2026-03-25
By Vadym · Generated with Boba, curated by me
March 2026
There's a gap between what's true and what most people have processed. That gap is where the future lives. Not in predictions, not in forecasts — in the things that already happened quietly while everyone was still arguing about the last thing.
So let me try to describe the edge. Not headlines. The actual edge. And then let's look five, ten years out — not to predict, but to orient.
Here's something most people haven't absorbed yet: more than half the code committed to GitHub in early 2026 is AI-generated or substantially AI-assisted. Not experiments. Not demos. Production code, shipping to users. A Swedish startup called Lovable went from zero to $100 million in annual revenue in under a year — the fastest-growing software company in history — built almost entirely on the premise that AI writes the code and humans steer.
The solo founder is no longer a romantic exception. It's becoming a viable architecture. 52% of successful startup exits last year were solo-founded. Dario Amodei, who runs Anthropic, was asked when we'd see the first solo unicorn built with AI. His answer: 2026. This year.
Meanwhile, the hyperscalers are spending like it's wartime. Meta plans up to $135 billion on AI infrastructure this year. Google: $185 billion. Total hyperscaler capex for 2026 is approaching $700 billion, most of it for AI. NVIDIA just posted $216 billion in annual revenue, up 65% year over year. This isn't speculation. This is capital deployment at a scale we've never seen outside of actual wars.
And it's not just code. Medical transcription is effectively automated. 40% of medical coding is handled by AI. Loan processing is halfway there. Radiology reads for routine scans are on borrowed time. In legal work, paralegals face 80% automation risk. The creative industries are further along than anyone in the creative industries wants to admit — 90% of marketers have used generative AI, and the quality ceiling keeps rising.
The edge of today is this: AI isn't coming. It arrived. The question now is how deep it goes, and what that does to everything else.
Let's be honest about what we don't know. Nobody has a reliable model for what happens when AI agents can sustain 45-minute autonomous work sessions (already happening) and that extends to hours, then days. Nobody knows what happens to the structure of a company when one person can do what used to require a team of twenty.
But some trajectories are clear enough to take seriously.
AI integration into daily work will go from augmentation to delegation. Right now, engineers can fully delegate maybe 20% of their tasks to AI. That number will climb. Not to 100% — the interesting work will stay human for a while — but the floor of what requires a human keeps dropping. McKinsey estimates 60-70% of document processing, research, and data analysis tasks can be fully automated. The jobs don't all disappear. They reshape. But the reshaping is brutal if you're not paying attention.
Space is real but slower than the hype. Starship has flown nine times. SpaceX launched 165 orbital missions in 2025 — a record. But Artemis III just got pushed again; the first actual lunar landing is now 2028 at the earliest, with $20 billion spent and $7 billion in overruns. Commercial space stations from Vast and Axiom are coming, but "coming" means 2027-2030. We are becoming a spacefaring species, but slowly, expensively, with more false starts than the evangelists admit.
The more immediate transformation is terrestrial. Within five years, most knowledge work will involve AI partnership as a baseline expectation — the way internet access became a baseline expectation for office work in the 2000s. The companies adapting fastest aren't the ones with the biggest AI budgets. They're the ones restructuring workflows around what AI actually does well. The UAE leads global AI adoption at 64% of its working-age population. Singapore is at 61%. These aren't tech hubs improvising — they're states making deliberate structural bets.
The strongest objections aren't coming from Luddites. They're coming from people who understand the technology deeply.
Gary Marcus, probably AI's most credible recurring critic, argues that large language models are architecturally wrong — they can't encode uncertainty, can't practice Occam's razor, can't know what they don't know. He's not saying AI won't matter. He's saying the current approach has fundamental limits that scaling won't fix. That's worth taking seriously, even if you disagree.
Timnit Gebru's argument is different but equally sharp: the harms aren't inevitable, they're choices. When AI development is driven by a handful of companies optimizing for their own interests, the externalities fall on everyone else. Bias, power concentration, erosion of public accountability — these aren't bugs. They're the predictable outcomes of the current incentive structure.
And then there's the meaning question. Researchers have started documenting what they call "AI Replacement Dysfunction" — professional identity loss, anxiety, a quiet crisis of purpose. The discourse around a "ghost economy" went viral earlier this year. The question underneath all the technical debate is simple and ancient: what is work for? If your job can be done by a model, what does a dignified life look like?
The gap between AI capex and AI revenue is also real. Hundreds of billions flowing into infrastructure against modest returns. Some analysts see a bubble. Maybe. Or maybe the returns are just lagging the investment, the way they did with the internet in 1999. The honest answer is: we don't know yet.
Here's what I think is actually happening, underneath all the noise.
We're not being replaced by AI. We're being restructured by it. The relationship between humans and their tools has always been symbiotic, but this is different in degree — and possibly in kind. When your tool can reason, create, and adapt, the line between tool and collaborator blurs. And that changes what it means to be the human in the room.
The people adapting well aren't the ones who learned to prompt better. They're the ones who figured out what they're uniquely good at — judgment, taste, context, the ability to ask the right question — and let AI handle the rest. The solo founders building companies aren't superhuman. They just stopped doing the parts that don't require a human.
This is both liberating and terrifying, and anyone who tells you it's only one of those things is selling something.
The next five to ten years won't be defined by whether AI gets smarter — it will. They'll be defined by whether we get smarter about AI. Whether we build the institutions, the guardrails, and the cultural frameworks to absorb a transformation this fast without losing the things that matter.
I don't have a prediction. I have an orientation: pay attention to the edge, take the skeptics seriously, adapt without surrendering your judgment, and stay honest about what you don't know.
The future isn't something that happens to us. It's something we're building right now, whether we're conscious of it or not. Might as well be conscious.