How does Moxo keep humans in control when AI agents run a workflow?
Quick Answer: Moxo keeps people on the decisions by design: approvals and other human steps are ones its product page says "only a person can close", and AI agents can fill preparer, advisor or reviewer slots around them (both read 15 September 2026). The checks on AI output are opt-in, though. In synthetic AutomationAtlas tests that day, an AI extract step's "Human review" and "Supervisor Agent" switches were both off by default, and the builder accepted the same role as a form's submitter and its approver.
Moxo says AI prepares and a person decides
Moxo states its model in one line on its AI page, read 15 September 2026: "AI prepares, humans decide, the record proves it." It's a good line, and the architecture mostly backs it up. Its Agent Foundry page gives every agent "a named human to hand work to the moment it steps outside it". The Description of Services, read the same day, places people as follows.
| Control | What the Description of Services says |
|---|---|
| Roles | "resolve to real participants at Flow start" |
| Assistive slots | Preparer, advisor and reviewer, "each fillable by a person or an AI agent" |
| Group decisions | "completion modes (any one, all, or majority)" |
| Deadlines | "due dates with configurable reminders and escalation routing" |
| External people | A Magic Link grants "account-free access to a specific assigned task" |
What the slogan leaves out is who carries the risk when a person approves something an AI prepared. Moxo's Terms and Conditions, read 14 September 2026, answer it plainly: "Company is solely responsible for any AI Content, including the accuracy, quality, appropriateness, and legality thereof", and Moxo AI is provided "AS IS". That is normal for the category; Zapier's Terms likewise make the customer "solely responsible" for Customer Content, which they define to include AI Inputs and Outputs. So the approve click is where the AI's work becomes your organisation's decision, with an employee's name on it. Design the step for that person, not for the demo.
The review gates ship switched off
In synthetic AutomationAtlas tests on 15 September 2026, a supplier-certificate template's AI extract step carried two review switches, and both were off by default: "Human review" ("Pause for a user to review and edit the AI output before it's used downstream") and "Supervisor Agent". Turning on Human review added a "Chat to change the draft before approving." switch and a required "Who reviews" picker, because "a flow can't be published with human review on but no one assigned."
That check is well designed, with a gap an internal auditor will spot: it confirms that somebody is named, not who. The picker let us choose the Requester role, the same role that submits the form, and the template later published with that role as its approver too. Keeping the person who prepares apart from the person who approves is something you design into the template, not something the builder enforces.
The Approval step's "On rejection" control listed "Revise and resubmit", "Run rejection steps" and "End the flow"; a fourth route, "Let the approver decide" ("The approver picks which steps to send back"), appeared only once "Revise and resubmit" was selected. The audit log labelled actors "AI Agent", "System", "Public link" or the person's name.
Other tools make the same default choice
Off-by-default review is a category habit, not a Moxo quirk. Zapier's AI by Zapier tools article says of "Require approval before running": "This toggle is off by default". Relay.app's documentation, read 14 September 2026 as that service was shutting down, said "By default, the option to require a human review is turned off for newly created steps." The reasoning is easy to see, since a gate that starts on slows every demo, but it hands the governance decision to whoever builds the flow, who is not necessarily the person answerable for the outcome.
Where the tools differ is who the human is. Zapier ties each review to a Zapier user: its Human in the Loop article, read 15 September 2026, says "Reviewers must have a Zapier account." and "Only one user can complete a review." Microsoft Copilot Studio can go the other way and make AI one of the approval stages; its preview documentation for multistage approvals in agent flows, read 14 September 2026, says: "By default, if the AI stage approves, the approval continues to the next stage. If the AI stage rejects, the approval ends as rejected." Moxo sits between them: a person on the human step, AI in the slots around it, and autonomous agent steps only where the organisation configures them.
Moxo lists Approvals on Team and review agents from Scale
Moxo's price list starts human approval and AI review agents on different plans. As of 15 September 2026, moxo.com/pricing lists Approvals and Magic links on Team at "$5,000/yr" with "100 flows/yr · $100 of AI/yr · Unlimited seats". Review Agent, Prepare Agent, Advise Agent and Audit log appear from Scale, a "Custom quote", and Supervisor Agents only under Enterprise. The human gate is the cheap part and the AI checker the expensive part, which is the right way round. The awkward bit is the audit log, the record that proves who decided, which also sits above the published price. Controls seen in testing are no guide to what Team includes.
