Short answer: in a same-prompt test on 12 November 2026, six AI assistants expected more capable agents in 2027, no AGI, power and permitting as a hard limit, open-weight models close behind the leaders, and regulation arriving piece by piece. They disagreed on how bad the first big AI incident gets, how many jobs it costs, and whether AI wearables sell.
How to read this. One run per assistant, through each app's public web version, no follow-up and no editing. The probabilities are each model's own estimates, not Satcove's forecast, and we have not verified them. The full comparison, with every table, is in our News write-up.
The questions
Will AI reach AGI in 2027?
No, according to the models that took a position. ChatGPT gives 85%, Perplexity 90% and Claude 95% that no widely accepted AGI arrives in 2027. Mistral puts a validated AGI claim under 10%. Gemini lists it among its "will not happen" items without a number. Grok does not address AGI directly; its own "will not happen" is a sustained 2x jump from recursive self-improvement.
What will AI agents be able to do in 2027?
Longer professional tasks with limited supervision. ChatGPT gives 85% to agents completing 30-minute to two-hour workflows. Perplexity gives 80% to an enterprise running one unsupervised on routine cases. Gemini gives 80% to a frontier model above 75% on SWE-bench Pro. Mistral gives 60% to a lab demonstrating multi-day software projects above 90% success. Claude is more cautious: 55% for software tasks worth three or more human days at 50% reliability.
Will AI take jobs in 2027?
The models agree on direction, not on size. Mistral gives 70% to one large Western company attributing 1,000+ layoffs to AI. Gemini gives 65% to three Fortune 500 companies each attributing 5,000+ cuts. Claude gives 50% to one Fortune 500 layoff of 5,000+ plus a measurable graduate unemployment gap. ChatGPT gives 75% to a 10%+ fall in graduate hiring in two sectors. Grok is the most aggressive: 60% that US white-collar unemployment rises by 1.5 points.
Will there be a serious AI incident in 2027?
Probably, if "serious" is loosely defined. ChatGPT gives 70%, Perplexity 65%, Gemini 60%, Grok 50%, Mistral 45%, Claude 35%. Most of that spread is definition, not belief: Grok counts a $10 million loss, Gemini $50 million, Claude $100 million of documented damage from an AI-enabled cyberattack.
Will open-source AI catch up with closed models?
All six say the gap stays narrow, from 65% (Claude, Grok, Mistral) to 85% (ChatGPT, Gemini). Gemini is the most concrete: 85% that an open-weight model sits in the Chatbot Arena top three for 14 consecutive days. Claude adds that a Chinese lab produces the best open-weights model (65%).
Will AI be regulated in 2027?
In pieces. Grok and Mistral both rule out a binding international AI treaty (Mistral: under 15%). What the models expect is local enforcement: ChatGPT 90% for enforceable frontier-AI obligations in at least one major jurisdiction, Claude 80% for at least 10 new US state laws, Grok 75% for EU AI Act findings against two non-European labs, Gemini 70% for binding watermark or provenance rules in two jurisdictions, Mistral 55% for an EU enforcement action on general-purpose models.
Will energy limit AI in 2027?
Five of six say yes. ChatGPT gives 90% that data-centre demand becomes a major electricity-policy issue. Claude gives 90% to a 1 GW+ nuclear or gas deal dedicated to data centres. Perplexity gives 85% to a flagship AI campus delayed, downsized or relocated over power or permitting. Grok and Gemini sit at 80% and 75%. Mistral did not address energy.
Will Nvidia keep growing?
Only Mistral took a stance against it: 50% that Nvidia's data-centre growth falls below 30% year over year as custom chips take over 20% of accelerator spend. The other five did not discuss chip share, so this is one model's bet, not a consensus.
Will an AI wearable sell in 2027?
The models contradict each other. Gemini gives 55% to a screen-free or heads-up AI wearable from Apple, Meta or Google selling over 5 million units. Claude gives 20% to an AI wearable selling the same 5 million. Grok gives 50% to always-on AI devices shipping over 20 million units combined, and Mistral 40% to an AI-native device reaching 10 million weekly users.
Can I trust these AI predictions?
Treat them as a shortlist of questions, not as forecasts. These are single runs, so each number is one draw. The models were not given shared definitions, so the same topic is measured with different thresholds. Their errors can also be correlated, because they learned from overlapping text. Agreement is a signal, not proof.
How can I check them later?
Most claims have a stated threshold: a benchmark score, a unit count, a dollar figure. That is why we asked for falsifiable predictions. In December 2027, each one can be scored as true or false against its own criterion.
What to take from it
The overlap is a reasonable planning shortlist: agents, energy, piecemeal regulation, a narrow open-versus-closed gap. The disagreements are the questions worth watching: incident severity, jobs, wearables. When six assistants give six different thresholds for the same word, the useful move is to ask them to agree on a definition first.