MPUG Editorial · August 26, 2026 · ~4 min read · Category: PM Platforms
monday.com AI Workflows can now stop mid-run and wait for a person to say yes. The company’s What’s New page lists the change on August 4, 2026: a human-in-the-loop block that sends an approval request and branches the workflow based on the response. For project managers who have wanted automation without surrendering the sign-off, that is the missing piece.
What the block does
monday.com’s description is compact: “Add a human-in-the-loop block in AI Workflows to send approval requests and branch based on the response.” Three things are worth pulling out of that sentence.
It pauses execution. The workflow does not complete optimistically and wait for someone to object afterwards — it holds at the block.
It asks a person, not a rule. Conditional logic based on field values already existed. This is a request routed to a human who has to answer.
It branches on the answer. Approved and rejected are distinct paths. You can send an approved change request to the next stage and route a rejection back to the requester with a reason, in one workflow rather than two.
It also sits in a clear sequence of platform releases. On July 16, 2026, monday.com shipped an MCP block for AI Workflows, letting workflows run actions against any third-party app exposing a public MCP server. On July 9, automations became manageable through Sidekick and MCP without UI access. The pattern is a platform giving workflows more reach — and now, an explicit place to put a brake.
Why it matters for project managers
Most PMs who have looked seriously at workflow automation have stalled in the same place. The mechanics are easy. The objection is governance: who approved this? A change request that auto-advances, a budget line that moves without a signature, a status that flips to Green because a formula said so — each is a small erosion of the audit trail you are accountable for.
A native approval block turns that objection into a design decision. You choose where the gate goes. Everything before it runs at machine speed; the gate is where a named human takes responsibility.
That framing matters more now that AI agents are taking actions inside these platforms rather than just suggesting them. An agent that files a ticket is convenient. An agent that closes a risk or reassigns scope needs a checkpoint. The approval block is the checkpoint — an audit point at the moment of consequence rather than a report generated after it.
If you are formalizing where those checkpoints belong on your own projects, the MPUG on-demand class on governing AI-assisted delivery in the enterprise works through change control and traceability for automated and AI-generated work.
How this compares across platforms
Approval routing is not new to project tooling — what is changing is that it is being built into AI workflow builders specifically, where agent actions need a gate. Atlassian’s Jira automation has long supported conditional rules, and the Jira 2026 Summer release leaned further into agents taking on recurring coordination work. Every major platform is converging on the same question: how much do you let an agent do before a person looks at it?
Practically, that means comparing tools on approval semantics, not feature counts. Can a workflow pause indefinitely? Who gets notified, and what happens on a timeout? Is the approval recorded somewhere an auditor can find it? Those questions decide whether automation survives your first audit. MPUG’s tool comparison resources and PM glossary are useful when you are building that evaluation.
What to do next
- Map your existing gates before you automate anything. List the decisions on your project that currently require a signature. Those are your approval blocks, already identified.
- Start with a low-stakes gate. A status change or a document publish, not a spend approval. Learn the notification and timeout behaviour before it matters.
- Decide what happens on no response. An approval request that nobody answers is a stalled project. Define the escalation path in the workflow, not in your head.
- Check where the approval is recorded. Confirm the record is retrievable months later, before you rely on it for governance.
- Build the platform skills alongside the governance. MPUG’s master class library covers automation and planning across major PM platforms, including Jira, and sessions count toward PDUs you can log in the PDU tracker.
Frequently asked questions
What is a human-in-the-loop block? A step in an automated workflow that pauses execution, sends an approval request to a person, and then continues down a different path depending on whether they approve or reject. It puts a human decision inside an otherwise automated sequence.
Is this available on all monday.com plans? monday.com’s What’s New entry announces the capability for the monday platform but does not specify plan-level availability. AI Workflows features have historically varied by plan and AI credit allocation, so confirm against monday.com’s current plan documentation or your account team.
Does this replace traditional approval automations? Not necessarily. It adds human approval routing inside the AI Workflows builder specifically, which is where agent-driven actions run. Existing board-level automations continue to work as before.
The bottom line
The interesting part of this release is not the block — it is what the block admits. Platforms are building agents that take real actions, and they now need somewhere to stop. A pause-and-branch approval step is a small feature that makes automated project workflows defensible, and it puts the design question back where it belongs: which decisions on your project should never happen without a person saying yes?
Source: monday.com What’s New — Route workflows based on human approval, August 4, 2026. Verify plan availability, AI credit consumption, and approval audit behaviour against monday.com’s current documentation before rolling this into a governed process.
