With everyone and their mother making predictions about 2026, I figured I’d throw my metaphorical hat into the ring and share what I see coming.
1. Small models keep winning.
A new technique from a paper by a Chinese lab dramatically reduces the cost of training. This forces a reset in how model development is priced and planned.
Lower-parameter, lower-cost models continue to improve.
2026 is the year this starts to become obvious. Practical agent systems and agentic browsers become normal.
2. Agents grow up.
Agents exploded in 2025, but most workflows relied on methodologies that were three to five years out of date.
In 2026, that gap closes... abruptly.
3. A new agentic tooling standard appears.
At least one new standard is proposed and adopted by nearly everyone.
It offers cleaner abstractions and more flexibility and less abstraction than MCP. Existing tooled systems start to look quaint by comparison.
4. China makes a chip power play.
Mid-year, and two years ahead of industry expectations, China moves aggressively to control the chip market.
The chips run hot and have non-trivial reliability problems. But... they are cheap enough to dominate American and European markets anyway, especially at the low end.
This triggers a cascade.
The edge explodes. Compute decentralizes at scale. Token costs collapse.
Data centers stop looking essential.
Business models built on token pricing fail quickly. Creative AI moves decisively to local hardware.
The number of self-described AI experts also explodes until the term loses all meaning. This is healthy.
5. AGI discourse stalls.
There will be no meaningful progress in the conversation around AGI and superintelligence. It stays loud, circular, and unresolved.
6. AI appliance PCs drop in price.
Dedicated AI machines move into reach fast.
Everyone you know will own at least one.
7. Apple does Apple things.
Apple publishes another paper about LLMs. It irritates everyone other than LinkedIn AI skeptics.
Later in the year, Apple ships new VR or AR hardware and the paper disappears.
8. Resistance in creative fields collapses.
Opposition to AI in music and other art forms beyond conventional illustration collapses quickly. Diffusion and LLMs become standard creative tools.
9. The power problem is solved, mostly.
In 2026, we'll start to see the light at the end of the tunnel here, and a way forward that might make sense. It's enough to stop dominating every serious discussion.
10. 3D and AI converge.
New world models accelerate the merge. A new version of Unreal Engine rewrites assumptions about how game engines work.
The fight over it is loud and public.
Looking ahead
All told, 2026 looks like an exciting year.