HubSpot AI Agents Can Now Build Workflows. Your Portal Isn't Ready.
HubSpot AI Agents Can Now Build Workflows. Your Portal Isn't Ready.
Quick answer: HubSpot's newest beta update lets AI agents create workflows directly instead of only suggesting them. That shifts the risk in a HubSpot portal from bad reporting to bad automation. If duplicate contacts, stale lifecycle stages, or unmaintained deal properties exist in a portal, an agent will build on top of them without knowing they're wrong. Data governance and workflow visibility are no longer optional maintenance. They are the precondition for using this feature safely.
For years, AI inside HubSpot could draft, summarize, and recommend. Building the workflow itself, the enrollment logic, the branching, the actions, stayed a human task. A person was always the last checkpoint before automation went live. That checkpoint is what just moved.
What Changed
The beta update lets an AI agent construct a workflow end to end: enrollment triggers, branching logic, actions, all without a person building it step by step first. The agent isn't drafting a recommendation for someone to click through. It's publishing automation directly into the portal.
That's a real jump in capability. It's also a jump in risk, and the risk doesn't come from the agent. It comes from what the agent is building on top of.
The Foundation Problem
An agent building a workflow makes decisions based on the data it can see: lifecycle stages, deal properties, list membership, contact records. It has no way to know a lifecycle stage hasn't been touched in two years, that a deal property meant one thing in 2022 and means nothing today, or that half the contacts in a triggering list are duplicates of contacts already further down the funnel.
A person building that same workflow might catch the problem. They've been in the portal long enough to know a specific property is unreliable and to double check before using it as a trigger. An agent doesn't have that history. It has the data as it exists right now, and it will build on top of whatever that is, accurate or not.
Dirty data used to produce dirty reports. A messy lifecycle stage field meant a misleading dashboard: annoying, but contained. Dirty data feeding an agent that builds automation produces dirty workflows: automation logic built on a broken premise, live in the portal, enrolling real contacts and touching real deals. The failure mode moved from reporting to execution.
What "Not Ready" Actually Looks Like
Three patterns make a portal unsafe to hand to an agent.
Duplicate contacts. The same person exists as three records because of a form fill, a list import, and a manual entry that never got merged. An agent building a workflow off contact count or lifecycle progression is now working with three way split data instead of one accurate picture.
Stale lifecycle stages. A contact marked as an active lead from an interaction eighteen months ago that was never updated. An agent trusting lifecycle stage as a trigger condition will enroll that contact into logic built for a stage they left long ago.
Deal properties nobody has maintained. A custom property built for a sales process that changed two reorganizations ago. The field is still there, still populated with old values, still available for an agent to select as a filter or branching condition, with no indication that the values no longer mean what they used to.
None of this is exotic. It's the normal accumulated state of a portal that's been in production for a few years without a governance pass. It was tolerable when the only consumer of that data was a person who knew to be skeptical of it. It stops being tolerable the moment the consumer is an agent that takes it at face value and builds.
This Problem Predates the Beta. It Just Got Louder.
Workflow debt was already the norm in most portals long before AI agents could publish automation on their own. Christian Baun, co-founder of Howly and an eight-year HubSpot power user, covered this in a recent conversation with Jonas De Mets of Koalify: portals accumulate workflow spaghetti over two to five years, race conditions and bad re-enrollment logic quietly degrade the customer experience, and static documentation tools like Lucidchart or Miro go stale the moment the portal changes again.
The part of that conversation most relevant here is the discussion of what an AI-first HubSpot demands from a portal's foundation: visibility has to exist before automation is trusted to run itself. That's the same argument this post is making, from someone who has spent years inside portals that got this wrong.
https://youtu.be/SPPRFN43U7w?si=22mUflB6qGsoO4Nk
Relevant timestamps: the workflow debt problem starts at 2:30, the AI-first HubSpot discussion is at 15:53, and re-enrollment best practices run from 17:12.
The Advice That Used to Be Boring
"Clean your data" has been standard RevOps advice for as long as RevOps has existed as a discipline. It gets nodded at in a QBR and rarely gets prioritized, because there's always a campaign to launch or a report due sooner.
That calculus changes with agents that write to the portal. Data hygiene stops being cosmetic and becomes the control layer for what an autonomous system is allowed to build. A portal with clean lifecycle stages, deduplicated contacts, and current deal properties gives an agent an accurate map to build from. A portal without that gives an agent a plausible looking map that's wrong in ways nobody has audited recently.
