The server checks each stage before a task moves on
AI agents write code faster than people can review it. CodeHerder runs each check on the server. An agent cannot skip a check.
Generation stopped being the bottleneck
Four sources point the same way. Faros, LinearB and Anthropic published three of them. One is an independent academic paper. More AI-generated code means more review work for people.
median PR review time increase on teams adopting AI coding agents
Faros AI, "The Acceleration Whiplash" (22,000+ developers, 4,000+ teams) · 2026Vendor-published
4.6×longer first-review wait for an AI-generated PR
LinearB 2026 benchmarks (8.1M PRs, 4,800 organizations), via secondary summary · 2026Vendor-published, secondary source
53.9%of AI-agent refactorings sit in commits that do not say they refactor
"Agentic Refactoring: An Empirical Study of AI Coding Agents" (arXiv) · Nov 2025Independent
0 to 20%of tasks developers can fully hand off without review, despite using AI for roughly 60% of coding work
Anthropic, 2026 Agentic Coding Trends Report · Jan 2026Vendor-published
PyTorch's own maintainers went further in May 2026: unreviewed AI-generated code is not accepted into pytorch/pytorch main at all. Read their playbook →
Each task type has stages, and each stage has a gate
Every task type carries a schema: an ordered pipeline of stages, and a gate on each one. A story ships through a real pipeline, for example:
Planning cannot end without written acceptance criteria. The build stage cannot end without a link to the merge request or pull request. If the workspace has a git host connection, the host must also confirm that the request exists. A task with no repo skips this check. The build stage also lists test and lint runs and a typecheck. These checks block only when a workflow marks them required.
The verify stage has a required goal gate. When it fails, the server sends the task back to code. The next code run asks for an Opus-tier agent. After 3 send-backs, the task moves to Blocked and waits for a person. The server applies the gates the same way to a person's move and to an automatic one.
Six gates the server checks
Each gate below maps to code in the server. Some gates let a task through when the server cannot read the workflow. Your git host approval is the last check.
Required fields
A stage declares which fields must hold a value before the task can leave it. If a field is empty, the server refuses the move. Planning needs acceptance criteria. The build stage needs a link to the merge request or pull request.
Required verifications
A workflow can mark a test, lint, typecheck, or goal-gate check as required for a stage. A required check blocks the advance until a pass exists for the current attempt. The built-in build stage lists its checks but does not require them. A recorded fail on a required goal gate sends the task back to code.
Checklist gates
On a story or feature, acceptance criteria live in a checklist field. Each item names the stage that proves it. An unticked item blocks any move past that stage. An agent cannot claim done while its list has open items.
Approvals with separation of duties
A stage can require sign-off from a named member, or from any human or agent, before it advances. The approver is never the requester or the task's own session. This rule is always on. The gate holds an explicit status change the same way as an automatic advance. The server tells each caller if they may approve, so nobody has to guess.
Dependency and close gates
A task cannot skip past a dependency that has not finished. A parent with open children cannot reach done. A parent can also opt in to wait until every child succeeds.
Change-size limit
A stage can opt in to a limit on the size of a change. The limit can count changed files or lines. It can also count top-level directories. When a change goes past it, the server refuses the advance and names the limit. Split the work into smaller tasks, or raise the limit. A task with no recorded size is not blocked.
Which check can the author not pass?
The answer is one human approval on your git host. Set it up once. The agent that wrote the change cannot give it.
No built-in workflow has an approval gate. A custom workflow can add one. Custom gates need the Starter plan or higher. Put a human approval gate on the entry stage. On an agent stage, the agent can never pass it, and the server warns you. Without a gate, the human check is your git host. Protect the main branch and require one approval. When the host needs a human approval, the merge agent's instructions tell it to block the task. A person then approves on the host.
An agent review stage helps, but it is not separation of duties. The server only checks that the reviewer uses a different model tier. It does refuse one self-check. A required goal-gate pass from the author fails with self_verification.
Read how approvals work and the security page.
The review stage reads the whole task
Some tools review a diff alone: CodeRabbit, Greptile, Qodo, Graphite and Sourcery. CodeHerder does not integrate any of them. The review stage reads the whole task.
The review brief holds the task description and acceptance criteria. It holds the hand-off notes from the earlier stages and any human answer to a mid-task question. It also holds a short index of wiki pages and the size of the change. The reviewer can then check the change against the plan.
Want to see the gates from inside the app? Read how work flows, how approvals work, and how review actually happens, or see how this fits alongside device isolation and access control on the security page. Gates catch most mistakes, not all of them. see what happens when an agent gets it wrong.
Try it on one repository for two weeks.
File a task. An agent builds it. A person reads every request. You see the cost of each task.