What I Think 2027 Will Look Like: 10 AI Predictions, Each With A Confidence Level

🔍 Read the full analysis: What I Think 2027 Will Look Like: 10 AI Predictions, Each With A Confidence Level on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the latest gadgets 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 a forecast for 2027 on October 15, 2026, listing 10 predictions about how artificial intelligence may change work, software development, research and computing infrastructure. Meyer assigns a confidence level to each forecast and says he will judge them against specified signals by October 2027; the source material available for this article describes only six of the 10 predictions.

Meyer says the 10 confidence ratings average 72.5%. He presents that figure as a way to assess his calibration: if his forecasts are well calibrated, he expects roughly seven to prove accurate by October 2027. That is the author’s proposed scorecard, not an independently verified prediction or an assessment of AI adoption.

The six predictions described in the available text concern agents handling multi-day assignments, AI producing most new code, increased demand for verification, assistants retaining work-related memory, constraints from electricity and computing capacity, and AI acting as a research partner. Their stated confidence levels range from 65% to 85%. Meyer also describes a possible move toward more routine AI tasks running on phones and laptops, while data centers handle more demanding work.

For each of those six calls, Meyer gives a condition that could count against the prediction. Examples include long-running agents remaining mainly demonstrations, production code still being written line by line by people, and mainstream assistants continuing to forget information between sessions. The source refers to evidence beneath each prediction, but the supplied material does not include the referenced infographics or the complete set of 10 forecasts.

How Meyer Plans to Score AI

The forecast matters as a dated set of claims that can be revisited, rather than as evidence that the predicted changes have already happened. Meyer’s confidence percentages and stated failure conditions give readers a way to compare his outlook with developments during 2027. They also make clear that the article is an individual forecast, not a consensus view from researchers or companies.

The predictions point to practical questions for workers and organizations: whether agents can handle longer assignments reliably, whether software teams shift from writing code toward defining and checking requirements, and whether AI systems can provide dependable evidence for their output. Meyer’s infrastructure forecast raises a separate issue: the cost and availability of power, chips and cooling could shape where AI runs and which services are available. These are the author’s expectations; the source does not establish that any of these outcomes will occur.

For readers evaluating the forecast, the distinction between confidence and certainty is central. Meyer says a 70% rating still allows for a substantial chance of being wrong. The average score is his own measure of accountability and does not, by itself, show that past forecasts were accurate.

The Six Forecasts Available

Meyer frames his piece as a forecast, not a prophecy, and argues that predictions should be stated in ways that can be tested. He says he will revisit the calls by October 2027, using signals associated with each prediction. The supplied text gives confidence ratings and examples of what could disprove six calls, but it does not reproduce all of the promised scoring signals.

The six available ratings are 70% for delegating week-long tasks to agents, 75% for most new code no longer being typed by people, and 80% for evidence becoming more important than an output that merely “should work.” Meyer assigns 85% to assistants remembering a user’s work across conversations, 80% to power and computing capacity limiting AI deployment, and 65% to AI becoming a research partner.

Several of the calls include qualifications. Meyer says agents may lose track of a goal over multiple days, that persistent memory raises questions about data control, and that generating more research ideas would not eliminate the need to test and judge them. The source also points to a short video accompanying the forecast, but the video’s contents are not independently established by the text provided here.

“A forecast, not a prophecy.”

— Thorsten Meyer

What the Forecast Leaves Open

The supplied source text ends partway through the sixth prediction and does not include predictions seven through 10. Their topics, confidence ratings and proposed scoring signals are therefore not available here, and the stated average of 72.5% cannot be independently checked from the six forecasts provided. The source also refers to infographics that are not included in the material.

More broadly, none of the 2027 outcomes is confirmed. Meyer offers his own estimates, not measured probabilities from a formal forecasting study. The source does not provide supporting market data, an independent review of the claims, or definitions for every measure of success—for example, what share of production code would qualify as “most” or how widespread agent use would need to be to count as a default.

It is also unclear how Meyer will account for changes in the technology during the scoring period, or who will verify the results. Questions raised by the predictions, including control over stored AI memory and the scale of power constraints, remain open in the supplied material.

The October 2027 Scorecard

Meyer says he will use evidence associated with each prediction to judge the results by October 2027. That review would allow readers to compare his stated confidence levels with what happened, although the source material does not explain whether the scoring will be conducted independently or how borderline outcomes will be treated.

Before then, the full forecast would be needed to assess all 10 calls and verify the reported 72.5% average. Readers can also watch for the specific indicators Meyer names: whether agents take on longer work with visible checkpoints, whether software teams rely more on machine-written code and testable specifications, whether assistants retain useful project context, and whether local devices handle more routine AI tasks. These remain measures proposed by the author, not confirmed milestones.

Key Questions

Who made the 2027 AI predictions?

Thorsten Meyer published the forecast on his website on October 15, 2026. The predictions are his own outlook, not an industry consensus.

How many predictions are described in the available source?

The source material supplied here describes six of the 10 predictions. It does not provide the remaining four, so their content and confidence ratings cannot be confirmed from this text.

What does Meyer mean by a confidence level?

Meyer presents the percentage as his estimate of the likelihood that a prediction will come true. He says a 70% confidence level means he would bet on the prediction while expecting to lose that kind of bet about three times in 10.

When does Meyer plan to assess the predictions?

He says the forecasts will be scored by October 2027, using evidence or signals tied to each prediction. The supplied text does not include all the referenced scoring materials.

Are the predicted changes already confirmed?

No. The article presents forecasts for 2027, not confirmed developments. Meyer describes possible outcomes and conditions that could show some predictions were wrong.

Source: ThorstenMeyerAI.com