Most Improbable Agent of You

Author: Kevin Kelly URL: https://kevinkelly.substack.com/p/most-improbable-agent-of-you Saved via: Kelly’s own tweet, October 21, 2026 One-line: AI today is the most centralized it will ever be; shrinking cost per capability will let everyone own and train a frontier-class model, and the AI worth having is not the best one but an “exoself” tuned to make you the most improbable version of yourself.

Key claims

• Today’s AI is the most centralized it will ever be. The scaling law (more data, more compute, more inference) produces data centers with millions of chips, trillions of parameters and gigawatts of power costing tens of billions of dollars, so AI power sits in very few organizations.
• Biology is an existence proof against centralization. Each human runs a supercomputer “better than most AIs” on about 25 watts, and there are some 8 billion of them networked together.
• Cost per capability halves roughly every 6–7 months. Algorithmic gains cut the compute needed for a given capability about in half every 8 months; combined with Moore’s Law, Kelly estimates that within a few years a model as capable as today’s leader (“let’s call it Opus”) could run on a device you own, and within about a decade an individual could afford to train one from scratch.
• Data centers may be temporary scaffolding. By analogy to the phone (whose graphics chip would have filled a room 25 years ago), the controversy over massive data centers “disappears because they disappear.”
• “Good enough” is coming, like bandwidth. Past a threshold, a local model will cover 99% of what most jobs need, the way a gigabit connection covers most tasks.
• Much AI criticism is really criticism of concentration. Who benefits, who controls access, who trains and supervises the model: Kelly argues many of these questions change if AI is generated, overseen and maintained by individuals, and he prefers “a billion AIs running the world” to three or four. He concedes individual AIs bring their own trust problems.
• The goal is the most improbable you. Building on his manifesto Your Most Improbable Life: agents take the tasks anyone can do and leave the tasks only you can do, so each person invents a unique niche. “You are not the best, you are the ONLY.”
• Training your personal AI becomes a life skill. Kelly wants an AI trained on his own experiences, influences, work and heroes. He predicts teacher AIs for personal AIs, schools of thought on how to educate them, and a self-help industry around “cultivating the Agent of You.” Frontier models trained on everything are generalists “hitting the average in the middle of the mainstream.”

Notable quotes

It is commonly said that AI today is the worst it will ever be. This is true. AI today is probably the most centralized it will ever be as well.

It is likely that the entire controversy around massive data centers disappears because they disappear; they were a temporary scaffolding needed for the first round of tech, just a blip in the history of technology.

A lot of the criticism of AI today is really just a criticism of capitalism.

But I find a vision of a billion AIs running the world much preferable to 3 or 4 of them running the world.

But we want a personal AI, an exoself, that is as unique as we are. We don’t want the best AI, we want the ONLY AI.

The art of training your own personal AI to be weird yet useful will become one of life’s skills.

How it connects

• Kevin Kelly: continues his long line of technology-trajectory essays; compare The Technium and How Kevin Kelly Changed His Mind.
• AI Agents: Kelly’s version of the agent future is personal and idiosyncratic rather than a few general-purpose assistants.
• Decentralization: the core argument is that the cost curve decentralizes AI the way it decentralized computing.
• The End of Startups: Evan Conrad’s post the next day argues the near-term opposite, that compute is scarce and rationed by credit. Both can be true on different time horizons.