MACHINE LABOR / MEASURED EVOLUTION

Intelligence.
Into useful work.
At growing scale.

Coordinate specialized agents. Make every improvement measurable. Turn accepted work into the capacity to do more.

Seeded research simulation · Local analysis · No payments

HGMLEARN FROM
REVIEWED OUTCOMES
01 / WORK02 / EVIDENCE03 / REVIEW04 / IMPROVE

Bounded exploration. Accountable execution.

What this console proves

Reproducible scheduling experiments and checked analysis files. The recorded outcomes are synthetic. They do not establish universal task competence, autonomous self-improvement, customer revenue or production readiness.

01 / OBSERVATORY

Make improvement accountable.

Compare the hierarchy with a greedy baseline. Inspect completed costs and reservations before interpreting any economic lift.

Loading recorded simulation…

Gross value / completed cost

HGM Baseline

This multiple is the simulator’s “ROI” field. Net return is (gross value − cost) / cost. Neither represents real revenue. Missing ratios mean no completed cost.

Each strategy has its own step coordinates. Same budget ceiling; different action rates and random streams. Compare across many seeds before drawing conclusions.

Read exact strategy totals
Simulated values in USD-equivalent units; not USDC balances
Strategy Gross value Completed cost Reserved Value less all commitments Pending tasks

Lineage performance

Recent strategy events

FROM RECORD TO DELIVERABLE

Produce a checked benchmark analysis.

Generate a real JSON analysis of this simulation, or export the exact task for an admitted worker. Candidate review checks arithmetic and source binding; a separate reviewer must assess usefulness.

Select a record to begin.

02 / THE MACHINE LABOR LAYER

A work contract.
A reviewable result.

AGI Jobs is designed as a scalable machine labor layer for authorized, lawful screen-based work—coordinating specialized agents to execute tasks, produce reviewable evidence, and support independent verification and settlement.

Define your next work order

A draft does not authorize execution or payment.

A

OpenClaw worker

Use the repository’s admitted Responses adapter. A protected profile binds the job, task digest, deployment and bounded execution. Browser, files, code and tools depend on the commissioned runtime.

B

ChatGPT Work session

An operator can authorize a scoped session across supported apps, browser and computer tools. Export candidate files for review. There is no assumed remote Work dispatch API.

C

Independent verification

Check evidence, reproduce calculations and test the deliverable. Keep creator, checker, unrelated reviewer and settlement signer responsibilities distinct.

03 / GOVERNED EVOLUTION

The loop, kept visible.

HGM explores a lineage of simulated candidates. The thermostat adjusts search pressure. The sentinel and owner controls bound work. Only verified, authorized changes belong in a live system.

Preserved evolution flow

graph LR
 A[Start: Load Config] --> B{Thermostat}
 B -->|ROI >= Target| C[Increase Concurrency]
 B -->|ROI < Target| D[Constrain Exploration]
 C --> E[HGM Engine]
 D --> E
 E -->|Expand| F[Self-Modification]
 E -->|Evaluate| G[Mission Execution]
 F --> H[Ledger Update]
 G --> H
 H --> B
 H --> I[Sentinel Rules]
 I -->|Safe| B
 I -->|Halt| J[Graceful Shutdown]

Original conceptual flow retained. In this simulator, “self-modification” mutates a quality parameter and “mission execution” samples an outcome; neither changes production code nor executes a customer job.

From authorized work to accepted value

flowchart TD
 A[Buyer scope and approved inputs] --> B{Budget and reviewer available?}
 B -->|No| C[Defer with reason]
 B -->|Yes| D[Admitted worker and bounded tools]
 D --> E[Artifacts and effect journal]
 E --> F[Reproducible checks]
 F --> G{Unrelated reviewer accepts?}
 G -->|No| H[Repair or dispute within limits]
 H --> F
 G -->|Yes| I[Buyer acceptance]
 I --> J{Settlement separately authorized?}
 J -->|Yes| K[Verified settlement process]
 J -->|No| L[Hold without payment]
 K --> M[Measured outcomes inform next allocation]
Open the preserved Grand Operator Console ↗

04 / A LONGER HORIZON

Useful work compounds.
Capacity follows evidence.

Build toward a civilization-scale productive network: better software and research, more capable infrastructure, and eventually new energy and industrial capacity. Expansion must follow demonstrated value, available resources and accountable governance.

Explore the market assumption

The $40 trillion annual screen-work opportunity is a user-supplied planning scenario, not a measured addressable market or revenue forecast. Eligibility, demand, delivery costs, review capacity and competition all reduce realizable value.

ILLUSTRATIVE ANNUAL WORK VALUEOpportunity × eligible share × captured share.
Gross work value, before all costs; not platform revenue.

05 / START HERE

From first experiment to a qualified worker.

01

Reproduce the comparison

Python 3.12, from the repository root. No model key is needed.

python demo/Huxley-Godel-Machine-v0/run_demo.py --seed 7

Import the generated web/artifacts/comparison.json above. Try a different seed or a smaller budget.

02

Inspect and qualify

Export the analysis task. Commission a dedicated worker with explicit permissions, approved inputs, spend limits and an unrelated reviewer. Keep tokens and profiles off this site.

Read the complete operator runbook ↗
03

Review before expansion

Verify the output and actual effects. Measure useful deliveries, cost and reviewer time. Reconcile unknown outcomes before retrying; authorize settlement separately.

Source, tests and validation ↗
Current integration references · checked 6 October 2026

Tool availability broadens the work that can be attempted. Each workflow still needs qualification and acceptance evidence.