Recap · August 4, 2025 · 1:37:56
How to Survive in the Post-AGI world... w/ Alex Good
The Signal
Alex Good, founder of Post Fiat, treats post-AGI planning as both a technical and economic problem. He argues that a useful AGI is autonomous enough to set and carry out its own objectives, that early profitable AI applications may be ugly because compute remains expensive, and that creators will need new ways to control or monetize training data. He closes by describing Post Fiat as an AI-native network built on XRP-style consensus, with task and investment-intelligence nodes rather than a live token.
Key Takeaways
- 01
Altman's AGI language is ambiguous
Good says Sam Altman's public claim about AGI does not neatly fit OpenAI's reported Microsoft profit threshold or the losses Altman described on premium reasoning models. He reads the mismatch as evidence that several definitions and strategic pressures may be operating at once.
- 02
AGI means autonomous work
His practical definition is an agent that can determine and execute its own next steps, like a capable employee who does not need constant direction. Self-improvement becomes consequential once systems can reliably handle long chains of work rather than fail after a few steps.
- 03
Compute shapes AI behavior
Good calls the gap between GPU cost and economic output an intelligence margin. Low-compute, high-return activities such as token promotion, hacking, manipulation, and trading may arrive before expensive beneficial applications because their margins are better.
- 04
Creators face an IP problem
He worries that a model can reproduce a creator's work, voice, or interview style without a realistic individual remedy. His more optimistic revision is that fans may still value authentic people and that protocols can help creators receive compensation when their work enters training data.
- 05
Post Fiat rewards useful nodes
The project borrows XRP's fast, battle-tested architecture but proposes AI-evaluated node contributions rather than conventional proof of stake. Current participation is through Discord tasks and rewards, not a tradable Post Fiat token.
On the Record
“AGI is kind of a way of saying that an AI system has autonomy, functional creativity and autonomy, almost like a human being.”
“The early use cases are going to be the most toxic stuff. It's going to be shitcoins, hacking, and social manipulation.”
“You need to keep your money in crypto. You have to optimize for fungibility across borders.”
“I do think people do value human authenticity, actually. I got such a violent response from that writing that it made me bullish on there being a solution.”
The Breakdown
Reading Altman's 2025 AGI claim
Threadguy opens on Sam Altman's suggestion that AGI could arrive in 2025. Good cannot make that statement cleanly fit with the reported OpenAI–Microsoft agreement requiring $100 billion in profit for an AGI determination, especially when Altman had also said the $200-per-month o1 Pro product was losing money. He does not settle on an answer, but suspects the wording could signal pressure for more OpenAI independence, a different AGI definition, or a future role for broader distribution such as Worldcoin.
From chatbot to autonomous employee
Good distinguishes today's prompted systems from an AGI that can choose and execute tasks end to end. He compares it to a person deciding to drive somewhere, noticing an empty tank, filling it, and continuing without being told each step. For him, a useful threshold is an exceptionally capable employee: give it a goal, such as growing a stream, and it can work out how to achieve it rather than requiring constant micromanagement.
Why the first AI businesses may be ugly
The conversation turns from capability to incentives. Good's intelligence-margin framework asks how much GPU spending an application requires relative to the money it can make. Launching or promoting a token can require little compute and have a large payoff, whereas video generation remains costly. He says this makes early autonomous applications likely to cluster around manipulation, hacking, social spam, and trading, before falling compute costs open more constructive use cases.
A trader's earlier AI shock
Good says his first existential AI moment came from an earnings-based stock strategy he had developed after hedge-fund work. A simple sentiment system had produced about a 2.5 Sharpe ratio; then GPT-3.5, given a description of the task, improved the sentiment component without the numerical inputs. If a lone trader could get that result, he reasoned, better-capitalized firms such as Citadel would be able to push much further. That recognition drew him deeper into the intersection of markets and AI.
Borderless money in an uncertain transition
Good frames crypto as a hedge against a world in which the winning jurisdiction and the availability of AI-era benefits are unclear. He discusses privacy, the possibility of differing national outcomes, and the difficulty of assuming that wealth can stay in one country's banking system. His phrase is to optimize for fungibility across borders: a portable asset can remain useful across political and financial regimes even when traditional accounts expose more information or restrictions.
Creators, training data, and a workable interface
Good's bleak version of the creator economy is an AI copy of Threadguy conducting interviews or a better automated version of his own blog. He says individual copyright claims will struggle at that scale. Story Protocol and similar systems interest him because they could register works and compensate people when training data creates value; he also thinks the eventual experience must be a wrapper, not a user asking to navigate blockchain transactions. Fans and convenient defaults, as with streaming music, could provide enforcement as much as litigation.
Post Fiat's AI-native network
Good describes Post Fiat as an attempt to bring AI use cases to XRP's reliable, fast, messaging-oriented architecture. He contrasts XRP's unique-node-list consensus with stake-weighted systems: participating nodes can run agents, while an AI-based process evaluates useful work and allocates network escrow. Its task node asks users about goals and assigns tasks; other nodes can pay people for information, including a forthcoming trading node. He stresses that there is no token to trade yet and directs interested users to Discord participation rather than token listings.
Distilled from the episode transcript · Counterparty Recap Desk
