AGI JOBSDEMO OBSERVATORY
← AGI Jobs home

FROM INTENT TO ACCEPTANCE

Design the work.
Define the proof.

Turn a useful outcome into a scoped work proposal. Download the exact task for operator review, then hand it to an authorized worker with clear acceptance criteria.

Build a work order ↓Browse ten categories ↓Execution guide ↗

Local planning only. No account, wallet or provider connection is needed. This page does not fetch your sources or dispatch work. Do not enter secrets or private data.

YOUR NEXT USEFUL JOB

Make success inspectable.

Save an editable draft to continue later.
Changing a field requires a new work order.

Pick up where you left off.

Save even an unfinished draft, then open it here to continue. Files contain your entered text and stay on your device. Reloading clears the form. Drafts carry no execution approval.

State the outcome the buyer will use. Your custom objective is kept when you change category.

One HTTPS URL per line. Up to 20. No credentials, query strings or fragments. The operator must verify rights, availability and the derived origin allowlist.

For OpenClaw, this must match a protected operator profile. A profile name does not create or authorize a worker.

An unfunded commercial proposal, separate from provider charges. Existing v2 contracts here use 18-decimal AGIALPHA; verify the real deployment currency before settlement.

These are planning allowances. Configure and verify actual runtime time, spending and stop controls separately.

TEN PRACTICAL STARTING POINTS

Work a buyer can use.

Templates describe proposed work.
They are not claims of completed customer engagements.

01

Performance optimization

A reviewable patch and reproducible before-and-after benchmarks.

Acceptance criteria
  • Reproduce the baseline and changed benchmarks with the same inputs and environment.
  • Pass the existing correctness tests and report regressions and resource costs.
02

Open-source feature

A patch, usage instructions and acceptance-test results.

Acceptance criteria
  • Demonstrate every agreed acceptance case, including failure cases.
  • Pass the relevant existing tests and document compatibility and setup.
03

API and SDK tooling

An integration patch with executable examples and error handling.

Acceptance criteria
  • Validate requests and responses against the pinned specification.
  • Exercise authentication failures, timeouts and rate limits using local fixtures; make no live purchase or account change.
04

Automated test suite

A test patch and evidence that the tests detect the target defect.

Acceptance criteria
  • Show the relevant test failing against the known defect and passing against the correction.
  • Run in a clean environment and document remaining coverage gaps.
05

AI evaluation package

Versioned cases, scoring rules and a limitations report.

Acceptance criteria
  • Separate evaluation data from tuning data and record source rights.
  • Recompute scores independently and report sample size, uncertainty and failure cases.
06

Public-data dashboard

Reviewable application source, normalized data and reproducible calculations.

Acceptance criteria
  • Reconcile displayed values and exports with the exact source data and units.
  • Exercise empty data, invalid inputs, keyboard use and mobile layouts.
07

Interactive application

Editable application source, setup instructions and a working user journey.

Acceptance criteria
  • Run the complete agreed user journey from a clean start.
  • Check boundary cases, accessibility and exports; distinguish computed results from examples.
08

Vendor and product intelligence

A dated comparison table and a source-linked recommendation.

Acceptance criteria
  • Cite a dated primary source for every material comparison and mark unknown values.
  • Recompute totals and eligibility independently; do not purchase, contact vendors or create accounts.
09

Executable documentation

An editable guide with checked commands and recorded results.

Acceptance criteria
  • Run the documented commands in order and verify their expected outputs.
  • Explain prerequisites, error recovery and version-specific limitations.
10

Scientific reproduction

A reproducible analysis, numerical results and environment record.

Acceptance criteria
  • Pin source data and the computational environment; specify units and numerical tolerances.
  • Have an independent reviewer rerun the calculation and explain any discrepancy.

BEFORE EXECUTION

From a draft to delivered value.

Why use AGI Jobs when a chatbot can help?

The intended value is a complete work lifecycle: a scoped outcome, assigned execution, reviewable artifacts, independent acceptance and accountable settlement. Use a chatbot directly when a conversation is enough. Use the job workflow when the deliverable and its verification need to stand on their own.

Can an agent do every task done with a keyboard and mouse?

Browser, desktop, code and file tools broaden the possible work substantially. Success still depends on available capabilities, permissions, inputs and measurable task performance. Qualify each work category with representative tasks, failure cases and independent review before committing to delivery.

How do OpenClaw and ChatGPT Work fit?

The repository has an admitted OpenClaw Responses adapter. Its worker owns the configured browser, desktop and native Codex capabilities. ChatGPT Work is an operator-led execution route with its own tool access and permissions. The repository does not expose a remote Work execution API. Read the supported integration paths →

What happens after I download?

Verify sources and rights, assign independent review, commission the runtime, then inspect and admit the exact task digest. The worker produces candidate evidence. Review the actual content, obtain buyer acceptance, and follow the configured settlement process only when authorized. A download grants none of these approvals.

What is ready for production?

The static website, local demonstrations and integration code can be built and checked. A live deployment still needs authentic provider commissioning, release trust, independent security review, current dependency assessment and target-network evidence. Inspect the current readiness index →

Inspect delivered work →Explore the preserved demo collection →Complete delivery workflow →