What does Epistema do?
Epistema helps an organization determine whether one piece of AI-assisted work has enough evidence, authority, and review to become organizational action. It shows what is supported, what is missing, and whether the workflow should proceed, be constrained, or receive further review.
Is it an AI monitoring tool?
Not primarily. Monitoring tools show prompts, responses, traces, costs, latency, scores, and errors. Epistema can consume those signals, but asks a different question: what does the evidence mean for reasoning, commitment, outcome, and calibration?
Epistema is not a general log analytics or search system. It asks for the smallest evidence set needed to evaluate one AI-assisted workflow.
What is a bounded workflow?
A specific piece of work with a clear context and decision question—for example, a customer-policy answer, an AI-assisted business requirement, a vendor recommendation, or an incident diagnosis. It should be small enough to inspect and important enough to govern.
What evidence is needed?
Epistema starts with the minimum evidence needed to evaluate one workflow: what happened, what the work was for, what commitment may have formed, and what happened afterward.
- activity evidence: traces, spans, chat logs, tool calls, or agent records
- context evidence: work items, acceptance criteria, policies, or business objective
- commitment evidence: approvals, delivery records, merged code, published reports, or customer-facing answers
- outcome evidence: corrections, rework, costs, incidents, or follow-up decisions
Not every source is mandatory. A small, clean bundle is better than a confusing data dump. Missing evidence should remain visible.
Does Epistema prevent bad decisions?
No system can promise that. Epistema makes the commitment boundary and evidence gaps explicit before weakly supported output is treated as something the organization approved.
Software engineering learned decades ago that critical work should not go directly into production. AI-assisted work deserves the same discipline before it becomes organizational action.
How would it help with an Air Canada-style chatbot failure?
Policy-sensitive question → authoritative policy evidence required → validation and authority checked → allow, constrain, escalate, or withhold
Epistema would not replace customer-service policy or decide the legal outcome. It may expose the missing evidence or authority before an answer creates customer reliance.
What is the Workflow Value Ledger?
A reusable record of why the work was done, what supported it, what human contribution occurred, what commitment was made, what risks or constraints remained, and what happened afterward.
What does a starter intake produce?
- a bounded workflow description
- an evidence inventory and normalized bundle
- a Workflow Value Ledger readout
- visible gaps, constraints, and withheld claims
- one recommended next action