A word on what these are before you scroll. I'm a Walt-Disney-level optimist with a doomsday clock in my back pocket, so read everything below in both registers at once — giddy and faintly terrified, like a kid who found the keys to the candy store and then read the ingredients as he gorged.
These aren't hot takes for the algorithm. They're me planting a flag on January 1 so that come mid-year, end of year, whenever, you can scroll back and tell me exactly how wrong I was. A prediction with no paper trail is just a vibe. Here's the paper trail. In no particular order, because the future doesn't queue politely either.
The vibe shift
There's going to be a real AI backlash — and it'll make "handmade" a luxury good. Expect a swelling tide of AI luddite-ism, "made by humans" badges, and a premium on the visibly hand-built. It won't slow enterprise down for a second — tech doesn't care and will steamroll ahead regardless — but you'll see a sharp line drawn between the apologists and the luddites. The luddites will lose. The horse has bolted; you can be mad about it from the saddle or from the dust, but it's gone.
AI becomes a hidden worker — and we'll love it more for it. The big move this year is AI doing things in the background, without invocation. We're going to shove the wizard firmly behind the curtain, which is exactly where we like our wizards. Nobody enjoys watching the levers get pulled and the dials get spun — it ruins the magic. Pushing the machinery out of sight makes AI both more accessible and, paradoxically, more magical. Watch for this to become the whole game.
"Good enough" eats the world. From coding agents to everything else, the hunt for perfect quietly dies this year. We accept the 1%-better principle and move on. Cheaper-and-now beats perfect-and-eventually, and most of the market won't even notice it made the trade.
The lab wars
Apple stays mediocre at AI, doubles down on hardware, and it's completely fine. Models are becoming a commodity, so Apple doesn't need to win the model race — it needs to win the use of models. Expect practical, mostly-invisible AI enablement baked into the hardware, a privacy story nobody else can tell, and a polite refusal to chase AI for its own sake. Everyone will call it a miss. It won't be.
Google wins — but not the way the org charts want it to. Vertical integration is the cheat code: top model, own chips, own infra, solid services, good harnesses spinning out. That's the mythical all-three-don't-pick-two outcome — better, faster, and cheaper — and it's brutally hard to run against. They won't sweep enterprise, though, because the wrong people make the decisions and pivoting a big org toward "better" is like turning a tanker with an oar. Many tears will be shed over AI capability that was right there and went unbought. Hot tip anyway: buy the stock. They should be almost everyone's AI cloud.
Microsoft and Amazon keep faltering, because generic is a trap. If it can do everything mediocre, it can do nothing well, and that's the corner both are painting themselves into. Amazon should lean into selling picks and shovels — be the easy on-ramp the way fal, Supabase, RunPod and OpenRouter are — and stop pretending it'll out-integrate Google (it's the better part of a decade behind on that). Microsoft should accept it's in the drapes-hanging business and get the wizard a bigger smoke machine. Neither will fully figure this out in 2026.
OpenAI slides to ~50% share, with Claude and Google taking most of the rest. The near-monopoly era is over. Watch the enterprise/API board especially — that's where the real money and the real movement is.
The models plateau on capability and compete on cost. LLM gains this year come more from the harness than the model. The next genuine capability leap is world models, not a bigger language model. Expect "cheaper and faster" headlines where you used to get "smarter" ones.
The build pipeline
AI ships code faster than humans can review it — so we start reviewing the reviews. Unchecked agentic PRs become common practice, not a confession. The interesting artifact stops being the PR and becomes the aggregate of reviews — review-of-reviews, dashboards of dashboards. Whoever owns that layer owns the bottleneck.
"100% AI-written codebase" announcements go from confession to flex. Expect a parade of them, plus a quieter parade of "100% of new code is AI-committed" claims that are technically true and quietly terrifying.
AI-powered QA lags badly. We'll accelerate the 16% of an engineer's day that's actually coding and forget that the other 84% — the planning, the PRs, the releases, the QA — doesn't get proportionally faster on its own. So the firehose gets bigger and the drain stays the same size. Companies that only speed up the coding will declare AI "broken." It isn't. Their plan is. They'll have turned themselves into factories that manufacture bottlenecks instead of software.
Agent memory becomes the hot topic — both kinds. Not just context memory (what the agent knows) but execution memory (what it did, why, and whether you can trace and stop it). The dirty secret arrives with it: memory you can't curate is memory that quietly poisons the agent. Remembering you liked pink and forgetting you changed your mind at 3am is a whole new class of bug.
Agents get trivially easy to build — and that's the problem. People will assemble multi-agent systems without knowing that's what they're doing. Marketplaces bloom, action fragments everywhere, and all of it accrues as what is functionally enterprise debt with a friendlier UI. The bill comes later, as bills do.
The subagent and MCP marketplace gets monetised in some form before the year's out. Where there's sprawl, there's a storefront.
Second half of the year, the focus shifts off engineering tools. The back half is about the rest of the lifecycle — the product side of engineering, and the post-engineering work: releases, infra, and especially cyber. The IDE was the easy part.
The world quietly reorganising itself
Robots get home-ish ready. Not in your living room folding your laundry — but pre-orderable, home-shaped, and demoing the right tasks. The "ish" is doing a lot of work in that sentence, and I'm comfortable with it.
Robotics advances are rapid — enabled by AI, not in spite of it. Watch for the moment hand-tuned control code gets replaced wholesale by a small learned model. That's the tell that this is real and not a highlight reel.
Fusion advances are rapid. Records fall, demo reactors stay on schedule, private money keeps piling in. Nobody puts net electricity on a grid this year — but the slope is unmistakably steepening.
Quantum advances are rapid too. (Filing this one honestly — it's the prediction I'm least equipped to grade myself on, so hold me especially accountable here.)
A huge wave of AI-driven medicine arrives, largely on the back of sequencing speed and an unprecedented grip on protein folds and binding sites. The flashy bit won't be a chatbot — it'll be systems finding druggable pockets we didn't know existed.
World models become something your non-technical mum has heard of. They go from research curiosity to consumer-adjacent product this year. Joe User won't understand them, but he'll know the phrase.
The money and the mess
Enterprise mimics the wrong things from big tech. Specifically: outsized layoffs without the outsized AI adoption or spend to justify them. Cargo-culting the headcount cut while skipping the transformation that was supposed to earn it. This is Peter Principle meets Conway's Law — the wrong people deciding, across boundaries they can't communicate across. The winners will be the rare few who read an x% efficiency gain as "let's eat x% more of the market" instead of "let's eat x% of our staff."
Thin-slice, boring problems get AI solutions, and those startups sprint straight for acquisition. Expect a ton of acquisitions — far, far more than the tiny handful of new AI billionaires minted. Lots of small exits, very few rocket ships.
AI-native products and wearables start reaching consumers in Q1. The category arrives early. Whether it sticks this year is a different question — but the door opens in the first quarter.
So set your watches
That's the flag, planted. Some of this will age like wine and some like milk, and the only honest thing to do is come back mid-year and mark my own homework — in public, no grading on a curve. I'll do exactly that.
Because here's the thing I keep circling: the timelines are collapsing into each other. AI, robotics, quantum, energy, and maybe-AGI used to be separate horizons and now they're close enough to overlap into compounding growth. That's either a Star Trek outcome — utopia, some distant new friends, some guy named Scotty beaming us up — or it's the other self immolation thing, and we do the immolation thing far more competently than we do utopia.
Either way, the move is the same: build forward, build fast, stay curious, stay humble, and keep experimenting. Run the Red Queen's race and try to outrun both endings at once.
See you at the halfway mark.