Agentic Engineering · WAI
Designed so the same error isn't solved by hand twice.
WAI is an observation layer that aims to classify application errors and match them against past cases. If a known fix exists, presenting a playbook is designed to happen automatically; if not, a human is asked.
- Status
- Prototype
- Last verified
- 2026-07-18
- Access
- TODO_CONTENT_CONFIRMATION
The problem it addresses
The same error gets investigated and fixed by hand, again and again, across different services at different times. WAI aims to accumulate incident records in a single layer and make the known fix reusable.
How it works
An event is sent to WAI via an SDK or a REST/OpenTelemetry ingest path.
The event is classified by its fingerprint and matched against past cases.
If a known playbook exists, it's proposed; if not, a human is asked.
The proposed fix is tried in an isolated environment; it doesn't reach production without human approval.
Verified capabilities
Event classification
Incoming events are classified by fingerprint and matched against past records.
Playbook proposal
A machine-readable playbook is proposed for a known problem; a new problem is routed to a human.
Evidence-bound diagnosis
Every diagnosis is designed to reference a real incident record; an unverified answer is treated as invalid.
Architecture and trust model
Each claim below is labeled with where it currently stands — none of these are confirmed, final guarantees.
Event ingest
- In development
Python and TypeScript SDKs, with a REST or OpenTelemetry ingest path, are designed for event capture.
- In development
Event processing is planned to run in parallel over NATS JetStream.
Doctor authority
- Design principle
A new "doctor" starts only in a watching "shadow" mode; the principle is that its authority grows in stages as it proves itself.
- Design principle
A fix headed for production not being applied without human approval is a design principle.
Known limits
Automated fixes aren't end-to-end verified
The flow for generating fix code, applying it in a sandbox, and testing it is still in development; no independently verified end-to-end demo has been published yet.
The doctor pool's maturity varies
The criteria for advancing between the shadow/advisory/acting stages will become clearer as the product matures.
Access and integration details aren't confirmed
The SDK distribution shape and access terms haven't been confirmed by the project owner yet (see the status block).
Evidence
A public set of tests, demos, or sample output for this product is currently being developed — it will be added here once published.