Human + Agent Operating Model

Introduction · how decisions become actions
Operating model Human in control

Human + Agent Operating Model

This page answers the practical question bp leaders will ask: what happens after a recommendation? It shows how decisions become actions, who stays accountable, where approvals live, how enterprise systems are updated, and how the audit trail is created.

How bp Sphere Thinks Platform Architecture

Discover → Understand → Recommend → Approve → Execute → Monitor → Learn

1
Discover
Signals, exceptions, and opportunities are detected across systems and events.
2
Understand
bp Sphere assembles context, evidence, policy, and business impact.
3
Recommend
Agents generate a governed recommendation with rationale and confidence.
4
Approve
Human owners approve, reject, modify, or escalate according to authority.
5
Execute
Approved actions flow to execution systems through governed gateways.
6
Monitor
Health, outcomes, exceptions, and policy adherence are continuously tracked.
7
Learn
Replay, overrides, and outcomes become memory that improves future decisions.

Who does what?

Analyst

Accepts, rejects, modifies, or requests more evidence on the recommendation.

Team lead

Challenges assumptions, requests reruns, and governs exception posture.

Controller

Owns material judgment, close integrity, control evidence, policy exceptions, audit readiness, and final accounting accountability.

Director

Overrides or approves decisions that cross policy or financial thresholds.

CFO / executive owner

Approves the final posture where enterprise authority or financial materiality requires it.

What happens after a recommendation?

Approval storage

Approvals, overrides, and escalations are captured in the decision ledger and tied to the acting human.

Execution update

Only approved actions pass through execution gateways to ERP, workflow, or operational systems.

Audit trail

Replay runtime, evidence contracts, policy lineage, and human actions make the decision reconstructable later.

The operating model is not “AI acts alone.” It is “AI recommends inside a governed human operating system.”