Introducing the Dispatch Inbox: the work is already drafted
Until this week, every change Dispatch governed started the same way: a person noticed something and wrote a request. The loop that followed was already the safest way to let an agent touch a production website, with a preview before anything shipped, a role-gated approval, attribution, and one-click restore. But the loop waited. Someone had to look at the site, decide a page had slipped, and type the intent. Today that changes. Dispatch now finds the work, drafts the fix, and puts it in front of the right person. You keep the approve moment. All of it is in production now; see the Inbox page for the walkthrough and the changelog for the dated record.
Why the loop had to stop waiting
A website run by agents changes faster than any one person's familiarity with it. Nobody remembers what the pricing page's structured data looked like three deploys ago, and nobody should have to. Meanwhile the checks that decide whether a page can be cited by an answer engine are mechanical: a title in range, a meta description, a single H1, a canonical, JSON-LD, social tags, indexability, a 200 status. A machine can measure them every day. The gap was never detection; it was that detection produced a dashboard rather than a request, and a dashboard is work for the reader.
The economic frame matters here. You already own the coding agent, the repository, the host, and the framework. What Dispatch adds is manageability, and the return is production capacity without a developer queue: the twenty comparison pages, the sitewide structured-data pass, the metadata cleanup that never makes it to the top of a sprint. The proactive layer is that capacity showing up on its own.
What shipped
Scheduled crawls and score history
Every governed site now has a crawl cadence, daily or weekly, set in its Policies card. Each crawl scores every page and writes the score to that page's history, so a regression is a fact with a date on it. The page detail view shows the check list, the sparkline, and everything Dispatch has found on that page.
Findings with a drafted fix
A finding opens when a check that used to pass now fails, when a page loses ten points or more since the last crawl, when a page answers with anything other than a 200, or when a page falls out of the sitemap. Every finding carries a suggested change request written in plain language: which route, what is wrong, what to change, and what not to touch. It is deterministic, not generated, so the same failure always produces the same request. Findings close on their own when the check passes again.
The Inbox
The Inbox is the daily-return surface. It has four sections: requests awaiting your approval, findings with a drafted fix ready, pages that slipped since the last crawl, and your work in progress. Draft fix files the request with the finding's intent, page, and suggested risk tier already filled in. Dismiss is an approver decision. Approvers land on the Inbox after login; operators land on requests; owners land on the overview with an Inbox strip showing the counts.
Autonomy levels
Each site can now auto-approve low-risk requests. When the policy is on, a low-risk request that reaches preview-ready is approved by the policy, and the record says so: the approval names the policy as the actor and the activity feed logs it. High-risk requests still need two distinct approvers, and the person who filed a request can never be one of them. That rule lives in the database, not in the interface. Everything a policy approves is one click from restore.
Alerts that carry the draft
Slack and email notifications now include what was asked, which page, how risky, and the one thing to do next, not just a link. An approver can decide from the channel, then open the preview or the request in one tap.
Note to request
Every page detail view has a box for what should change. Write the note in your own words; Dispatch drafts the request with the page route, current title, and failing checks attached, so the agent works from the page as it was last crawled.
The agency handoff playbook
For agencies running client sites on Dispatch, the runbook we use ourselves is now public: onboard, configure, hand off, plus the weekly rhythm the client runs from the Inbox after the engagement ends.
What did not change
Dispatch still governs the work and never executes it. Your coding agent writes the code, your repository holds it, your host builds it. The proactive layer produces requests, and a request is only ever a request until a person or a policy you set approves it on a real preview. Nothing ships silently, nothing is un-restorable, and the record is complete whether the initiator was a colleague or a crawl.
The one-line version
The agent builds it. Your team approves it. Dispatch keeps it safe. And now Dispatch also finds it, so the approve moment is the only moment you have to show up for.
Connect a website on the free plan and the first crawl runs on its own. The Inbox will be waiting.
What is the Dispatch Inbox?
The Inbox is the page in Dispatch where work waits for you: change requests with a preview ready for your approval, findings from the scheduled crawl of every governed page (each with a drafted fix), pages whose score slipped since the last crawl, and the work you have in flight. Approvers land on it after login.
Does Dispatch find problems on its own?
Yes. Dispatch re-crawls every governed page on a schedule you set per site, keeps a score history for each page, and opens a finding when a check that used to pass now fails, a score drops, a page stops answering, or a page falls out of the sitemap. Each finding carries a drafted change request an agent can execute.
Can Dispatch approve changes automatically?
Only when you turn it on, and only for low-risk requests. A per-site policy lets low-risk requests go live without waiting for a reviewer; the approval names the policy as the actor, the activity feed logs it, and restore is one click. High-risk requests still need two distinct approvers, enforced in the database.
Is the proactive layer available now?
Yes. Scheduled crawls, score history, findings with a drafted fix, the Inbox, autonomy levels, role-aware landing, alerts that carry the draft, and note-to-request are all in production today, on the free plan for one governed website.
Dispatch Team
Writing about AI-powered websites, AI governance, and collaboration, helping teams turn AI from scattered experiments into shared organizational capability.