Welcome back to the prediction scorecard, our monthly exercise in holding the AI industry accountable for the things it promised. Last time we graded September 2025’s biggest claims. The overall score was a dismal 2.4 out of 10.
This month we’re looking at what the industry’s loudest voices were saying in October and November 2025 — and checking those claims against the world we’re actually living in.
The Scorecard
Prediction 1: “AGI by 2026, 10% Probability with Grok 5”
Who said it: Elon Musk, October 2025
The claim: Musk estimated a 10% probability that Grok 5 would achieve AGI, with the model expected in Q1 2026. He also predicted AI would surpass individual human intelligence by the end of 2025.
Reality check: Let’s start with the obvious: Grok 5 hasn’t shipped. The Q1 2026 window has closed. Prediction markets put the odds of a June release at just 12%, and the model is reportedly still training on xAI’s Colossus 2 cluster.
But here’s what makes this prediction genuinely remarkable. Since Musk made it, all 11 of xAI’s co-founders have left the company. Every single one. Musk himself admitted xAI “was not built right” and is “being rebuilt from the foundations up.” SpaceX absorbed xAI in a $1.25 trillion all-stock deal.
It’s hard to achieve AGI when the entire founding research team walks out the door.
Score: 0/10 — Worse than last time. At least in September the team was still there.
Prediction 2: “Widespread Tesla Robotaxis by End of 2026”
Who said it: Elon Musk, various statements October-November 2025
The claim: Fully autonomous Tesla robotaxis operating without safety drivers, expanding to multiple cities, with hundreds of vehicles on the road.
Reality check: As of March 2026, Tesla has expanded its Austin geofence to 245 square miles. Progress, right? Except only 4 to 8 Model Ys are actually running unsupervised. And “unsupervised” means no human in the car — Tesla still uses chase cars and remote monitoring.
Eight months into the program, the service has just 19% availability, roughly three dozen vehicles total, and prices that recently jumped for a 5-mile trip. The Cybercab — the dedicated robotaxi with no steering wheel or pedals — is supposedly starting mass production this month.
For comparison, Waymo operates fully driverless in multiple cities with over 250,000 paid rides per week. Tesla has maybe 8 cars.
Musk now says “widespread” by end of 2026. We’ve heard that deadline before.
Score: 1/10 — The geofence grew. Everything else is smoke.
Prediction 3: “AI Writing 90% of Code Within 3-6 Months”
Who said it: Dario Amodei, Anthropic CEO, March 2025 (reiterated through October)
The claim: AI would be writing 90% of code within 3-6 months, and “essentially all of the code” within 12 months.
Reality check: We’re now 13 months past the original claim. Futurism ran a headline in October: “Exactly Six Months Ago, the CEO of Anthropic Said That in Six Months AI Would Be Writing 90 Percent of Code.” The answer was no.
When pressed, Amodei claimed the prediction was true “within Anthropic”, then walked it back to “on many teams.” Redwood Research’s analysis found this claim doesn’t hold up even at Anthropic.
The actual numbers? About 42% of code in codebases with AI tools enabled is AI-assisted. 51% of professional developers use AI tools daily. But “AI-assisted” means a human prompted, reviewed, edited, and approved it — that’s augmentation, not replacement. And 46% of developers say they don’t fully trust AI-generated code.
The industry is crossing 40-50% AI-assisted code for developers who actively use these tools. That’s impressive. It’s also half of what Amodei promised, among a self-selected group.
Score: 2/10 — The trajectory is real. The timeline was fantasy. And the goalpost-moving was shameless.
Prediction 4: “Gemini Will Be the Only AI That Matters”
Who said it: Sundar Pichai, October-November 2025
The claim: Pichai predicted Gemini 3.0 before end of 2025, and that AI would act as an agent for users within the next year. Google positioned Gemini as the model that would lead every benchmark.
Reality check: This one is complicated because parts of it actually landed. Gemini 3 Pro shipped November 2025. Gemini 3.1 Pro launched February 2026 and genuinely leads 13 of 16 benchmarks Google measured: 77.1% on ARC-AGI-2, 94.3% on GPQA Diamond, 80.6% on SWE-Bench Verified.
The benchmark numbers are real. Google actually delivered a technically impressive product on something close to schedule.
But “the only AI that matters”? ChatGPT still dominates consumer usage. Claude holds a loyal developer base. GPT-5.4 scored 75% on the OSWorld desktop autonomy benchmark. The market is more competitive than ever, not less.
And the “agent” prediction? Google’s AI agent products remain scattered across Workspace, Android, and various APIs with no unified experience that matches the original promise.
Score: 6/10 — The benchmarks delivered. The dominance narrative didn’t. Partial credit for actually shipping something impressive.
Prediction 5: “Superintelligence in a Few Thousand Days”
Who said it: Sam Altman, originally September 2024, reiterated through late 2025
The claim: Superintelligence could arrive within a few thousand days (5-14 years from September 2024). By 2026, “systems that can figure out novel insights.” By 2028, “more of the world’s intellectual capacity could reside inside data centers than outside.”
Reality check: The “few thousand days” timeline is conveniently long enough to avoid accountability in any given month. But let’s look at where we actually are.
GPT-5.4 is impressive on benchmarks — 75% on OSWorld, beating the 72.4% human baseline for desktop tasks. But “figuring out novel insights”? Not yet. Models are getting better at executing known tasks, not generating genuinely new knowledge.
