10 Things To Watch In AI By 2027, With Confidence Ratings

🔍 Read the full analysis: 10 Things To Watch In AI By 2027, With Confidence Ratings 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 10 predictions for AI in 2027 on October 15, assigning each a confidence rating and describing evidence he says he will use to judge whether the forecast came true. The list argues that AI use could shift from one-prompt interactions toward delegated work, while presenting these developments as forecasts, not confirmed outcomes.

Meyer says his ratings average 72.5%, and frames that figure as a way to hold himself accountable: if the ratings are well calibrated, about seven predictions should be borne out by October 2027. He also identifies conditions that would count against each call. The average is his own summary of the ratings, not an independent probability assessment.

The predictions include multi-day AI agents, AI-generated code becoming the majority of new code, and a greater emphasis on checking AI output with tests and other evidence. Meyer assigns these calls confidence ratings of 70%, 75% and 80%, respectively. He expects AI memory that carries information between sessions to become a purchasing consideration, rating that prediction 85%.

Other forecasts address constraints and new uses: Meyer rates an energy and computing-capacity bottleneck at 80%, with routine AI work moving onto personal devices; AI becoming a lab partner receives 65%. The supplied source material cuts off during the discussion of that research prediction and does not include the full list of ten calls. The remaining predictions and their ratings cannot be reported from the material provided.

How the Forecasts Could Be Tested

The list matters as a set of specific claims that can be revisited, rather than a general assertion that AI will advance. Meyer names possible tests for several predictions, including whether agents remain mostly chat interfaces, whether production code is still mainly typed by people, and whether assistants retain useful information between sessions.

Several calls also point to practical decisions for organizations and users. If AI systems take on longer tasks, teams may need clearer checkpoints and ways to review progress. If generated output becomes plentiful, testing and verification could take up more of the work. And if persistent assistant memory becomes common, data control and portability may affect which products people choose. These are implications of Meyer’s forecast, not established effects.

Meyer’s Scoring Framework

Meyer describes his list as a forecast that should be judged against later evidence. He says a 70% rating means he would bet on a prediction while expecting to lose roughly three times in ten; a 45% rating means he leans slightly against the call. The stated average of 72.5% applies to all ten ratings, although the supplied text contains details for only part of the list.

For the multi-day-agent prediction, Meyer says he would count against his call if long-running tasks remain demos rather than a routine way to use AI by the following October. For code generation, he points to whether most new production code is still typed line by line by people. For memory, he asks whether mainstream assistants continue to forget between sessions or whether users rely on memory features. These are Meyer’s proposed measures; the source does not describe an external review or scoring process.

“A forecast, not a prophecy.”

— Thorsten Meyer, writing on ThorstenMeyerAI.com

Which Predictions Will Hold Up

None of the 2027 outcomes in the article are confirmed: they are predictions published in October 2026. Whether agents can reliably manage work across several days, whether AI-written code will exceed human-typed code in production, and whether users will trust persistent assistant memory remain open questions.

The source text provided here is incomplete. It describes six predictions, with the sixth section cut off, despite stating that there are ten. The full wording and confidence ratings for predictions seven through ten are unavailable, as are the complete details of the lab-partner forecast. Meyer’s proposed scoring signals are also not independent verification; the material does not identify an outside evaluator or a formal scoring method.

The October 2027 Checkpoint

Meyer says he intends to score the predictions by October 2027, using signals described under each call. That review would provide a point at which readers can compare the forecasts with developments in agents, coding, verification, memory, power use and research.

For a full account of the ten calls, the complete source article is needed: the supplied material does not show all predictions or their measures. Until then, the ratings should be read as Meyer’s stated confidence, not as a consensus outlook or a measured probability of what will happen.

Key Questions

Who published the AI predictions?

Thorsten Meyer published them on ThorstenMeyerAI.com on October 15, 2026.

What does the 72.5% figure mean?

It is the average of Meyer’s ten confidence ratings. He says that if he is well calibrated, about seven predictions should come true by October 2027. It is his own estimate, not an external assessment.

Are the predictions confirmed developments?

No. They are forecasts for 2027. The source presents possible tests for some calls, but the outcomes have not yet been established.

What AI changes does the supplied material describe?

It discusses multi-day agents, AI-generated production code, verification, assistant memory, energy and computing limits, and AI-assisted research. The provided text is incomplete and does not include the full ten predictions.

Source: ThorstenMeyerAI.com