medical Group rolled out AI scribes to its doctors: the tool listens to the visit and drafts the clinical note. Over one year and more than 2.5 million patient visits, it saved an estimated 15,791 hours of documentation, and doctors reported better conversations with their patients. The important part is that it was built around how clinical work already happens. The doctor still reads the note and can edit it for accuracy before it counts.

Same kind of technology. One was dropped on top of an industry. The other was designed from inside it.

Where the work is now

I have spent my career designing for regulated places: banking, insurance, wealth management, export finance. What those years taught me is that the hard part was never the screen. It was the disclosure a lawyer insisted on, the approval step nobody could explain but everyone depended on, the rule that changes from one province or state to the next.

That knowledge used to be the slow, unglamorous part of the job. It is becoming the whole job. When building is cheap and the interface is a conversation, the people who win are the ones who can map an industry end to end: every workflow, every rule, every exception. Then they design something that speaks, and knows what it is allowed to say.