AI By 2027: My 10 Predictions And How Certain I Am About Them

🔍 Read the full analysis: AI By 2027: My 10 Predictions And How Certain I Am About Them on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get audio and creator gear delivered free with Prime

• Fast, free delivery on millions of items
• Prime Video, Amazon Music and more included
• Member-only deals all year

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

Thorsten Meyer published 10 predictions for AI in 2027 on October 15, assigning each a confidence rating and describing evidence he says he will use to score it by the following October. The forecasts cover autonomous agents, AI-written code, verification, persistent memory, energy limits and research, but they remain Meyer’s expectations, not confirmed developments.

Meyer says his ratings average 72.5% and presents them as a way to make the forecast testable: if well calibrated, he expects roughly seven predictions to come true. He describes the piece as a forecast rather than a prophecy and says signals for scoring appear under each prediction. The source excerpt supplied here includes details for six of the 10 calls, so the remaining four predictions cannot be independently described from that material.

Among the listed forecasts, Meyer gives 85% confidence that AI assistants will remember users’ projects, people and decisions across conversations. He assigns 80% to the idea that verification will become a central workplace skill, as AI-generated answers become easier to produce, and another 80% to electricity, chips and cooling limiting AI capacity and influencing which services run locally on devices.

He rates the prediction that most new code will not be typed by people at 75%, arguing that developers will increasingly specify requirements and tests while agents generate implementations. He puts 70% on agents handling work across several days, and 65% on AI becoming a research partner that proposes approaches and runs simulations. These figures are subjective confidence ratings, not measured probabilities or results from a formal forecasting study.

How the Forecast Could Be Tested

The article matters as a scoreable forecast rather than a report of new AI capabilities. By attaching confidence levels and naming possible evidence, Meyer gives readers a framework for checking whether his claims hold up over the coming year. That makes the piece more accountable than a list of predictions without a stated way to assess them.

The forecasts also point to practical questions for workers, software teams and businesses: whether agents can complete long assignments without drifting, whether generated code can meet clear tests, and who controls the information stored in an assistant’s memory. Meyer’s arguments about power and computing capacity raise another issue: AI’s growth may depend not only on software improvements but also on infrastructure and cost. The article does not establish that these changes will occur; it sets out the author’s view of what to watch.

Meyer’s Case for Measurable Calls

Meyer’s introduction argues that a prediction without a number can be difficult to judge. He says a 70% rating means he would bet in favor but expects to lose that kind of bet about three times in 10; a 45% rating means he leans against the prediction. He uses the average of his ten ratings as a further calibration check, with roughly seven predictions expected to come true if the ratings are well calibrated.

For multi-day agents, Meyer identifies a specific concern: systems may hold a goal for hours but drift over several days, quietly working toward a different result. He says useful tools will need visible work logs and checkpoints. On AI-generated code, he argues that defining “done” and proving it through acceptance tests may matter more than typing the implementation. These are Meyer’s analysis and criteria; the source excerpt does not provide independent data confirming that most software teams already work this way.

For persistent AI memory, Meyer focuses on privacy and user control, including whether information can be inspected, edited, exported and stored where the user chooses. His infrastructure prediction links limited power, chips and cooling to a possible division of labor: routine tasks handled on laptops or phones, with larger data centers reserved for more demanding work. The article also says he made a short video covering all 10 calls.

“A forecast, not a prophecy.”

— Thorsten Meyer

What the October 2027 Check Must Show

The predictions have not yet been tested against their stated deadline. The excerpt gives details for only six of the ten calls and does not include the full infographic signals Meyer says will be used to score them. It also does not specify how ambiguous outcomes—such as what counts as “most” new code being machine-written or an agent completing a multi-day task—will be measured.

The confidence figures are Meyer’s own judgments. The supplied material provides no external validation of their calibration and no independent evidence for the broad claims about current software teams, AI infrastructure or assistant memory. It is also unclear how Meyer will treat partial progress, changes in product availability, or forecasts that come true only in some sectors. Those limits matter when comparing the final score with the stated 72.5% average confidence.

The October 2027 Scorecard

Meyer says readers can use the signals listed beneath each prediction to assess the calls by October 2027. A meaningful review will require those criteria to be specific enough to distinguish a widely adopted capability from a limited trial or demonstration. The supplied excerpt does not state whether Meyer plans to publish a formal scorecard or how he will resolve borderline results.

Until that review, the forecasts should be read as a set of claims to monitor, not as a description of what AI will certainly do. The next evidence will come from whether agents take on longer tasks in ordinary workflows, whether teams rely more on tests and proof, and whether companies make persistent memory and local computing practical buying considerations.

Key Questions

Who made the 2027 AI predictions?

Thorsten Meyer published the forecast on ThorstenMeyerAI.com on October 15, 2026.

How many predictions are included?

The article presents 10 predictions. The source material provided here describes six of them in detail.

What is Meyer’s average confidence rating?

Meyer says the ten ratings average 72.5%. He says that, if well calibrated, this would imply about seven predictions come true; it is his stated expectation, not a verified result.

Which prediction received the highest confidence rating in the excerpt?

Meyer gives 85% confidence to his prediction that AI assistants will retain information about users’ work across conversations.

When will the predictions be assessed?

Meyer says he will use signals associated with the predictions to score them by October 2027. The supplied material does not say whether he will publish a formal scorecard.

Source: ThorstenMeyerAI.com