AI In 2027: Ten Possible Futures And My Confidence In Each

🔍 Read the full analysis: AI In 2027: Ten Possible Futures And My Confidence In Each on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on monitors, keyboards and dev gear

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

Thorsten Meyer published ten confidence-rated predictions for AI in 2027 on October 15, saying he plans to assess them against stated signals by October 2027. The forecasts cover agents that handle multi-day work, AI-generated code, verification, persistent memory, computing power and research; they are Meyer’s projections, not confirmed outcomes.

Meyer says the ten confidence levels average 72.5%. He presents the numbers as subjective odds: a 70% prediction, in his framing, is one he expects to be right about seven times in ten. He says each forecast includes signals for judging whether it has come true, and he describes the overall exercise as a forecast rather than a prophecy.

Among the five fully supplied predictions, Meyer gives persistent AI memory his highest confidence, 85%. He expects assistants to retain project information across conversations and says control over where that memory is stored could become a buying decision. He assigns 80% confidence to a greater emphasis on verifying AI output and to energy and computing limits shaping what AI products can run. His other calls are that most new code will be written by agents (75%) and people will delegate work over several days rather than submit one prompt at a time (70%).

The source begins a sixth forecast, “AI becomes a lab partner,” at 65%, describing agents proposing research approaches and running simulations. It says testing and judgment may become constraints as the number of hypotheses grows. However, the supplied text cuts off partway through that prediction and does not include the final four calls. The complete list and its evidence criteria cannot be reconstructed from the material provided.

How the Forecasts Could Be Tested

The article’s significance lies less in any single prediction than in its attempt to make a broad technology outlook measurable over a set period. Meyer says the signals beneath each forecast are meant to show what would count as evidence and what would prove him wrong. That provides readers with a stated basis for revisiting the claims by October 2027 rather than treating them as open-ended speculation.

The forecasts also point to practical decisions for workers and organizations. If agents take on longer assignments, people may spend more time defining goals and checking work. If AI-written code grows, clear specifications and reliable tests become more consequential. And if persistent memory becomes common, users and buyers will have to weigh convenience against privacy, portability and control. These are implications drawn by Meyer, not established effects.

His power prediction connects AI’s software ambitions to physical infrastructure: electricity, chips and cooling. Meyer expects those constraints to affect availability, cost and where computing takes place, including a possible shift of routine tasks to phones and laptops. The source offers this as a forecast; it does not provide independent data or a quantified estimate of future power demand.

From Prompts to Delegated Work

Meyer frames the coming year as a change from using AI one question at a time toward assigning systems a goal and reviewing their eventual output. His first prediction describes a hypothetical workweek in which an agent researches, builds and checks a project before presenting a result. He also identifies a risk: over several days, an agent may drift from the task it was given. He expects work logs and checkpoints to help users see what happened, but whether those features become standard is uncertain.

Several forecasts are linked in the article. More agent-written code would make acceptance criteria and testing more important, Meyer argues, while cheap plausible answers would increase the value of evidence that an answer is correct. Persistent memory could spare users from repeating background information, but it would also concentrate sensitive work data with whoever controls the memory system. Meyer calls for memory that can be inspected, edited and exported, with storage chosen by the user.

The source supplies only the opening of the research prediction and does not include predictions seven to ten. The published framing says there are ten forecasts and a scoring approach, but the available material does not support reporting the missing details or presenting the partial list as complete.

“A forecast, not a prophecy.”

— Thorsten Meyer

What the Source Leaves Open

These are one author’s forecasts, and the supplied material does not cite independent research or establish that the predicted changes will occur. Meyer says each prediction has signals for scoring, but those infographic details are not included here. The source also omits four predictions and cuts off during the sixth, so their subjects, confidence levels and criteria remain unknown.

Several forecasts depend on developments that could vary by product and industry. It is not clear how Meyer will define “most” new code, which assistants count as mainstream, or what evidence would establish that multi-day delegation has become a default rather than a demonstration. The article’s complete scoring rules and how borderline cases will be handled are also not available in the supplied text.

The October 2027 Scorecard

Meyer says the predictions are intended to be judged by October 2027, using the signals attached to each call. That later assessment would show how many forecasts met his stated criteria and whether the confidence levels were well calibrated, according to the approach he describes.

For now, the next step for readers is to consult the complete original article for the missing predictions and scoring details. Until those are available and the forecast period has passed, the claims should be treated as an agenda for tracking AI developments—not as a report that those changes have already happened.

Key Questions

What is the news in Meyer’s article?

Meyer published ten confidence-rated forecasts about AI in 2027 and says he plans to score them against evidence by October 2027.

Are the predictions confirmed developments?

No. They are Meyer’s forecasts. The article describes them as projections, and the supplied material does not establish that the predicted changes have occurred.

Which prediction has the highest confidence in the supplied text?

Meyer assigns 85% confidence to the prediction that AI assistants will remember users’ work across conversations. He also raises questions about who controls and stores that memory.

Are all ten predictions included in the source material?

No. The material provides five predictions in full, part of the sixth, and no details for predictions seven through ten. The missing forecasts and their scoring signals cannot be confirmed from the supplied text.

When will Meyer assess the predictions?

He says the forecasts will be assessed by October 2027, using the signals associated with each prediction. The supplied text does not include all of those criteria.

Source: ThorstenMeyerAI.com