Every time someone uses AI at work, they create something valuable. Prompts, corrections, workflows, and judgment. That has created a new ownership fight. Model providers want to learn from that data. Companies want to keep it.
Microsoft’s Satya Nadella just wrote about why companies need to own their learning loop. His warning is that companies should not have to give away the learning that makes them unique just to use AI. He describes two forms of capital a firm needs. Human capital is the knowledge, judgment, relationships, ingenuity, and pattern recognition of its people. Token capital is the AI capability it builds and owns through its models, agents, data, evals, workflows, and institutional memory.
But that is a firm-level framework. It explains what the company needs to protect and compound. It leaves out the people whose work is creating both. What do they get to keep?
I think the answer is personal capital. It’s the intelligence you own, the judgment you embed in it, and the economic upside you retain as it creates value. Personal capital is everything that builds up for you through work. The salary you earn, the equity you vest, your reputation, relationships, ownership, and track record. In the AI era, it may also include something new: a personal harness, an agent, memory, evals, and decision rules that carry how you think from one job, company, and model to the next.
Thinking Machines manifesto gets at why this matters. AI should extend human will and judgment, not flatten everyone into one centrally chosen system. A personal harness is how an individual keeps shaping the intelligence that acts for them.
AI makes this question more urgent. As execution gets cheaper, the valuable part of work shifts from producing output to exercising judgment. If that judgment compounds only inside the company, the person gets better at the work while the asset gets built somewhere else.
So what could change? I do not know what the final version looks like. Company IP is real, important work is done by teams, and no one should get to copy a company’s systems when they leave. But I can imagine at least five things that sound strange today becoming normal:
Bring your own agent. Your work agent stays with the company. Your private harness, memory, preferences, methods, and long-term goals follow you.
Personal learning accounts. Employers fund a private agent, memory, compute budget, and eval environment that belongs to the employee, much like a 401(k). It never contains company data, but it gets better with the person over time.
The right to fork. Not the right to take code or customer data. The right to retain the general methods, personal evals, and non-proprietary judgment systems you built while working.
Personal evals become part of a resume. The proof of judgment will not just be a list of companies you worked at. It will be a system that shows what you think good looks like and how you make decisions.
Vesting becomes less about time. Four years at a company is a crude proxy for contribution when someone can improve systems that run long after they leave. Some form of learning dividend, equity, or durable-outcome compensation could emerge.
The future worth building is not one where the model learns, the company compounds, and the person starts over. It’s one where human, token, and personal capital all compound together.