Alber

🆕 Codex now officially supports skills Skills are reusable bundles of instructions, scripts, and resources that help Codex complete specific tasks. You can call a skill directly with $.skill-name, or let Codex choose the right one based on your prompt.

The Negotiation Revolution Nobody's Talking About I keep seeing the same weird pattern in AI governance discussions... Everyone's panicking about AGI centralizing power, building digital panopticons, requiring massive government control. But what if we're looking at this completely backwards? What if AI's real superpower isn't control, but coordination? Here's the thing that's been nagging at me: We've been stuck with the same dreary choice for centuries. Either suffer through market failures (your neighbour's leaf blower at 6am, anyone?) or submit to ham-fisted government rules. Nobel laureate Ronald Coase spotted the problem decades ago. In theory, you and your leaf-blowing neighbour could just... negotiate. You pay him to stop, or he pays you to tolerate it. Everyone wins. Except it never happens. Why? Transaction costs. It's too bloody expensive to find each other, hash out a deal, and enforce it. So we default to blanket bans or resigned suffering. But that's not the interesting part... What if every person had a hyper-competent AI agent - not some corporate spy, but a genuine digital advocate that knew your preferences down to the penny? An agent that could negotiate with millions of other agents in milliseconds? Suddenly, Coase's impossible dream becomes Tuesday afternoon. Your agent knows you'll tolerate construction noise on weekdays but not during your baby's naptime. The construction company's agent instantly sees a higher "price" for those sacred hours. They either pay up or reschedule. No council meetings. No angry letters. Just instant, precise coordination. Think about how this changes... everything. That hideous tower block developers want to build? Instead of a political bun fight, it becomes simple economics. The developer's agent must discover the actual price at which every single resident would accept it. If residents truly value their neighbourhood's character, that price will be astronomical. The project only proceeds if the developer values the location more than residents value the status quo. Either way, the true costs are revealed and paid. "But wait," you might think, "couldn't wealthy people just buy their way through everything?" Here's where it gets clever... These agents don't just do individual deals. They can form instant coalitions. Picture 10,000 residents whose agents spot a factory's pollution proposal and instantly band together: "Our collective price is £50 million. Non-negotiable." The poor suddenly have the coordination power of the rich. Even better: wealthy areas can't just export their problems anymore. A billionaire closing a public beach must make a public, auditable offer to every single person affected. The cost of selfishness becomes explicit. (Though knowing billionaires, some will still pay it. At least then we're compensated, eh?) Now you might wonder: "What about the really big stuff? Climate change? Pandemics?" Fair point. You can't negotiate with someone planning to release a bioweapon. That's where the state still matters - but in a leaner, focused way. The state stops trying to micromanage parking disputes and focuses on what only it can do: preventing violence, enforcing contracts, stopping existential threats. Think of it as the state becoming a bouncer instead of a busybody. The framework even suggests how agents themselves stay aligned... It's not one giant AI overlord (terrifying) but millions of personal advocates operating within nested constraints: law (outermost boundary), market competition (middle layer), and your preferences (innermost core). Like Russian dolls, but for governance. Each layer keeps the others in check. Here's what really gets me: This isn't just about efficiency. It's about revealing the true shape of human preferences that have been hidden by transaction costs for centuries. What do people actually value? What trade-offs will they really make? We're about to find out. Imagine the implications: Sugar taxes? Your health insurer's agent offers immediate premium discounts if your agent limits junk food purchases Airline peanuts? Agents negotiate flight-specific policies based on actual passenger needs Urban planning? Real-time preference aggregation replaces five-year-old council votes We're talking about governance that adapts as fast as human needs change. A permanent referendum on everything, but one that actually works. The mad thing is, this starts small. Office thermostats. Shared parking. Roommate chores. As these systems prove themselves, they scale up. Same mechanism that solves your neighbour dispute could one day help manage river rights between nations. Makes you wonder what else we're missing... What other "impossible" coordination problems are actually just transaction cost problems in disguise? What other human potential has been locked away, waiting for the right key? If you're working on AI alignment, maybe the question isn't "how do we control AGI?" but "how do we use AGI to finally coordinate properly?" Different question. Different future. Try this thought experiment: Next time you're annoyed by an externality - noise, pollution, whatever - ask yourself: "What would I actually accept to tolerate this?" Then imagine a world where that question matters. That's the world we could be building. (And yes, I know this sounds like techno-optimism gone wild. But compared to the "let's centralize all AI in a government bunker" proposals floating around, I'll take distributed negotiation over centralised control any day. At least this way, we muddle through together.)