HubSpot Just Shut Down Every Custom AI Assistant — What That Means for the No-Code Agent Market

On 13 July 2026, HubSpot flipped a switch and made every custom AI assistant in its ecosystem read-only. No warning pop-up. No migration wizard. Just a knowledge base note and a redirect to something called Breeze Studio.

If you're a no-code builder who'd spent months configuring assistants for sales teams, customer support queues, or marketing workflows, that date mattered. Those assistants didn't break gracefully. They went silent. And now you're being told to rebuild them inside a platform you didn't choose.

This isn't a bug. It's the biggest no-code AI platform consolidation of 2026, and it's part of a pattern that's reshaping who gets to build AI agents, where they get to build them, and what happens when the platform owner changes the rules.

What actually happened on July 13?

HubSpot had been telegraphing this move since mid-June. On 19 June, accounts that had never created a custom assistant lost the ability to create one entirely. Then on 13 July, the hammer dropped: every existing custom assistant became read-only and was automatically migrated into a new entity called a Breeze project. The old interface? Gone. The old editing tools? Locked. Your assistants still technically existed, but you couldn't touch them anywhere except inside Breeze Studio.

And the migration wasn't clean. Welcome messages and conversation starters didn't transfer automatically. HubSpot's own documentation told users to "manually port" them over. If you had three or four custom assistants built across different teams (sales here, customer success there), each one needed a manual audit and rebuild.

That's not a migration. That's a relocation with extra steps.

What actually broke?

The practical damage fell into a few buckets.

First, the UI change was disorienting. Custom assistants had lived in a dropdown menu that teams used daily. Breeze projects live inside Breeze Studio, which is a different navigation path and a fundamentally different interaction model. Niall Heath, a HubSpot partner, asked the question a lot of people were thinking: "Will we have a snazzy big dropdown to easily choose? That was the advantage of making new custom assistants."

Second, Google Drive connectivity apparently got lost in translation. One user flagged that Breeze Assistant could access Google Drive like a knowledge vault, but Breeze projects couldn't. If your team had built assistants that pulled context from Drive documents, those workflows broke.

Third, there's now a naming problem. Domenique Bruno put it bluntly: "So now there will be 3 different 'Projects' in portals. It's a good word, but would love to see a little more differentiation!" HubSpot already had a Projects object. Now Breeze projects. Plus whatever other Projects were floating around. Good luck searching for anything.

And the community reaction wasn't just about inconvenience. Scott Fahy, who runs a customer care team, said: "Not gonna lie here, not a huge fan of this one. I've got a whole team of customer care folks who use the Custom Assistant every day as a fast way to find answers on a knowledge base, and now that seems like it's being deprecated or rolled into another place that's harder to get to."

That's the thing. These weren't experimental toys. Teams had built operational workflows on top of custom assistants. When you yank the UI and tell people to rebuild, you're not shipping a feature update. You're creating a migration project nobody budgeted for.

Was governance the real reason?

HubSpot's official line is governance. And to be fair, there's a real argument here.

Custom assistants had become a sprawl problem. Different teams built different assistants with different instructions, different knowledge sources, and different guardrails. A marketing assistant might pull from outdated brand guidelines. A sales assistant might recommend messaging that legal had never reviewed. Without a central governance layer, there was no way to audit what these things were saying or doing.

Breeze Studio solves that by giving admins one place to configure behaviour, instructions, and approved information sources. It also introduces proper team collaboration (shared projects with view, edit, and manage permissions) and connects to external tools like G2, Linear, Gong, and Amplitude through MCP servers.

But the timing tells a more interesting story. HubSpot is also sunsetting five specialised agents on 23 July: Social Post Agent, ABM Landing Page Agent, RFP Agent, Cross-sell/Upsell Agent, and Sales to Marketing Feedback Agent. That's a lot of agent deprecation in a single month. And Breeze pricing is credit-based: $0.50 per resolved customer conversation, $1.00 per prospecting lead, $0.10 per data agent answer.

The governance argument is real. But so is the revenue architecture underneath it. When every agent action routes through a governed platform that also happens to be the billing surface, the incentives are not subtle.

Is it just HubSpot doing this?

What makes the HubSpot story significant isn't that one company consolidated its AI tools. It's that everyone is doing it, and they're all using the same language to justify it.

Salesforce launched the new Agentforce Builder in February 2026, replacing its legacy Setup experience with a unified Agentforce Studio. It introduced Agent Script, a hybrid language that combines natural language prompts with deterministic business logic, all governed through the Einstein Trust Layer. Build anywhere else? You're on your own.

ServiceNow went bigger. At Knowledge 2026 in May, it unveiled ServiceNow Otto, unifying Now Assist, Moveworks, and AI Experience into one AI assistant. Its Build Agent now works inside Cursor, Claude Code, and GitHub Copilot, but everything deploys through the ServiceNow AI Platform, governed by AI Control Tower and metered per operation. The governance layer and the billing surface are the same object.

SAP took the most aggressive approach. Its April 2026 API policy explicitly bars third-party AI agents from making autonomous multi-step API calls. Want to connect a Copilot or a Claude agent to SAP? You route through Joule (SAP's own AI) over the A2A protocol. Forrester publicly urged CIOs to push back on this as gatekeeping rather than security.

Workday launched Developer Agent and Agent-Ready Tools in June 2026, all governed through the Workday Build platform with per-call Flex Credits metering. CEO Aneel Bhusri told analysts that AI is "bigger than SaaS itself." He was reading the revenue split.

Futurum Research's June 2026 Signal report crystallised the pattern: Microsoft, Salesforce, and ServiceNow are the Elite Zone vendors for enterprise agentic AI. The report argued that "competitive differentiation no longer hinges on model quality alone, but on how effectively platforms translate model intelligence into controlled, repeatable action."

Translation: the winners won't be the companies with the smartest AI. They'll be the companies with the best tollbooths.

What does this mean for no-code builders?

I think there are three things worth taking seriously here.

First, platform-native AI building is becoming a walled garden. The era of "build whatever you want, wherever you want" is ending. Every major enterprise platform is converging on the same model: you can build agents, but you'll build them inside our studio, on our governance rails, and you'll pay per operation. The freedom of the early no-code agent market, where you could spin up a custom assistant in an afternoon and run it for free, is being engineered away.

Second, the governance argument is not entirely a smokescreen. The Credo AI State of AI Governance 2026 report found that 60% of enterprises are scaling AI, but only 4% are governing it at scale. When agents take autonomous actions across hiring, financial transactions, and customer interactions, someone needs to answer for what they do. The question is whether that governance should be owned by the platform vendor or the enterprise deploying the agent.

Third, interoperability is the battleground that matters most. The vendors embracing MCP and A2A open standards (ServiceNow, Salesforce) are positioning themselves as the orchestration layer for heterogeneous agent environments. The vendors blocking external agents (SAP) are betting that their ecosystem lock-in is stronger than the market's demand for interoperability. For no-code builders, this means the platforms you choose today will determine which agents you can connect tomorrow.

The takeaway

HubSpot shutting down every custom assistant on a single Sunday in July isn't a one-off. It's the most visible example of a structural shift that's been building all year. Enterprise platforms are herding no-code AI builders into governed studios because that's where the control, and the revenue, lives.

If you're building AI agents on someone else's platform, you should assume the rules can change. Not might. Can. And when they do, the migration work will land on your desk, not theirs.

Build accordingly.