Meanwhile, OpenAI is bleeding executives ahead of its planned IPO. COO Brad Lightcap was reassigned, CMO Kate Rouch stepped down, and AGI head Fidji Simo took medical leave — all in the same week. Revenue hit $25 billion annualized, but the path to profitability on $852 billion valuation remains unclear.
Altman recently accelerated his own timeline, suggesting superintelligence could be “only a couple of years away.” The original “few thousand days” prediction is being revised before it can be graded. Classic.
Score: 3/10 — The models improve incrementally. The goalposts move constantly. And “novel insights” remains undefined by design.
Prediction 6: “Enterprise AI Will Finally Deliver ROI in 2026”
Who said it: PwC, Deloitte, McKinsey, every AI vendor
The claim: 2026 would be the year enterprises stopped experimenting and started seeing returns on their AI investments.
Reality check: Only 5% of enterprises are seeing real returns from AI in 2026. A MIT study found a 95% failure rate for enterprise generative AI projects, defined as no measurable financial returns within six months.
The numbers tell the story: 86% of companies increased their AI budgets this year — spending is projected to hit $2.52 trillion globally. But only 14% of CFOs report measurable ROI. And 61% of CEOs say they’re under growing pressure to justify the spending.
The clearest data point: 74% of organizations say they hope to grow revenue through AI. Just 20% actually are.
Score: 2/10 — The enterprise AI ROI story is the same one we’ve been told since 2023. “Next year” never arrives.
Prediction 7: “AI Will Transform the Job Market”
Who said it: Everyone, constantly
The claim: AI would cause either mass job displacement or massive job creation, depending on who was talking.
Reality check: The job market data from the past six months is sobering. About 55,000 job cuts were directly attributed to AI in 2025. In the first two months of 2026 alone, 32,000 technology jobs have been cut.
But here’s what makes this prediction especially insidious: companies are laying off workers because of AI’s potential, not its performance. Harvard Business Review found that layoffs are driven almost entirely by anticipated AI capability, not demonstrated results. Companies are firing people based on what they think AI will be able to do, not what it actually does today.
Entry-level workers are hit hardest — a 15% year-over-year decline in entry-level job postings, and unemployment among 20-to-30-year-olds in tech-exposed occupations has risen nearly 3 percentage points since early 2025.
The transformation prediction technically scored, but in the worst possible way: real people losing real jobs based on speculative capabilities that haven’t materialized.
Score: 5/10 — Jobs are being eliminated. But the cause is hype-driven corporate decision-making, not actual AI capability replacing workers.
The Pattern (It’s the Same Pattern)
If you read last month’s scorecard, this will sound familiar:
-
Timeline compression continues: Every major prediction was early. AGI, robotaxis, 90% AI code — all missed their deadlines by months or years.
-
The goalpost shuffle intensifies: Musk moved AGI from 2025 to 2026. Altman compressed “a few thousand days” to “a couple of years.” Amodei redefined who he was talking about. The only prediction that never fails is the next prediction.
-
Capability and deployment remain divorced: Gemini 3.1 Pro genuinely leads benchmarks. GPT-5.4 genuinely handles desktop tasks. The models are better. But the gap between what they can do in demos and what they do in production remains enormous.
-
The human cost is real and accelerating: Unlike previous scorecards, this month’s data shows clear job losses driven not by AI succeeding but by the expectation that it will. People are paying the price for predictions that haven’t come true.
What Actually Happened Since October 2025
The past six months have been defined less by breakthroughs than by reckoning:
- xAI imploded. All 11 co-founders gone. SpaceX absorbed the company. The organization promising AGI can’t hold a team together.
- Enterprise AI spending keeps rising while ROI stays flat. The industry is spending $2.52 trillion annually. 95% of projects aren’t delivering returns.
- Models got better, quietly. Gemini 3.1 Pro and GPT-5.4 are genuinely capable. But “better” isn’t “revolutionary,” and nobody runs press conferences for incremental improvement.
- Workers are losing jobs preemptively. Companies are cutting headcount based on AI hype, not AI results. This is the prediction that became self-fulfilling.
What This Means
The AI prediction industry has a credibility problem, and it’s getting worse.
When we ran this scorecard in March, the overall accuracy was 2.4/10. This month it’s slightly higher — 2.7/10 — but only because Google actually shipped something that matched expectations. Strip out Gemini, and we’re back to the same dismal average.
The executives making these predictions face no consequences for being wrong. No one at OpenAI, Anthropic, or xAI gets penalized for a missed timeline. The media cycle rewards bold claims and forgets broken ones. And the people who do face consequences — entry-level workers laid off because of AI hype — had no say in the predictions at all.
Here’s the discount formula we proposed last time, updated with six more months of data: when an AI executive says something will happen “next year,” assume 3-5 years. When they say “a few months,” assume 2-3 years. When they say “within our company,” assume “on one team, once, under ideal conditions.”
The technology advances. The predictions don’t improve. And the gap between promise and delivery is where regular people get hurt.
The Bottom Line
Overall prediction accuracy from October 2025: 2.7/10
Marginally better than September’s batch, entirely because Google delivered on benchmarks. Every other prediction followed the same pattern: too bold, too early, too conveniently vague.
We’ll be back next month. The predictions certainly won’t stop, so neither will the scorecard.