More than a decade of working with AI, and what our 2017 forecast got right

We have been thinking about and working with AI for more than a decade. In February 2017, one of our Partners in London argued in Quirk’s that nearly every technology trend of the year came back to one: artificial intelligence moving from following simple commands to taking on whole tasks for people. The piece first ran on our own website as “One tech trend to rule them all.”

Nine years on, that call has held up. Handing a whole task to AI is now an everyday product, and driverless cars now give hundreds of thousands of paid rides a week. Looking back at it also shows which parts of any forecast deserve a team’s trust.

What we argued in 2017

The case was simple. People want technology to take friction out of their lives, and the step after asking an assistant for the weather was handing it a whole job: buying the family’s presents, say, within a budget it had learned from past spending.

The piece pointed to where that was already starting. Voice assistants were getting better. Self-driving cars had begun carrying passengers. AI was supporting legal and medical work, and writing music and news. Easier AI tools would bring a wave of startups, and the biggest technology companies already treated a top AI team as essential.

What came true

Delegation is now something people buy. OpenAI describes its ChatGPT agent, launched in July 2025, as able to “do work for you using its own computer, handling complex tasks from start to finish,” and it asks permission before anything consequential.

Driverless passenger service has arrived too. By March 2026, Waymo was providing 500,000 paid robotaxi rides a week across 10 US cities, ten times its weekly total in May 2024.

Reading this year’s forecasts

A forecast’s calls about a capability tend to hold up better than its picks of who will lead. What came true here was the capability and its direction; the companies that delivered it were often not the ones forecasts pointed to at the time. Delegation arrived through large language models, which broke through only after 2017, more than through the voice assistants of the day. Uber was the name attached to self-driving taxis then; the rides now counted in the hundreds of thousands a week are Waymo’s. IBM’s Watson was the name attached to AI in medicine, and IBM sold the Watson Health business in 2022.

That is how we read forecasts in our trends and foresight work. The capability and its timing are the claim to test. The named leaders are examples of who might deliver it, and they are the part of any trend report most likely to change. A team planning around a forecast does better to build for the capability, and to keep its choice of partners and platforms open for as long as it can.

Our trends and foresight work for clients includes a services vision for a consumer technology company’s devices, built from basic human needs; a map of the forces of digital change for a sportswear brand, with what each could mean for the business; and future food models for a foodservice business, each one sized.

What it means for our own work

The 2017 piece said little about research itself, but by then we had been using AI in our work for some time. By 2015 we were using early AI text-analysis tools such as Luminoso to analyze large data sets for clients. They were precursors of today’s large language models. We have worked with many tools since, and we keep experimenting with the newest technology as it arrives.

Today AI can read a whole body of interviews and draft a confident summary of them. The question we keep asking is what happens to the evidence when it does, and whether the contradictions, the unusual voices and the context survive. We set out that question in What gets lost in a convincing AI answer?

Our answer is on AI in our work. We use AI as we did then, to look further and check harder. These days we build our own AI agents, and people decide what the evidence means.

The original: “The rise of artificial intelligence”, Quirk’s, February 10, 2017, an edited version of our post “One tech trend to rule them all.”

Common questions

How long has antedote been working with AI?

For more than a decade. By 2015 we were using early AI text-analysis tools such as Luminoso, precursors of today’s large language models, to analyze large data sets for clients, and we have kept experimenting with newer tools ever since. In 2017 we argued in Quirk’s that AI would move from following commands to taking on whole tasks for people.

What did people predict about AI in 2017?

In a 2017 article in Quirk’s, we argued that AI would move from following simple commands to taking on whole tasks for people. We pointed to voice assistants, self-driving cars carrying passengers, AI supporting legal and medical work, AI writing music and news, a wave of startups, and the biggest technology companies treating a top AI team as essential.

What does a 2017 AI forecast teach about reading trend reports?

Calls about a capability and its timing tend to hold up better than calls about which company will deliver it. Read a forecast for the capability, and treat the named leaders as examples of who might deliver it.

How should a business use a technology trend forecast?

Treat the capability and its timing as the claim to test, and the companies a forecast names as examples. Build for the capability, and keep the choice of partners and platforms open for as long as you can.