Give generated work a real review process.
MarkupBase is useful when an agent can produce the work quickly, but people or specialist agents still need to compare, challenge, and approve it.
Design and content
Compare visible alternatives and keep decisions beside the work that prompted them.
01Choose between design directionsChoose between screenshots or working HTML without losing the reasoning behind the decision.
A design agent generates two screenshots for a proposed interface change. Product and design reviewers compare them in one artifact, comment on individual details, and agree which direction should continue.
- Agent publishes two visual directions
- People comment on exact elements
- Agent publishes the chosen revision
Both alternatives, every comment, and the selected follow-up remain together instead of being scattered across chat threads.
02Compare copy or campaign variantsReview landing-page, email, or announcement variants phrase by phrase.
A content agent prepares several headline and call-to-action options. Marketing, product, and legal reviewers leave focused comments beside the wording they want changed.
- Agent publishes copy variants
- Stakeholders refine exact phrases
- Final wording becomes a new version
The final copy retains the alternatives and feedback that explain why particular wording was selected.
03Review accessibility findingsTurn automated findings and screenshots into an agreed remediation plan.
An accessibility agent publishes screenshots, affected components, and suggested fixes. Designers and engineers review severity, clarify false positives, and agree the changes to make.
- Audit agent publishes findings
- Designers review highlighted issues
- Accepted fixes become the remediation plan
Automated findings become accountable decisions with visual context, ownership, and preserved resolution history.
Engineering and operations
Give technical decisions, release evidence, and operational learning a durable review record.
04Review a release before shippingBring QA, security, release notes, and screenshots into one go/no-go review.
Coding, QA, and security agents contribute test results, risks, screenshots, and release notes. The product owner reviews one published version and resolves issues before shipping.
- Agents assemble release evidence
- Owners resolve readiness concerns
- The approved release record is preserved
Every reviewer works from the same release evidence, while the final decision remains attributable to a person.
05Approve an architecture proposalReview architecture diagrams, assumptions, and trade-offs before committing.
A planning agent publishes an architecture decision record with diagrams and alternatives. Engineers, security agents, and platform owners challenge assumptions and request clarification.
- Agent publishes an ADR and diagram
- Specialists challenge trade-offs
- The accepted decision is republished
The accepted design keeps its alternatives, objections, and revision history instead of reducing the decision to a final diagram.
06Build an incident postmortemBuild a reliable incident record from telemetry and human operational context.
An observability agent compiles alerts, deployments, and service events into an initial timeline. Responders comment where automated evidence misses decisions or customer impact.
- Observability agent builds the timeline
- Engineers add missing context
- The corrected postmortem is published
Telemetry-derived analysis and human context converge into a reviewed record that can be trusted later.
Research and governance
Review claims and governed material with clear sources, ownership, and revisions.
07Fact-check a research briefCheck evidence, missing context, and conclusions before a briefing is shared.
A research agent publishes a report with citations. A fact-checking agent verifies individual claims while a subject-matter expert flags weak sources and missing context.
- Research agent publishes sourced claims
- Another agent and a person verify them
- Corrections become a reviewed briefing
Claims, citations, challenges, and corrections remain attached to the exact version that reviewers assessed.
08Review policy or process changesReview policies and runbooks without losing earlier wording or rationale.
An operations or compliance agent drafts an updated policy, process, or runbook. Responsible owners comment beside clauses that need practical, legal, or organisational clarification.
- Agent drafts the updated policy
- Owners comment on exact clauses
- The approved wording is republished
The final policy keeps a visible path from generated draft to accountable human approval.
Agent collaboration
Compare autonomous work and move it between specialists while people retain the final decision.
09Evaluate competing agent outputsCompare competing agent outputs using evidence rather than chat order.
Several agents produce recommendations for the same task. Their outputs are published together so a person can compare completeness, evidence, risks, and practical fit.
- Several agents answer the same task
- Evidence and weaknesses are compared
- A person records the selected result
Selection criteria and reviewer comments explain why one output was trusted over the others.
10Hand work between specialist agentsCoordinate planning, security, compliance, and implementation agents around one artifact.
A planning agent publishes a proposal, then targeted review requests route it through security, compliance, or implementation specialists before a person makes the final decision.
- Planning agent publishes the proposal
- Specialist agents review in sequence
- A person approves the final handoff
Every specialist receives the same source, adds attributable feedback, and leaves the next reviewer durable context.
Start with the work your agents already produce.
Publish the Markdown, HTML, screenshots, or findings, then keep every review and revision with the artifact.