TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
A 2026 retrospective on Taylor Town’s review of Kevin Kelly’s 2016 AI forecasts revisits predictions about Google, cloud-based AI, specialist software and conversational systems. Some themes now look familiar, but the source does not provide a systematic scorecard or establish that every prediction came true.
A 2026 retrospective on Kevin Kelly’s 2016 AI predictions revisits his forecasts that AI would become Google’s main product, that a few cloud providers could dominate general-purpose intelligence, and that most AI would remain specialized software. The review, published by Taylor Town, frames the decade-old passages against developments since the book appeared, but does not offer a full evidence-based audit of which forecasts have been fulfilled.
Kelly’s prediction about Google was unusually direct: “By 2026, Google’s main product will not be search but AI.” In the passage quoted by Taylor Town, he argued that search activity could help train Google’s AI, reversing the conventional view that the company would primarily use AI to improve search. The retrospective notes that Google researchers published Attention Is All You Need in 2017, a paper that introduced the transformer architecture, but that fact alone does not establish whether Kelly’s broader prediction about Google’s main product has come true.
Other excerpts focus on how AI might be delivered and who might control it. Kelly anticipated an “oligarchy of two or three” large, general-purpose commercial intelligences, while comparing the future of AI infrastructure to Amazon Web Services: cheap, reliable computing that runs largely out of sight. He also argued that AI should ideally be cheap or free, describing access to intelligence as a shared utility. Taylor Town presents these passages as relevant to current debates about cloud-based AI and concentration, rather than documenting market shares or proving that only a few providers will prevail.
Kelly also forecast that most AI would be special-purpose software, such as systems that translate languages or perform a narrow task without human-like conversation. Alongside that, he imagined people asking cloud-based systems questions in ordinary conversational language and receiving explanations when answers were known. The retrospective groups these predictions as distinct strands: specialized capabilities, conversational interfaces and large-scale infrastructure. It does not supply quantitative measures for the forecast that 99 percent of AI interactions would involve narrow specialists.
The Forecasts Behind Today’s AI Debate
The retrospective matters because Kelly’s predictions were not limited to a single product launch or technical milestone. They described a possible structure for the AI economy: intelligence delivered through cloud infrastructure, concentrated among a small number of providers, with specialist systems handling much of the work. Revisiting that framework helps readers distinguish forecasts about how AI is built and distributed from claims about machines becoming human-like.
It also puts the tension between concentrated commercial services and open access into view. Kelly anticipated powerful platform companies while arguing that AI should be inexpensive or freely available, a goal that Taylor Town says he advocated in the book. Those are not the same prediction: the first concerns who might control infrastructure; the second is a preference for broad access. The source does not establish whether that ideal has been achieved.
For readers evaluating claims about AI’s trajectory, the exercise is useful only if predictions are kept precise. A conversational interface does not by itself show that an AI can reliably answer any question, and the use of AI across a company’s products does not prove that AI has replaced search as its main product. The retrospective invites comparison, but the evidence presented is selective rather than a complete assessment.
As an affiliate, we earn on qualifying purchases.
What Kelly Wrote in 2016
The material reviewed by Taylor Town comes from The Inevitable, Kelly’s book about broad technological trends. The source quotes passages on Google, commercial cloud intelligence, narrow AI applications and conversational question-answering. Kelly’s Google forecast drew on an anecdote he recounted about speaking with co-founder Larry Page around 2002, before Google’s initial public offering. According to the book passage, Page told him, “Oh, we’re really making an AI.”
Kelly connected that exchange to his view that search queries and user interactions could contribute to training AI. He also cited Google’s acquisition of AI and robotics companies and comments by CEO Sundar Pichai about applying AI across Google products. These are details from the 2016 book excerpt reproduced by Taylor Town, not a new statement from Google. The later milestone noted in the retrospective is the 2017 publication of the transformer paper by Google researchers.
The review’s title, “Kevin Kelly’s 2016 AI Predictions, 10 Years Later,” marks a decade between the book’s forecasts and the retrospective. Its method is to revisit selected passages, not to report a formal study, interview Kelly or score each forecast against predefined criteria. That distinction matters when treating the article as evidence about what has changed.
“By 2026, Google’s main product will not be search but AI.”
— Kevin Kelly, as quoted from The Inevitable
As an affiliate, we earn on qualifying purchases.
Which Predictions Hold Up?
The retrospective does not establish whether AI is now Google’s main product, how to define “main product,” or what measure should be used to judge the prediction. Google’s use of AI across services and the 2017 transformer paper are relevant developments, but they are not, on their own, proof that AI has displaced search in the company’s business or identity.
It is also unclear from the source whether two or three providers dominate general-purpose commercial AI, how the forecast’s “99 percent” figure should be measured, or whether narrow systems account for that share of people’s direct and indirect interactions. The review supplies no current market data, usage figures or detailed comparison of outcomes. Its selected excerpts support a retrospective discussion, not a conclusive verdict on Kelly’s accuracy.
As an affiliate, we earn on qualifying purchases.
A Fuller Scorecard Remains
The next step for a more conclusive assessment would be to compare each forecast with clearly defined 2026 evidence: the role of AI in Google’s products and revenue, the number and market position of major AI providers, and the balance between specialist systems and general-purpose assistants. The Taylor Town piece, as provided, does not set out that scorecard or announce further reporting.
For now, the development is a look back at a set of influential forecasts, not a new finding about the state of the AI industry. Whether Kelly’s predictions have come true in full remains open to examination, particularly where the terms “main product,” “bulk of AI” and “oligarchy” need explicit definitions.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Kevin Kelly predict about Google?
Kelly predicted in The Inevitable that “By 2026, Google’s main product will not be search but AI.” The retrospective revisits the forecast but does not conclusively determine whether it has been fulfilled.
What is the 2026 article about?
Taylor Town’s article revisits selected AI-related passages from Kelly’s 2016 book, including forecasts about Google, cloud infrastructure, specialized software and conversational question-answering.
Did Kelly predict that a few companies would control AI?
Yes. The excerpts say he expected an “oligarchy of two or three” large, general-purpose commercial intelligences. The retrospective does not provide market data to show whether that forecast describes the 2026 market.
Did Kelly expect AI to be human-like?
Not as the main form of AI. He wrote that most AI would be special-purpose software, while also predicting that people would ask cloud systems questions in conversational language. A conversational interface and human-like intelligence are different claims.
Does the retrospective prove the forecasts were right?
No. It revisits selected passages and notes developments such as Google researchers’ 2017 transformer paper, but it does not provide a full audit, defined benchmarks or comprehensive current market evidence.
Source: hn
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
