Backlog, Cloud Agents, Review, Then Ship
Short answer
A software delivery line can run in this order: a backlog item, a cloud agent that drafts the change, a person who reviews the pull request, then ship. Cloud agents are the hosted workers on that line. They are not the line. This page describes the stations.
Last updated: September 22, 2026.
Key takeaways
| Question | Answer |
|---|---|
| What is the line? | Backlog, then a cloud agent draft, then human review, then ship. |
| What is a cloud agent here? | A hosted worker that can draft from a written ticket. Not the whole factory. |
| What does the person do? | Review the pull request and decide the merge. |
| What does this page refuse to claim? | A cycle time, a throughput ratio, or a headcount result. |
| Where is the cloud agent definition? | On the cloud agents page. This page is only the delivery order. |
The four stations
A backlog is a list of jobs, not a factory. It becomes part of a delivery line when each job has a next station that is the same every time. One such order is the one in the URL: backlog, cloud agents, human review, ship.
| Station | What happens | What it does not mean |
|---|---|---|
| Backlog | Someone writes a job with an edge: a bug, a failing check, or an acceptance note | A pile of ideas with no definition of done |
| Cloud agent | A hosted agent reads that job and proposes a diff | The agent has accepted the change |
| Human review | A person reads the diff and the check, and can send it back | A rubber stamp, or a claim that review is free |
| Ship | The person merges, under whatever release rule you already use | The agent deploying because a pull request exists |
The line itself is a software factory shape. The coding stations, with or without a hosted runtime, are on the code factory page.
What the cloud agent is for on this line
The agent's job is the middle station. It should be able to read the ticket, stay inside a boundary you wrote, run the check you named, and open a pull request that shows the diff and the check result.
It should not be the author of the ticket and the approver of the ticket. Those are different stations on purpose. Agentic AI is the behavior that lets an agent pursue the drafting goal with tools. The pipeline is the order of stations around that behavior.
If the ticket is a product debate, it is not ready for the middle station. Send it back to a person. Cloud does not make an unbounded job bounded. Dark factory software in production lists the pieces a line needs before you rely on it: job, check, gate, record.
What review is for
Review is the gate, not a courtesy. The person looks for a wrong approach, a missed edge, and a diff that passed a check but changed the wrong thing. They can request a smaller job.
This is also why the page does not claim that review removes a bottleneck or multiplies output. Review is often the scarce station. A line that drafts faster than it can be read just moves the queue. Seeing that is useful. Pretending it is not a queue is not.
The record is the pull request: the diff, the check, and the person who merged. If those are not findable, the line did not happen in a way you can explain later.
xSquad runs a managed version of this order, with human review before ship. The offer is on xsquads.ai. How several agents split the middle station is how we run agent swarms.
How to lay out backlog, cloud agents, review, and ship
Keep the order boring. If a ticket skips a station, write down why, or it will become the normal path.
Write the backlog item as a job with an edge
Say what is wrong or what must be true, and how a stranger would notice a failure. An untitled idea stays in the list. It does not enter the agent station.
Send only bounded jobs to the cloud agent
Give the agent the repo, the check, and a fence around what may change. If you cannot fence it, a person still owns the draft.
Require the check result on the pull request
The reviewer should see pass or fail without asking the agent whether it feels done. A pull request with only a summary is not the handoff.
Leave merge with the person
The agent opens the request. The person accepts it, sends it back, or splits it. Ship follows your existing release rule, not the existence of the request.
Stop the line when review cannot keep up
If drafts wait unread, pause the agent station or narrow the jobs. A growing unread queue is a signal. It is not a success metric, and this page will not turn it into one.
Frequently asked questions
What is the backlog, cloud agents, review, ship line?
It is a delivery order. A written backlog item goes to a cloud agent for a draft, then to a person for review, and only then to ship. The cloud agent is one station.
Are cloud agents the same as a software factory?
No. Cloud agents are hosted workers. A software factory is the production system, including intake, check, review, and ship. This line is one way to arrange that system.
Does the agent ship the change?
No. The agent opens a pull request with a diff and a check result. A person reviews and merges. Shipping still follows the release rule you already have.
Will this clear a backlog?
Not as a promise. Drafts can move only as fast as jobs are bounded and review can read them. If review is the scarce station, the queue moves there. This page states no cycle time.
How is a cloud agent different from a local coding agent?
A cloud agent runs on hosted infrastructure, so the draft is not tied to one open laptop. The job shape is the same: a bounded change, a check, and a human merge. The definition is on the cloud agents page.
What should not enter the agent station?
Product debates, unbounded rewrites, and anything you cannot check. Those stay with a person until the job has an edge and a fence.
Is human review optional once tests pass?
No. A passing check means the diff is worth reading. It does not mean the change is the right one. The person can still reject it.
Does this replace engineers?
It moves routine drafting to an agent when the job is bounded. Engineers still write jobs, review diffs, and own what merges. No headcount result is claimed.
How is this different from agentic AI?
Agentic AI is the ability to pursue a goal with tools and a check. This page is an order of delivery stations that can use that ability in the middle. You can have agentic tools and still have no line.
Where do agent swarms fit?
A swarm is one way to staff the middle station with several roles and still keep one human gate. It does not add stations, and it does not remove review.
Sources
- What are cloud agents? for the runtime definition this line depends on.
- What is a software factory? for the production system around the four stations.
- What is a code factory? for the coding line with or without a hosted runtime.
- What is agentic AI? for the behavior inside the draft station.
- xsquads.ai and how we run agent swarms for a staffed review gate.
No cycle time, throughput, or headcount figures are claimed here.
If you want this order with a person on review, start at xsquads.ai.
running a shopify store? → /shopify
Last updated: September 22, 2026. First published: June 14, 2026.
Ready to Scale Your Development Team?
Try a squad on your repo. You review every PR.