Low-volume and unattended automation fall outside Moxo's published price
At Team's "$5,000/yr" (moxo.com/pricing, 15 September 2026), 20 flows a year cost $250 each (AutomationAtlas arithmetic). The page lists Triggers, Webhooks and REST APIs only from Scale, so app-to-app automation with no human decision point has no published Moxo price. If nobody needs to decide anything, you are paying for the one thing this product does best and not using it.
Editor's Note: The builder recently did one thing I liked and one I didn't. On a synthetic supplier-certificate template, we switched on Human review for an AI extract step, and Moxo made the reviewer picker mandatory, warning that "a flow can't be published with human review on but no one assigned." Exactly right. It then let us pick the Requester, the same role that submits the form, and the template published with that role as approver too. The tool insists that a human signs, but keeping the preparer and the approver apart is your policy to build. Build it that way and the product lives up to its own line: "AI prepares, humans decide, the record proves it." — Rafal Fila, ShadowGen
Related Questions
Related Tools
Moxo
Process orchestration for cross-company work: AI prepares the steps, people sign them off, outsiders join by link
Workflow AutomationZapier
Automate workflows between apps without coding—connect 9,000+ tools with simple, reliable automation.
Workflow AutomationActivepieces
No-code workflow automation with self-hosting and AI-powered features
Workflow AutomationAutomatisch
Open-source Zapier alternative
Workflow AutomationRelated Rankings
Best Human-in-the-Loop Automation Platforms 2026
Eight platforms are ranked here on a single mechanism: how a person enters an automated path, and what the system can prove afterward about the decision they made. Moxo scores 7.7 and ranks first; Camunda 8 scores 7.2, Microsoft Power Automate 7.1, Pega 6.9, n8n 6.7, Zapier 5.9, Kissflow 5.8 and ServiceNow 4.8, scored 22 September 2026 on the five weighted criteria. Those criteria cover the step type, the approver's account or licence, the routes available on a rejection, the audit record, and the meter that charges for the decision, and none of them can tell a review from a rubber stamp, so the methodology adds a check readers can run on any product in a few minutes: whether the decision record tells a considered approval from an instant one. Every figure was read from a vendor-owned surface and carries its own date.
Best Automation Platforms for AI Orchestration 2026
This ranking answers one question: how many real business applications can an AI agent act on out of the box? It evaluates nine platforms as of August 2026 on the reach they give an agent, not on the workflow logic they can express. That boundary is deliberate, because two neighbouring pages on this site answer different questions. Best Process Orchestration Platforms 2026 scores multi-step process control, error handling and state management. Best AI Agent Platforms 2026 scores building and hosting the agent itself. This page scores the layer between them: the connective tissue that lets an agent already built elsewhere reach the applications a business actually runs on. A platform that leads one of those pages can place low here, and two of them do. Scores derive from application and action catalogue counts, the exposure model each platform uses to publish those catalogues to an agent, setup effort, failure handling and cost per agent action. Every figure was retrieved from a vendor-owned surface on 11 August 2026 unless an earlier date is stated against it.
Dive Deeper
Moxo vs Lindy in 2026: Who Presses Approve When the Agent Asks
Lindy, founded in 2023, sells an AI teammate that lives in Slack and in each person's inbox; Moxo, whose predecessor Moxtra was co-founded in 2012, runs multi-party processes where approvals are steps only a person can close. Moxo wins three rounds to two: who closes the step, work across companies and the record. Lindy wins reach and testing. Vendor pages and docs read 28 September 2026.
Moxo vs Kissflow in 2026: What an Outside Approver Costs, and What the Record Proves
Moxo and Kissflow both sell approvals to people who do not write code, and each wins one round outright: Moxo's Web and Headless SDKs run the approval inside the buyer's own product, under the buyer's own sign-in. Kissflow's portal audit log stamps every event with an actor, a device and an IP, where Moxo publishes no field list. Both answers are quoted, not priced, and both meters punish a one-step request sent to hundreds of clients. Cost, the decision record, rejection routes, AI metering and governance, verified 14 to 23 September 2026.
Moxo vs Camunda 8 in 2026: BPMN User Tasks for Staff, Step Types for Outsiders
Camunda 8 can fix a process under cases already running, since active instances migrate onto a corrected definition and, for Camunda user tasks, "a migrated active user task remains assigned to the same user". Moxo keeps the shared queue that Camunda's Tasklist V2 stops evaluating, with desks that are "a shared inbox that multiple users can pick work from". Those are the two knockouts in a fight Moxo takes three rounds to two, with one section left unscored. Portability, AI review gates and a published price decide the rest, verified 14 to 23 September 2026.