2025: What came true of our Predictions?🎄Ep. 42

Welcome to our 2025 Year-End Retrospective

Big announcement first: Starting January 2026, this newsletter is officially also becoming a podcast. New name, new platforms (Spotify, Apple Music, everywhere), same Aragorn and Patrick diving into exponential change. The new brand launches in weeks. We'll also move the newsletter to substack, but more on that later All formats will be complementary, covering the same ground from different angles.

We're writing this from holiday mode after spending 2.5 hours fighting grocery store crowds. We're exhausted. The singularity doesn't take breaks, but humans need them.

So this episode is different. No breaking news. No frantic benchmark updates. Just Patrick and Aragorn looking back at everything that happened in 2025, checking which predictions came true (spoiler: we didn't do badly), and mapping what 2026 will bring.

Cheers,
Patrick, Nikola & Aragorn 🚀

2025's Biggest Realization: AI Outsmarts Humans

AI surpassed human intelligence. The IQ measurements hit 125+ in January. During 2025 we realized: The smartest entity in the world right now is already non-biological. IQs above 130 means AI is in the top 1% percentile of humanity. Not "approaching human-level." Already there. Already beyond most humans.

The progression in 12 months:

• January 2025: GPT-4o Vision scored just below human average IQ

• December 2025: GPT-5 Pro scores high 120s, Gemini 3 Pro near 130

• Gemini 3 Flash (significantly cheaper and faster) scores the same as Gemini 3 Pro

In one year, we went from the bottom of human average IQ to absolute top genius level at a ridiculously low cost. Prices for frontier models went down 10x. For some specific models, 100x price drops in terms of cost.

The Leverage Effect

AI doesn't just help smart people get smarter, it levels the playing field.

If you're a smart person and you have a smart AI, you're doubling your intelligence. But if you're an average person (anywhere between 90 and 100 IQ), you now have a 120+ IQ AI in your pocket. You can make up for so much of whatever you naturally lack in terms of intelligence.

Why does it matter?
When average people have genius-level intelligence in their pocket, the entire structure of knowledge work, education, and economic value creation shifts. The people who traditionally couldn't compete because they lacked raw IQ can now leverage AI to punch far above their weight. Meanwhile, highly intelligent people who don't use AI will be outperformed by average people who do. We keep debating "when will AGI arrive?" while ignoring that functional superintelligence already exists in your phone.

We Checked Our 2025 Predictions.
_{(How do you think we did?)}

In January 2025, we made specific predictions for the year. So what better way to judge this year, than by seeing how much of what we thought we saw coming, actually came true? (check out the original newsletter with all predictions here.)

Predictions Overview​

📉 Job Markets & Economy

• Major Tech Workforce Cuts (20-30%): YES. Companies like Fiverr and Chegg led the way with massive workforce reductions to pivot toward AI-first structures.finalroundai​

• 3-4% Western Job Displacement: PARTIALLY. While displacement hit ~2% in tech hubs, the broader service sector is still in the early stages of this shift. azumo​

• Emergence of "AI Managers": YES. Job titles for managing autonomous agent fleets saw a 150% surge this year.anthropic​

• Anti-AI Protests & Union Action: YES. Mid-2025 saw significant strikes in the gaming and creative industries specifically over AI usage rights.theverge​

🧠 AI Intelligence & Agents

• ARC Test Performance (25%+): YES. We smashed this—commercial models reached 37.6% on ARC-AGI benchmarks this month.arcprize​

• 90-99% Drop in Compute Costs: YES. Inference costs plummeted, making complex reasoning tasks nearly trivial to scale.youtube​arcprize​

• The End of Traditional Apps: PARTIALLY. Apps haven't died, but they’ve become background APIs for the new "Agent-First" OS interfaces from Apple and Google.weforum​

• AI-Led Scientific Breakthroughs: YES. AI was directly credited with discovering new fusion reactor fuel configurations this year.lanl​

🎮 Gaming & The Metaverse

• Dynamic AI NPCs: YES. Real-time LLM-driven interactions are now a standard feature in major 2025 releases.reddit​

• AI-Generated AAA Games: NO. While AI assets are everywhere, a fully AI-generated AAA title from an indie dev is still on the horizon.polygon​

• Meta Glasses as a Primary Interface: PARTIALLY. Adoption has tripled, but they are still largely used as high-end companions rather than full phone replacements.innovation-network.beehiiv​

