Frontier AI in 2026 – What’s Not Going to Change

So far in this series, I’ve looked at how AI capabilities are evolving and what that implies for 2026. Equally important is what didn’t change in 2025—and is therefore unlikely to change next year either.

AI will remain unreliable

Generative AI will continue to hallucinate, be overconfident, and be difficult to fully secure. There will be incremental improvements, but no breakthrough that suddenly makes AI “safe by default.”

As a result, AI will remain boxed in by guardrails, human oversight, and scaffolding. Full automation will stay hard; augmentation and assistants will continue to dominate real-world use.

Organizations will keep struggling to absorb AI

AI progress will keep outpacing organizational change. Adopting new ways of working takes time: processes, tools, governance, and people don’t transform overnight.

This creates a persistent capability overhang—things AI can do long before most companies can use them effectively. That gap didn’t close in 2025, and it won’t close in 2026.

AGI that replaces jobs isn’t imminent

AI is getting better at automating tasks, not entire jobs. There is still no system that can take full responsibility for most roles end to end.

That kind of AGI would change everything—but it’s not in sight.