“Did anyone follow it?”
The wiki says a manager approves. Nothing shows whether one did.
An AI agent SOP, or agent operating procedure, is a standard operating procedure written as workflow steps the agent executes one at a time, with gates where an approval is required. ConvOps turns the SOP your team keeps in a wiki into a process the engine holds, so every run follows it and edits reach the next run.
claude · codex · cursor · any ai client · one step at a time · approvals recorded at the gate
Vendor onboarding · example
A standard operating procedure written as steps an AI agent runs one at a time. In ConvOps the engine holds the steps, the gates and the record.
ConvOps is the operations layer for AI agents. It holds your team's process as workflows with gates, a shared memory, scheduled and unattended runs, programs, a plan board, metrics and a full audit trail. It does not sell or host models.
key facts · october 2026
best for
not for
updated
The procedure lives in a wiki. People skim it, agents never see it, and the steps drift.
“Did anyone follow it?”
The wiki says a manager approves. Nothing shows whether one did.
“Which version is current?”
Three copies of the procedure, edited by three people, in three places.
“Does the agent know?”
Paste the SOP into a prompt and the model decides which parts matter today.
An example procedure, turned into a workflow from Claude Code, then run on a real case.
You hand the agent the wiki page. It proposes a workflow: one step per action, a gate where an approval is required, a step that must record its reasoning.
The same procedure, now held by the engine. The agent sees one step at a time, so it cannot jump ahead to the part it likes.
The agent runs the checks the SOP names and writes the findings down. The step closes only with a reason recorded.
Checks
The run stops at the approval gate. The manager reads the findings and decides. Until it is approved, the vendor is not created.
Approve this vendor?
Checks passed · contract over threshold · reasoning recorded
Every step lands in the history with its reason. Field changes on the task keep the agent label and the signed-in person apart.
agent label · implementer | person · verified from the credential
Each step is served when the run reaches it. The approval sits where the SOP requires a sign-off.
Legal name, tax id, bank details and contact, as the SOP lists them.
Against the policy the SOP names. Findings recorded on the task.
The step cannot close without the reason written down.
decisionAn approval gate holds the run until the vendor is approved.
needs approvalOnly after approval. The step history shows every stage.
Labelled examples from different kinds of procedure.
A telecom's refund SOP checks the amount first. Refunds under the threshold go on to processing; larger ones branch to a team lead's approval gate.
Decision step and approval gate
A manufacturer's document control SOP requires a review report for each controlled document. The step waits until the report is filed, and the quality manager approves the new revision.
Required report type
Each maintenance visit is its own task running the visit SOP: safety check, inspection, parts used, sign-off. The technician's notes land on the task, step by step.
One task per visit
The offboarding SOP keeps the access-removal step open until the agent records which systems were closed and why any were kept. HR approves the final checklist.
Recorded reason and approval
The runbook opens one child task per affected service, each with its own recovery steps. The incident stays open until every child is completed or cancelled.
Child tasks
Each claim runs the expense SOP: receipts checked, policy limits applied, exceptions written up. Claims over policy wait at the approval gate for the budget holder.
Policy check and approval gate
Every incoming sample runs the intake SOP: label check, storage condition, chain-of-custody entry. There is no approval gate: the record on the task is what the auditor reads.
Ordered steps, no approval
Details collected, checks run against the policy the SOP names, the reasoning recorded, then a manager approves before the vendor is created. The same steps for every buyer.
Recorded reason and approval gate
The reasons behind each step, kept by the team and recalled when the next case arrives.
Why the threshold is where it is, recorded at the step that applied it.
why we do it
The check that caught a bad bank detail last month, so the next run looks for it.
what went wrong
The policy, the vendor sheet and the contract folder. Found once, kept.
where it lives
Policies attached to a workflow reach the agent with the step they apply to.
guidance, not a gate
Many SOPs run on a calendar. ConvOps starts them, groups them and measures them.