Teams that treated data governance as an ongoing practice, not a cleanup project run once before a migration, are the ones positioned to use this capability safely. That's the whole point of the practice.
What to Check Before an Agent Touches Your Workflows
Audit for duplicate contacts and companies across every object type an agent is likely to touch: Contacts, Companies, Deals, Tickets, and any custom objects.
Review lifecycle stage accuracy on a sample of records, not just the field's existence. A populated field is not the same as a correct one.
Map existing workflow dependencies before adding new automation on top. An agent building a new workflow has no visibility into what it might collide with downstream: a duplicate enrollment, a property update another workflow depends on, a list-based trigger that overlaps with something already live.
Flag orphaned and stale workflows already running in the portal. An agent building new logic doesn't know which existing workflows are dead weight and which are quietly load-bearing.
Steps three and four are where most manual reviews break down. Dependency mapping across a portal with more than a handful of workflows is not something anyone does by scrolling through the workflow list.
This is where Howly fits. It's a read-only app, connected via OAuth, available on the HubSpot Marketplace. It builds the full dependency map, direct enrollment, list-based, and property-based, in 25 seconds depending on portal size. The Health Checker scores the portal from 0 to 100, and the Impact Analyzer shows the blast radius of a property change before anyone, human or agent, makes it. An AI Audit, powered by Claude, surfaces structural issues in about 15 seconds. Howly doesn't touch the portal. It shows what's already there, clearly enough to fix before an agent starts building on top of it.
The trial is 7 days, no credit card required, at howly.io.
Frequently Asked Questions
Does HubSpot's AI agent feature let AI make changes to my portal without approval? The beta update allows AI agents to create workflows directly, rather than only suggesting them for a person to build. Exact approval and permission settings depend on how the feature is configured in a given portal, but the core shift is that agents can construct working automation logic, not just recommend it. Anyone enabling this should confirm what approval gates exist in their specific setup before turning it on broadly.
Why does dirty data matter more now than it did before AI agents could build workflows? Previously, dirty data mainly affected reporting accuracy: a messy dashboard, a misleading count. Now that agents build automation directly from portal data, dirty data produces dirty automation: workflows built on stale lifecycle stages, duplicate contacts, or outdated deal properties, live and running against real records. The consequence moved from a reporting error to an operational one.
What's the fastest way to audit a HubSpot portal for readiness before enabling AI workflow agents? Start with a duplicate contact and company audit, followed by a lifecycle stage accuracy check on a sample of records. Then map existing workflow dependencies and flag any orphaned or stale workflows already running. Howly generates the dependency map and a portal health score in 10 to 25 seconds, which turns what would be a multi-day manual audit into a starting point for cleanup.
What is an orphaned workflow and why does it matter for AI-built automation? An orphaned workflow is an active workflow with no upstream or downstream connections to other automation in the portal. It matters here because an AI agent building a new workflow has no way to know an orphaned workflow exists or what it does. It can't account for logic it can't see. Surfacing orphaned workflows before adding agent-built automation reduces the chance of unexpected overlap or conflict.
Can I see how AI-built workflows connect to my existing automation? HubSpot's native workflow view doesn't show cross-workflow dependencies at a glance, whether a workflow was built by a person or an agent. Howly maps those connections, direct enrollment, list-based, and property-based, across the entire portal, so any new automation, agent-built or otherwise, can be checked against what's already running before it causes a conflict.
Summary
HubSpot's beta update lets AI agents build workflows directly, which means the accuracy of the underlying data now determines the accuracy of the automation built on top of it. Duplicate contacts, stale lifecycle stages, and unmaintained deal properties were tolerable when only humans were reading the data. They are not tolerable when an agent is building live logic from it. The portals in the best position for this shift are the ones where data governance and workflow visibility were already a real, ongoing practice, not a project run once before a migration.
Before enabling AI agents to build workflows, connect the portal to Howly and see the full dependency map and health score first. It's the fastest way to know what an agent would actually be building on top of.
Howly is a read-only HubSpot workflow mapping and audit tool. It maps workflow connections, flags structural issues, and shows the impact of property changes before you make them. Used by RevOps teams and HubSpot agencies managing complex portals at scale.
