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.
MACHINE LABOR / MEASURED EVOLUTION
Coordinate specialized agents. Make every improvement measurable. Turn accepted work into the capacity to do more.
Seeded research simulation · Local analysis · No payments
Bounded exploration. Accountable execution.
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
Compare the hierarchy with a greedy baseline. Inspect completed costs and reservations before interpreting any economic lift.
Loading recorded simulation…
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.
| Strategy | Gross value | Completed cost | Reserved | Value less all commitments | Pending tasks |
|---|
FROM RECORD TO DELIVERABLE
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
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.
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.
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.
Check evidence, reproduce calculations and test the deliverable. Keep creator, checker, unrelated reviewer and settlement signer responsibilities distinct.
03 / GOVERNED EVOLUTION
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.
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.
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]
04 / A LONGER HORIZON
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.
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.
05 / START HERE
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.
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 ↗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 ↗Tool availability broadens the work that can be attempted. Each workflow still needs qualification and acceptance evidence.