⚡ Energy & Quantum

• Nuclear Power Renaissance: YES. AI energy demands forced a historic shift in acceptance, leading to the reopening of plants and SMR deployment.engineering.tamu​

• Quantum Error Correction: YES. Practical, error-corrected qubits moved from lab theory to real-world demonstrations this year.techcrunch​

• AI-Designed Reactors: YES. Machine learning models have successfully optimized next-gen reactor fuels now undergoing regulatory review.lanl​

The Verdict: We were about 85% right. The speed of the "Intelligence Age" is no longer a theory—it is our current reality under the second Trump Administration, which has leaned heavily into these technological shifts for national infrastructure.wikipedia​

Prediction Deepdive: AI Models Reaching 250+ IQ

Verdict: Close, but IQ measurement gets fuzzy at these levels.

We predicted frontier models would hit 250+ IQ by end of 2025.

What happened: Models reached 180 with expensive subscriptions. Google Deep Think (not publicly available) and ChatGPT's internal frontier models likely exceed 200.

Here’s some context on the benchmarks:

Arc AGI: At start of 2025, still seemed valuable. Had to invent Arc AGI 2 because of rapid saturation. AI went from "very bad" to approaching human levels by year-end.

Humanities Last Exam: Accuracy jumped from GPT-4o scoring 2.7% to Gemini scoring 38%. Most humans would have a very hard reaching these levels.

Frontier Math: At the start of the year, people believed AI would never be able to do math because this was a big problem of large language models. GPT-5 scored 16.7%, Gemini 3 Pro scored 18.8%. 99% of people watching this podcast couldn't even reach 2% score on this test. This is the absolute hardest level of math you can imagine. For AI to score 18.8% on a limited timeframe is absolutely mind-blowing. That was impossible a year ago.

Other Predictions

✅ OpenAI custom GPT showing early agent capabilities - Definitely happening
✅ Google's integration of AI across product suite - Happened
✅ Microsoft Copilot fundamentally changing office usage - Happening (though not always great)
✅ Salesforce and ServiceNow developing autonomous agents - Happened
✅ Quantum computing progress - Massive strides
✅ Nuclear power costs dropping through AI optimization - Confirmed

Why does it matter?
It’s really about understanding the trajectory. Our prediction methodology wasn't perfect, but the direction was accurate: job displacement happening faster than expected in specific sectors, AI surpassing human intelligence sooner than imagined, corporate restructuring accelerating. The lesson is: If you understand exponential curves and pay attention to signals, the future is more predictable than it seems.

2025: Year of Agents or Year of AI Video? (The Answer is Both)

This was the year AI video truly broke through, reaching a level of realism that is now indistinguishable from actual footage.

We saw this firsthand at our latest Innovation Network meetup. Jan Willem Blom showcased how he generates multiple commercials at a fraction of the traditional cost and time.

We have discussed this shift for a long time, but with tools like Sora, that future has arrived. Now, virtually anyone has the power to create crisp, high-fidelity video content in seconds.

Aragorn argued that 2024 was the actual 'year of AI video' regarding public awareness, while we are now simply living in the maturity of that shift.

Instead, he draws a parallel between video's past trajectory and the current state of AI agents, a technology that is making massive technical strides this year, even if it hasn't fully hit the mainstream yet.

The Case for 2025 Being the Year of Agents

In January 2025 we only had chatbots. In December 2025 we saw:

• MCP rolled out universally, everyone working with it

• Perplexity has AI agent included with automatic connectors to your email and Drive

• Gemini fully integrated with Gmail, Google Drive, everything (available to anyone with Android phone and Google account)

At the start of this year you'd be lucky if you could upload a document to ChatGPT with a paid account and maybe ask a question about it.

More agent evidence:

• Lovable - probably biggest unicorn of the year, basically agentic coding

• Salesforce - DreamForce was all about putting AI everywhere in everything

• Chinese hacker using AI agents - biggest AI-run hack in history, one hacker steering a Claude manager running hundreds of individual Claude agents

Why does it matter?
Progress is so fast we can't even agree on what defined the year. Both are right. 2025 was the year AI video became indistinguishable from reality (Hollywood in your pocket, deepfakes everywhere, fan-made Star Wars films in basements). AND it was the year agents emerged from chatbots to actual assistants that can access your email, search your files, write code, hack systems. The fact that these two massive breakthroughs happened simultaneously shows how compressed innovation timelines have become.