Put a recurring SOP on a schedule: a weekly safety walk, a monthly control check, a quarterly policy review. Each schedule has its own time zone, and you can preview the next five runs. Group the schedules into a program, and its verdict shows when an SOP is late, silent or failing.
Place the one-off runs on Plan, where capacity comes from four weeks of completed work and spread shows a preview before anything moves.
Define the numbers the SOP must report in Metrics, such as rework per case, and bind them at the step that produces them. An agent can propose a metric too, and it stays a draft until approved. Markers on the trend show every edit to the SOP workflow.
The engine serves the current step only. Steps inside one workflow never run at the same time.
An approval gate refuses the advance until an explicit approval is given, and the response lists what is missing.
require_context keeps a step open until the reason is recorded, so the decision travels with the work.
Step history with reasons, an activity feed, and field-level before and after on tasks and workflow instances.
The same procedure, kept four ways.
| ConvOps workflow | A markdown SOP file | A prompt or skill | A wiki page | |
|---|---|---|---|---|
| Who decides the next step | The engine, after checking the gates | The model, after reading the file | The model, each time | Whoever reads it |
| Approval gate | A gate holds the step until it is approved | A line asking for one | An instruction to ask | A note on the page |
| When the SOP changes | Edit one step; the next run follows | Edit every copy of the file | Re-share the prompt | Edit the page and hope it is read |
| Record | Step history with reasons and approvals | None | The chat history | Page history |
| Whether it works | Metrics on every run, with change markers | Unknown | Unknown | Unknown |
Connect the AI tool your team already uses, install the workflows, and run one real piece of work through them.
Pick one SOP your team repeats weekly and paste the wiki page into your AI client
Ask the agent to draft it as a workflow: one step per action, a gate where a sign-off is required
Run one real case through it and read the step history afterwards
Fix the step that was unclear once; every later run follows the new version
In ConvOps, an SOP for AI agents is a standard operating procedure stored as a workflow: ordered steps with instructions for each, and gates where an approval or a report must come first. The agent receives one step at a time, and the engine checks the gates before the run moves on. Edit a step once and every later run follows it.
Write one action per step, in the order it happens, and say what done looks like for each. In ConvOps, mark the hard rules with MUST and the preferences with SHOULD, so the agent knows which ones it may weigh. Put a gate where a sign-off is required and name where a rejection goes. Then run one real case and fix the unclear step.
A markdown SOP is read by the model, which then decides which parts to follow. A ConvOps workflow is held by the engine: the agent receives one step at a time, the gates are checked before the run moves on, and every step leaves a record. The file describes the procedure; the workflow runs it and shows afterwards what happened.
Yes. ConvOps serves each step only when the run reaches it, so there is no later step to jump to. A gate holds the step until an explicit approval is given, recorded with the account that gave it, and a required report in enforce mode holds it until the report is filed. A refused advance returns what is missing, so the agent knows what comes next.
Measure it. In ConvOps you define the metrics the SOP must report, such as rework per case, and bind them at the step that produces them, and each value carries who recorded it and at which step. ConvOps also records the cost, duration, tokens and outcome of every run by itself, and agents cannot write those. Markers on the trend show each SOP edit, so you can tell whether a change helped.
Yes. In ConvOps, a schedule starts a task with the SOP attached at the cadence you set, such as weekly or monthly, in its own time zone. It can spawn a new task on each fire or wake an existing one. Group the schedules into a program and its health verdict shows when a run is late, silent or failing.
In ConvOps, it depends on the step's approval gate. If the step names an on_rejected target, a rejection moves the run to that step, for example back to the checks, and the step history records it. If the step has no target, the rejection is refused and the run stays at the approval, so name a target for every gate that can say no.
Not inside one workflow. In ConvOps, the steps of a workflow run in order, one at a time. Independent pieces of work are split into child tasks that start together, each on its own workflow, and the parent waits until all of them finish. That keeps each piece on its own record, with its own gates and its own history.