AI Predictions for 2026
It’s a new year and time for new AI predictions. Let’s start by looking back at what I expected to happen last year, and what actually happened. After that, we’ll move on to predictions for 2026.
Looking Back at 2025
The year AI influencers take over
Partially correct, but limited in scope. AI‑generated personas and influencers were used more than before, but adoption concentrated almost entirely in adult(ish) content. My original bet was on product placement tools enabling monetizable AI‑generated content at scale, but that did not materialize. And outside of pure content generation, AI influencers did not meaningfully break into mainstream in 2025.
Humanoid robots will level up to real-world impact
This prediction largely held. 2025 was the year of large‑scale proof that humanoid robots can operate reliably in constrained environments. Figure robots participated in building over 30,000 BMWs, Tesla ran specialized Optimus test lines, and Digit handled more than 100,000 warehouse fulfillment operations. This did demonstrate the viability in scale, which is the required step before wider rollout.
Software development is about to accelerate and get interesting
Correct. Models became fast enough for continuous use, context windows grew large enough to support real tasks, and tooling matured real quick. Another part of this prediction concerned long‑term reasoning, which also proved to be accurate. By the end of the year, models were increasingly evaluated by how long they could work on a problem without losing focus.
Predictions for 2026
Another year of “AI progress is plateauing” takes
Progress will appear slower because improvements are spreading out and becoming less flashy. Most gains will come from reliability, integration, and longer‑running systems rather than new headline capabilities. This gap is common when technologies start to mature and get into operational use, looking less like breakthroughs and more like quiet compounding.
Agentic AI goes mainstream
Agentic systems will start to see wider adoption in restricted use cases and environments. These systems will not resemble full autonomous employees, but they will reliably execute multi‑step tasks within tightly scoped environments, shifting human roles more toward oversight and decision‑making. Focus starts to move to scaffolding platforms (“ChatGPT wrappers”) which will add value through narrow orchestration, system access, and tool integration. Specific to a task.
AI video in advertising becomes normal
AI‑generated and AI‑augmented video will become a standard part of advertising workflows. Most of the underlying mechanics are already validated, and in late 2026 they expand into brand‑safe contexts. The majority of output will be low‑cost, fast‑iterated, and performance‑driven rather than creatively ambitious. Not impressive. Not controversial. Just normal. And often barely noticeable.
Software development shifts to ticket‑driven workflows
High level tickets increasingly become executable units of work. Development workflows move toward ticket‑driven, agent‑executed implementation, where humans define goals, constraints, and agents handle execution. Agents also begin to demonstrate a basic understanding of architectural structure, both during implementation and in review.
The year of AI code review
AI‑assisted code review sees meaningful adoption. Compared to code generation, review delivers faster and more consistent quality improvements by enforcing standards, detecting errors, and reasoning about changes in context. It also becomes a necessary control mechanism for agent‑driven development.
First steps beyond toy-level non-technical development
While non‑technical users can already build with AI, current results are still mostly fragile and often unsuitable for production systems. Ticket‑level agent execution combined with improved AI code review enables the first real production‑grade outcomes, even if still limited in scope.
Next-generation humanoid robots start delivering real impact
Humanoid robot deployments move from scaled testing into operational use in specific environments. The limiting factor is not raw capability, but identifying tasks that align well with current learning methods and system constraints. Where those conditions are met, robots begin delivering measurable economic value.
Final thoughts
If 2025 was mostly about proving things can work, 2026 will be about figuring out how they can reliably produce value in the messy reality we actually operate in.
