Human–Agent Responsibility Matrix

A conventional RACI does not fully describe an agentic workflow. It usually says who is responsible but not whether an agent observes, prepares, recommends or executes.

Suggested responsibility codes

Human roles

• O — Outcome owner: accountable for the business or transformation result.
• D — Decision authority: makes the material decision.
• E — Executor: performs the approved action.
• V — Verifier: independently checks evidence or outcome.
• C — Contributor: provides expertise or context.

Agent roles

• OB — Observe: retrieve, classify or monitor.
• AN — Analyze: compare, calculate or detect.
• PR — Prepare: create a draft, transaction or action package.
• RC — Recommend: propose an option with evidence.
• EX — Execute: perform a bounded, approved action.
• MO — Monitor: observe outcome and exceptions.

Build the matrix

List meaningful activities and decisions—not entire workstreams—in rows. Add the relevant client roles, consulting roles, agents and governance authorities as columns.

For each row, verify that:

• one human outcome owner is identifiable.
• material decisions have a human authority.
• agent authority matches its approved autonomy.
• preparation and approval are separated where required.
• execution is independently verified when impact warrants it.
• exceptions and unavailable humans have a defined route.

Example

For a proposed deviation from SAP standard:

• Process / Fit-to-Standard Agent: AN / RC
• Clean Core Guardian: AN / RC
• Process Consultant: C
• Solution Architect: C / V
• Business Process Owner: D / O
• Design Authority: D where architecture thresholds apply

The matrix exposes the real shift: agents take on more preparation and analysis, while experienced people spend more time on context, challenge and consequential decisions.