Agentic AI: the age of autonomous intelligence
Going into 2026, we will witness a continued shift from assistive AI to agentic AI. Not copilots that wait for instructions, but systems that can entirely think, act, and finish work on their own.
What is agentic AI?
Copilots draft, suggest, and wait for you. Artificial intelligence (AI) Agents take a high-level end goal and attempt to achieve it work without constant supervision, handling full tasks, and operating with the advantages of software. In short, control moves from ‘human-in-the-loop’ to ‘human-in-command’.
Why does this matter?
Progress has always been held back by time. Humans can reason, imagine, and experiment, but a relatively slow pace. Agentic AI can analyse, test, and iterate without ever stopping. The real gain comes from speed, not necessarily its ability to outthink humans.
Is this a new industrial revolution?
The steam engine multiplied physical force. Agentic AI multiplies cognitive work. General models gave us raw capacity, but specialised agents offer the real value. Once systems can learn continuously and scale instantly, entire sectors may reorganise around these capabilities.
How are tech giants positioning themselves?
Google and Meta favour systems that act for the user. Apple and Microsoft focus on tools that amplify the user. OpenAI and Anthropic live in the middle. The outcome will depend as much on business models as on technical progress.
Are we still underestimating AI’s curve?
Benchmarks say yes. The curve is unmistakably exponential, with capabilities doubling roughly every seven months. By mid-2026, models are projected to sustain full eight-hour workdays autonomously. By 2027, they could routinely outperform experts across a wide range of industries.
What happens next?
Intelligence becomes abundant and specialised. It blends into daily work the way electricity once did, running in the background and substantially extending what humans can get done.
What does this mean for investors?
We are still early. We currently prefer focusing on the infrastructure that makes AI possible: chips, energy, transmission, storage. That is the real bottleneck. However, we also see selective opportunities among companies already putting AI to productive use.
13 febbraio 2026
