2026 08 25 AIRops HUG #7 recording

Session #7 of the AIRops HUG, the HubSpot User Group focused on AI for revenue operations. HubSpot launched Agent Hub and Agent Builder a few weeks before this session. So we went straight to the source: two of the product managers building it. Becca Stamp (Senior Product Manager, HubSpot) and Davis Mastin (Product Lead for agentic automation and Agent Builder) walk through the whole thing live — the thinking behind it, a guided tour, two agents built on screen, and honest answers on the roadmap. What we covered: Why Agent Hub exists: everyone's told to adopt AI, but teams need one shared place to discover, activate, and manage agents — and to prove where the value is actually captured. Building an agent is hiring a teammate. Job description (role, goal, method), the actions and tools it may use, and the knowledge it carries. Manage agents the way you manage humans. Live build #1 — Customer Health Agent: a health rubric as agent knowledge, custom properties for its scores, Breeze Assistant reviewing your instructions, version history to roll back a bad edit. Test mode is the trust builder: it reads your CRM but never writes, never sends, never charges credits — refine until the output is right, then deploy. (One caveat: write actions through MCP tools can't be intercepted in test.) The new workflow builder vs legacy: a forgiving, Canva-like canvas that's object-agnostic — trigger from a schedule, a Slack emoji, a Google Sheet update, not just CRM record enrollment. Branch on agent output, pull in associated data beyond the trigger. The big roadmap answer, on the record: yes, this replaces legacy workflows. Compatible workflows get a safe one-click conversion first (announcements rolling out by INBOUND), mass migration later, with plenty of notice. Live build #2 — Customer Handoff Agent: three steps, browses the web plus your CRM, writes the context note the next owner always wishes they had. Credit control, demoed: portal-wide credit limits, per-feature limits, per-agent run limits, average-credit-per-run estimates from test runs, and run history — plus why the cost of an agent run varies with the size of the task, and confirmation that regular workflow actions still don't consume credits. Agent Hub vs MCP + CLI: prompted by a Salesforce comparison question, a genuinely useful decision framework — where the data lives, who the end user is, and why in-context governance (audit logs of what an agent changed) favors building inside the portal. Both approaches are legitimate; most orgs will use both.