Everything started with one conjecture:

Conjecture (Post-AGI Economics): Since P(dominant agent execution within finite T)>0, responsible strategy requires planning for the then-world—where any workflow with a measurable V, evaluators, and actuators is executed by agents. In that world, economics pivots from "can agents do the work?" to "who sets V, who owns the capital/models/data, and how do prices and policy route the surplus?"

This single insight led to the entire framework below.

What this is: an open, testable framework for thinking about when software agents take over execution of valuable work—and what that does to the economy. No hype, no vibes. We turn the question into measurements, a countdown, and scenarios you can actually argue with.

• Core idea: If you can (1) say what "good" looks like, (2) score it, and (3) give the agent the tools, then agents can match or beat human execution. What stays human is governance (deciding what "good" is), not the keystrokes.
• We don't pick a date. We compute a mechanical countdown from a data-driven model and update it as signals change.
• Two endings matter: a future where buyers are rich (surplus is routed back to households), and a future where buyers are thin (output rises but demand erodes). Our model tracks probabilities of each.

The original conjecture breaks down into three testable components:

1. Agent Execution Hypothesis (AEH) — "any workflow with a measurable V, evaluators, and actuators is executed by agents"
   If a task has a clear success metric (V), a reliable way to score outputs (evaluators), and the agent has the right permissions/tools (actuators), then the agent can, in principle, do the job as well as or better than a human working under the same constraints.

2. Finite-Window Hypothesis (FWH) — "responsible strategy requires planning for the then-world"
   Changing rules, infrastructure, and safety rails takes time. Because tech rolls out fast and systems "lock in," there's a finite window to set up evaluation, actuation rules, and fair distribution before the new equilibrium hardens.

3. Surplus Routing Principle (SRP) — "how do prices and policy route the surplus?"
   Automation only helps society if buyers can still buy. The economy works when productivity gains are routed back to households (wages, dividends, transfers). No buyer floor → thin demand. This is the economic pivot: from "can agents do it?" to "who owns it and how is surplus distributed?"

• We model the world as regimes (e.g., "glide," "sprint," "grid-choked," "governance-bounded") and let data decide which we're in.
• At each update, we compute a countdown to "agent-dominant execution" using standard absorbing-Markov math. It's a clock, not a prophecy: new data → new clock.
• There are two absorbing end states:

• Buyer-Rich (agents dominate and households keep purchasing power)
• Buyer-Thin (agents dominate but demand thins out)

We publish the split probabilities for both each release.
• Signals that move the countdown include: compute/energy constraints, algorithmic efficiency, automation share in production, governance gates, buyer capacity, and market concentration. The exact list is fixed by a Pre-Analysis Plan.

• Is: logic-first scaffolding, math you can check, a transparent update process, falsifiable claims.
• Isn't: a prediction about an AGI date, a sector hot-take, or a "% jobs replaced" blog post.

• Start with the navigation index:
Thesis/Index.md (complete chapter guide and overview)

• If you want the intuition:
Thesis/01-Introduction.md → Thesis/12-Conclusion.md

• If you want the engine:
Thesis/04-Absorbing-Math-Clock.md (the countdown)
Thesis/05-Path-Dependence-Lockin.md (why the window is finite)

• If you want to plug in data or contribute:
Thesis/08-PAP-Preanalysis-Plan.md (signals, lags, rules)
Thesis/09-Implementation-Runbook.md (what we publish and how)
Contribute.md (ground rules and contribution workflow)

The original conjecture asks the critical question: in a world where agents execute all measurable work, "who sets V, who owns the capital/models/data, and how do prices and policy route the surplus?"

"AI takes your job" is only half a sentence. If the system doesn't route surplus back to people, who buys the output? Our framework makes that question measurable: it pairs the execution clock with a buyer floor and shows which end state we're tracking toward.

The economics pivots from capability ("can agents do the work?") to coordination ("who decides what work gets done and who gets the gains?"). This is why we track both the countdown AND the absorption probabilities—timing matters, but so does the path we take.

This is open source. Keep changes evidence-gated and spec-consistent: • Don't hardcode timelines or probabilities in prose—numbers are injected by the pipeline.
• Symbols and acronyms must match Thesis/G-Glossary-Notation.md.
• Changes to signals or estimation require a new PAP.
See Contribute.md for the full checklist.

If you use this framework, cite the repo and the references listed in Thesis/References.md (Hamilton 1989; Filardo 1994; Grinstead & Snell; Rabiner; Acemoglu & Restrepo; Hayek; Gneiting & Raftery; Diebold & Mariano; etc.).

This project is licensed under the Apache License 2.0.

Author: Harsh Joshi ([email protected])
Maintainer: RawThoughts Enterprises Private Limited (RTEPL)

Questions, critiques, or data you think should be a signal? Open an issue. The goal isn't to win a debate—it's to make the debate measurable.