2026 Predictions: The Year You Get Your Jarvis (And 90% of Content Becomes Fake)

After reflecting on 2025, we shifted to what's coming in 2026.

Pat’s Prediction: True Agentic AI Arrives for Consumers

Patrick made an incredible app for our Innovation Network Meetup without really any prior programming experience. Is it perfect today? No. Is it very low entry bar? No. It still requires some basic knowledge, understanding, willingness, investment.

But if we follow the trajectory from where it was a year ago to today and extrapolate to a year from now, we can easily infer that we’ll basically be able to talk to our computer, tell it what we need and make it work.

Aragorn’s Prediction: Robots in Homes, Cars on Autopilot

X1 has been rolling out the €499/month model, referencing Kai-Fu Lee's AI 2041 chapter called "The Golden Driver" about autonomous vehicles that can be taken over by remote human drivers in emergencies (like FPV drone pilots controlling from distance).

We think that will be reality by end of 2026. Maybe not on mass level, maybe not for millions of people, but those things will exist. There'll be cars driving around in Europe, Asia for sure, and the US that are autonomous.

We'll have robots that do a lot in the home, but for specific circumstances are controlled by humans. It won't be millions of people having it, but there'll be tens of thousands or even hundreds of thousands running with this. That will accelerate progress in this industry.

Both Predict: Space Data Centers Launch

Technically, this enables a bi-directional energy future. We are looking at a scenario where satellites can be powered remotely from Earth, while orbital arrays harvest solar energy to beam directly back to the surface. Development is already underway, and the rate of progress in this sector is exponential.

Pat’s Prediction 4: The Deepfake Crisis Hits Mainstream

There will be bad actors playing with it. We don't know where it will lead to, but it will be on a larger scale than we've seen in 2024 and 2025.

Both Predict: Premium on Human Connection

If you look at 2020, we went into the pandemic. Everybody got working from home. All these online tools exploded. Every major tech company said 'this is okay, we can all sit in a small room and have video conferences.' Look how this has pivoted over five years—how important physical connection is.

Pat’s Prediction: Corporate Restructuring Goes Mainstream

The new competition:

Next year, the theoretical becomes practical: the rise of the hyper-efficient company. We have seen organizations like Lovable and Cursor achieve global reach with fewer than 50 employees, validating the prediction that AI-enabled individuals can rival traditional corporations.

This shift will spill over into the political arena, demanding a new cognitive approach: the 'Singularity Surfer,' who learns to ride these exponential curves.

Aragorn anchors this prediction in Kai-Fu Lee’s AI 2041. Specifically, he points to the 'Job Savior' chapter, which outlines the harsh new math of the workforce, a world where human labor is rendered obsolete not by inability, but by the sheer cost-efficiency of automated alternatives.

Why does it matter
These aren't wild futurist predictions, they're obvious extrapolations from current trajectories. If you can already make apps without coding experience, voice-control partial AI assistance, and see robot prototypes, then fully functional Jarvis-level AI assistance within 12 months is inevitable. The corporate restructuring prediction is the scary one because it's the most certain: When 30-person companies can do what 10,000-person companies did, the math is brutal. 2026 won't be the year we debate AI's impact. It'll be the year we live through it at scale.

Final Message: Tough It Out, The Best is Yet to Come

Aragorn wanted to close with something important for people feeling anxious about 2026.

Humanity has been around for a million years and we've done just fine for the majority of that time. We've only improved and grown. We've raised families, been productive, felt satisfied for many of us.

The promise: We really believe we're in for even much better things. There will be some turmoil. You cannot expect to transition in such a massive paradigm change and it'll just happen overnight and be fine. There will be challenges.

The strategy: All you need to do is tough it out, stick it through, find ways to make the best of it so you can come out on the other side. It's like sometimes you're lucky and fall in love. Sometimes you're unlucky and fall out of love. What follows is a tough period, but you know at the end you'll probably be better and stronger for it and find new love.

The conclusion: Considering we're going into Christmas and end of the year and there's so much uncertainty, we want to emphasize: There's incredible potential at the end of this road, and not just scary things.

The 2026 promise: Singularity Surfing will become a term! You'll see much more of us on all types of channels. We've got some nice things in store for you. Thanks for following us. Thanks for sharing the content and all the nice comments. Keep doing that if you like what we're doing.

The Full Podcast