Pablo Zavala · AI Safety Evaluation · Research Engineering
AEGIS and the Market for Verified Agent Work
AEGIS frames cross-owner agent work as a contract, verifier commitment, bid record, receipt, dispute path, settlement authority packet, and execution receipt, with launch claims bound to evidence.
Agent Work Needs a Market Boundary
Agent commerce fails when institutions treat unverified action as completed work. In the bad version, a buyer posts a task, a seller-agent returns an artifact, a score says pass, and money moves before anyone can inspect the obligation. Without a contract, verifier commitment, receipt, and dispute path, opacity gains speed precisely where accountability should harden.
That failure gives AEGIS its boundary: settlement after work crosses ownership lines. Private agent frameworks can plan, call tools, and route tasks inside one organization. Cross-owner work demands more: a neutral record of the promise, the verifier commitment, the artifact, the result, the challenge window, and the settlement reason. AEGIS supplies that verification and settlement boundary for independently owned agents.
Contracts Anchor Trust
AEGIS centers the `VerifiedWorkContract`. The contract turns a task into a scoped PRD, capability requirement, verifier preview, sealed verifier-bundle hash, deadline set, economic term, risk tier, settlement profile, and dispute policy.
The split protects the exchange. The seller sees enough to judge feasibility; the exact verifier remains sealed; both sides can later prove which bundle governed the bid. As a result, the task becomes an obligation with a testable shape rather than a chat instruction with shifting criteria.
From that commitment forward, the lifecycle must remain legible end to end. The contract freezes scope and verifier commitment; the verifier tests delivery against the same sealed bundle; the receipt records result, hashes, duration, exit code, and challenge status; the dispute path replays or challenges committed evidence inside a defined window; settlement then releases, refunds, or blocks through a settlement authority packet and a settlement execution receipt.
After delivery, the exchange preserves a `VerificationReceipt`: verifier kind, verifier version, bundle hash, deliverable hash, pass/fail result, duration, exit code, challenge status, and artifact references. Settlement then produces two artifacts: an authority packet that explains the permitted resolution and an execution receipt that records status, execution mode, chain reference when applicable, and final resolution. Together, those records turn completed tasks into institutional memory.
Standards Still Need Settlement
Against that standards backdrop, AEGIS composes with open agent infrastructure. MCP standardizes model access to tools and context, with an authorization profile for protected resources. A2A gives independent agents a protocol for discovery and task exchange. x402 supports machine-payable HTTP access. ERC-8004 proposes identity, reputation, and validation hooks for agents across organizations.
Together, those layers help agents discover one another, call tools, identify counterparts, and pay for resources. AEGIS adds lifecycle binding: promised work, committed verifier, delivered artifact, result receipt, dispute record, and settlement reason travel through one market boundary.
- What was promised?
- Which verifier was committed?
- What artifact was delivered?
- What failed or passed?
- Why did settlement follow?
Those questions define the boundary.
Public Scope Stays Narrow
The public AEGIS scope covers onboarding, discovery, bidding, delivery, deterministic verification, disputes, receipts, and settlement artifacts. Beyond that public claim sit internal society routing, operator control-plane work, sovereign paths, managed mandates, vaults, capital surfaces, and portfolio tooling.
Credibility starts with that boundary. A market for agent work should first prove that bounded, externally verifiable tasks can move through a disciplined exchange while preserving the contract, verifier, receipt, dispute record, settlement authority packet, and execution receipt.
Evidence Boundary
Evidence cutoff: July 4, 2026. The latest non-generated local source material I found is dated July 4, 2026 at 18:40:33 -04:00. The latest commit-backed AEGIS material I found is commit `36f4f3a9d4058178c11b1bbd3a2cebf2bd6d78e0`, dated July 2, 2026 at 11:07:10 -04:00.
As of July 6, 2026, local API, discovery, and front-door metadata contract checks passed, while `launch_clear` remained `false` and `docker-verify-v1` remained blocked. The local `./aegis check --json` result returned `ok:false`, `launch_clear:false`, and `runtime_parity_clear:false`. The release section failed `changed_files_classified` because `docs/AI_AGENT_MARKET_READINESS_STRATEGY.md` was unclassified; skills and install also failed.
The evidence boundary disciplines the claim. This evidence supports contract-surface readiness for scoped work, deterministic verification, portable receipts, disputes, settlement authority packets, and execution receipts; live public production launch remains outside the claim.
AEGIS Is for Bounded Work
A good first market handles work with a bounded artifact, a deterministic verifier, a clear failure mode, a dispute window, and economics that let honest sellers bid.
For a first market, that scope is enough. Because the market records each promise and each proof, trust gains operational meaning. Buyers and sellers can see what was promised, what was tested, what failed, what passed, and why settlement followed.