My View Of AI In 2027: 10 Predictions, From High To Low Confidence

🔍 Read the full analysis: My View Of AI In 2027: 10 Predictions, From High To Low Confidence on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the little things that make your day 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, giving each forecast a confidence score and describing evidence he plans to use to judge it. The essay argues that AI use may shift from chat-based prompts toward agents that perform longer tasks, while emphasizing that the forecasts are claims about the future, not established outcomes.

Meyer says his ten confidence scores average 72.5%. He frames that figure as a test of calibration: if his estimates are accurate, he expects roughly seven predictions to come true by October 2027. He also says each forecast includes signals for scoring it. The source text provided for this article contains only part of the list, however, and cuts off during the sixth prediction; it does not provide the remaining four predictions or all of the scoring criteria.

Among the forecasts visible in the source, Meyer assigns 85% confidence to AI assistants retaining information about users’ projects and decisions across conversations. He gives 80% to the idea that teams will demand evidence—such as tests, screenshots or independent review—rather than accept plausible AI-generated work without verification. He also predicts that electricity and computing capacity will constrain AI deployment, with routine tasks shifting toward local devices.

Meyer assigns 75% confidence to his forecast that most new code will be produced by AI agents rather than typed by people. His 70% prediction is that people will delegate work lasting days, then review the result, instead of directing AI one prompt at a time. He gives a 65% score to AI becoming a research “lab partner,” generating approaches and simulations while people and machines test and evaluate results. Each percentage is Meyer’s own estimate, not an independently measured probability.

From Chat Prompts to Delegated Work

The forecasts matter because they describe a possible change in how people organize work around AI. If agents can handle tasks over several days, users may spend less time issuing individual instructions and more time defining goals, checking progress and approving results. Meyer identifies a practical obstacle: agents can lose track of the original goal during long tasks. He argues that visible work logs, checkpoints and explanations would help users monitor what an agent has done.

His predictions also place verification and control alongside generation as central issues. Faster production of code, reports or research ideas would not by itself show that the output is correct. And if assistants retain information about a person’s work, their owners will need to decide where that memory is stored and who can inspect, change or export it. These are implications raised by Meyer’s forecast, not changes the source establishes as already widespread.

The infrastructure prediction adds a physical constraint to the discussion. Meyer expects power, chips and cooling to affect the availability and cost of AI services, and suggests that some routine tasks may run on phones or laptops. Whether that happens at scale will depend on technology and investment beyond the user experience of any one assistant.

How Meyer Plans to Score Forecasts

Meyer presents the article as an accountable forecast rather than a prophecy. He says a prediction should be specific enough to fail and attaches a confidence score to each of his ten calls. In his framing, a 70% estimate means he would take the bet while expecting it to be wrong about three times in ten; a 45% estimate means he leans slightly against the prediction’s headline.

For several forecasts, he names a possible result that would count against him. The agent prediction would be weakened if AI remained mostly a chat interface and long tasks stayed demonstrations rather than a normal way to work. The coding forecast would be challenged if people continued to type most production code line by line while AI remained limited to autocomplete. The source also describes potential disconfirming signals for memory and verification, including assistants that still forget between sessions or customers who accept answers without tests or traces.

The available source is dated October 15, 2026 and is prospective: the stated review point is October 2027. It also references an accompanying video and infographics, but the provided material does not include those visuals. The article’s title promises ten forecasts, yet only six are represented in the supplied text, with the sixth ending mid-sentence.

“A forecast, not a prophecy.”

— Thorsten Meyer

Forecasts Still Await Their Test

None of the 2027 outcomes is confirmed by the source; the article records Meyer’s expectations as of October 2026. It supplies no independent evidence that AI agents are already completing multi-day assignments as a default, that AI writes most new production code, or that persistent memory will become a routine buying consideration. The stated confidence levels are the author’s judgments, not results of a published forecasting assessment.

The supplied source also does not contain the full ten-prediction list. It stops during the sixth item, after introducing the idea that human judgment will remain important in AI-assisted research. The remaining predictions, the rest of that discussion and the referenced scoring infographics are unavailable here. No details are provided about a formal scoring process beyond Meyer’s plan to compare the forecasts with signals by October 2027.

It is also unclear how Meyer will define terms such as “most new code,” “default” use of agents or mainstream AI memory. Those definitions could affect whether a forecast is counted as fulfilled, partly fulfilled or wrong. His essay describes possible indicators, but the available text does not set out a complete measurement protocol.

October 2027 Is the Checkpoint

Meyer says he will use the signals listed under each prediction to assess the forecasts by October 2027. That review is the next stated milestone. Readers will be able to compare the predictions with developments in agent tools, software production, AI verification, assistant memory and computing infrastructure, though the supplied material does not describe a separate audit or external scorekeeper.

For now, the forecasts should be read as a dated set of expectations. A later assessment would need to show which predictions met their stated tests and how partial results were handled. The full list and its accompanying indicators would also be needed to evaluate all ten calls rather than the six included in the available source.

Key Questions

What did Thorsten Meyer publish?

He published 10 predictions about AI in 2027, assigning confidence scores and describing evidence he plans to use to evaluate them. The source provided here includes only part of the list.

Which prediction has the highest confidence score in the available text?

Meyer gives his forecast that AI assistants will retain information about users’ work across conversations a score of 85%. That is his estimate, not a verified probability or an outcome that has already occurred.

Does Meyer say AI will definitely take over multi-day work?

No. He assigns 70% confidence to the forecast that people will delegate work lasting days to agents. He also identifies agents losing track of goals over time as a reason for uncertainty.

When will the forecasts be assessed?

Meyer says he expects to judge the predictions by October 2027, using signals described under each forecast. The supplied source does not provide a formal independent review process.

Source: ThorstenMeyerAI.com