HubSpot has opened agent building to everyone on Professional and Enterprise, and the early numbers suggest teams were more ready for it than the forecasts assumed. Agent Hub and Agent Builder entered public beta on 23 July 2026, and by the time HubSpot reported second quarter results on 5 August, more than 2,700 customers had already switched them on. That is a quick start for a product measured in weeks rather than quarters, and it points at something useful: the appetite for building custom agents inside the CRM is considerably larger than the appetite for buying yet another standalone AI tool.
The wider adoption picture supports the same read. HubSpot told investors that more than 55 per cent of its Professional and Enterprise customers now use either its agents or Breeze Assistant, that monthly agentic actions across the customer base have grown more than threefold since January, and that weekly active use of Breeze Assistant has doubled over the same period. Agent usage on its own climbed from 9 per cent at the start of the year into the high teens by July. Those are adoption curves most enterprise software features never reach, and they were achieved in roughly seven months.
So the practical question for revenue leaders has changed shape. It is no longer whether agents belong in the CRM. It is which workflow to hand over first, and how to demonstrate the value quickly enough that the second and third follow without a budget argument. HubSpot has made that noticeably easier than the previous generation of AI features did, and the reason comes down to two design choices: a build surface that does not require a developer, and pricing that meters actual completed work rather than potential access.
What is HubSpot Agent Builder?
HubSpot Agent Builder is a no-code canvas, released in public beta on 23 July 2026 for Professional and Enterprise customers, that lets teams assemble custom AI agents and agentic workflows using plain language rather than code. It draws on the customer context already held in Smart CRM, so an agent can read contact, company, deal and ticket records without a separate integration project, and it can reach into connected external systems where the work requires it. Agent Hub sits alongside it as the single place where every agent runs, whether HubSpot built it, a partner built it, or the customer built it, with live status and recent results on one screen and agents grouped by business outcome rather than by product module.
The interesting part is not the canvas itself. It is who can now use it. Twelve months ago, an agent that read deal records, checked an external inventory system and then drafted a follow-up was a scoped development project with a partner statement of work attached. Today a revenue operations manager who knows the data model can assemble a first working version in an afternoon and iterate on it live. That collapses the distance between the person who understands the process and the person who builds the automation, which historically has been the single biggest source of delay and disappointment in CRM automation work.
It also changes what a good backlog looks like. When each agent took a quarter to deliver, the sensible strategy was to pick one large, high-value process and invest heavily in it. When an agent takes a week, the better strategy is a portfolio: several small agents covering high-volume, low-variance tasks, each one measured independently, with the weak performers retired without ceremony. That is a healthier operating model, and it is now available to teams who could not previously afford the experiment.
Why a shared control surface matters more than any single agent
Agent Hub answers a question that has quietly troubled every organisation running more than two or three AI tools: where does someone go to see what the agents actually did today? HubSpot’s answer is one screen showing every agent in the portfolio, its live status, and its recent results, organised around outcomes such as building demand, winning deals, delighting customers and scaling growth. Crucially, the same surface covers agents HubSpot shipped, agents a partner configured, and agents the customer built themselves, so there is no separate console to check for each origin.
This is the sort of infrastructure that looks unglamorous on a product roadmap and turns out to matter enormously in month six. Teams that adopt agents successfully tend to be the teams who can answer three questions on demand: what is running, what did it produce, and who owns it. A single control surface makes those questions answerable by a manager rather than by an administrator digging through logs, which in turn makes it realistic to expand the portfolio without adding oversight headcount. Governance stops being a policy document and becomes a screen someone actually looks at.
How much does HubSpot Agent Hub cost?
Agent Hub is included with eligible HubSpot Professional and Enterprise subscriptions, and the agents themselves consume HubSpot Credits as they complete work. One credit is priced at one US cent. The featured agents publish their rates openly: the Customer Agent charges 50 credits for each resolved conversation, roughly 50 cents; the Prospecting Agent charges 100 credits per outreach recommendation, roughly one dollar; and the Data Agent charges 10 credits per response, roughly ten cents. Custom agents built in Agent Builder draw on the same credit pool when they take configured actions. Each eligible tier includes a credit allowance, and additional credits can be bought as volume grows.
Published per-outcome rates are genuinely useful to a buyer, and it is worth saying plainly why. A cost of 50 cents per resolved support conversation can be compared directly against the loaded cost of a human resolution, which most service teams already track. A dollar per outreach recommendation can be compared against the cost of the research time it replaces. Neither comparison requires a modelling exercise or a vendor spreadsheet, and both can be checked against reality after a fortnight of live usage. That is a far more comfortable position for a buyer than per-seat AI pricing, where the cost is fixed on day one and the value has to be argued for afterwards.
HubSpot’s own chief executive framed the shift in exactly those terms on the earnings call, noting that customers adopting AI want proof of value before they commit, and predictability in what it costs. Metered pricing gives both. It also means a pilot can be genuinely small. A team can allocate a few thousand credits to one workflow, run it for a month, and make the expansion decision on measured output rather than on a projection.
How do you measure the return on an agent you built yourself?
Measuring CRM return on investment gets easier when the unit of work is metered, because the denominator is handed to you. For each agent, track four numbers over a fixed window: the count of actions completed, the credits consumed, the proportion of outputs accepted without human rework, and the time saved per accepted output measured against the previous manual process. Multiply accepted outputs by time saved and the loaded hourly cost of the person who used to do the work, then subtract the credit spend. That gives a defensible net figure per agent per month, and because it is calculated per agent rather than per platform, it tells you which parts of the portfolio to expand and which to retire.
The acceptance rate is the number worth watching most closely, and it tends to be the most encouraging one. An agent that produces output people use without editing is doing genuine work. An agent at 60 per cent acceptance is usually not broken, it is under-specified, and the fix is normally a clearer instruction or one additional data field in its path rather than a rebuild. Teams that treat acceptance rate as a tuning dial rather than a pass or fail verdict tend to reach a productive portfolio considerably faster.
What does the data underneath a good agent look like?
CRM data quality means the records an agent depends on are accurate, complete, consistent in format, and current enough for the decision being made. The reassuring part, and the part that often goes unsaid, is that an agent does not need a clean database. It needs the specific fields on its own path to be reliable. A prospecting agent needs industry, employee count, and recent activity to be trustworthy. It does not care whether a legacy custom field from a 2019 campaign is still populated. Scoping data work to the fields an agent actually reads turns a multi-quarter cleanup programme into a focused task that takes days.
This reframing is the most practical shortcut available to teams sitting on years of accumulated CRM debt, and it is why the portfolio approach works so well. Each new agent brings a short, specific data requirement with it. Meet that requirement, ship the agent, and the improvement is permanent and immediately paid for by the agent it enabled. After six or seven agents, an organisation finds it has quietly repaired the fields that matter most to its revenue processes, funded entirely by work that delivered value along the way. That is a far more sustainable path than a standalone data quality project that has to justify itself on principle.
How do you get your first agent into production in a fortnight?
Improving CRM adoption starts with choosing work people are glad to hand over. For a first agent, pick a task that is high in volume, low in variance, and currently done by someone who would rather be doing something else: routing inbound enquiries, enriching newly created companies, drafting first-touch replies to known question types, or flagging deals that have gone quiet against a defined threshold. Build it in Agent Builder, run it in suggestion mode so a human accepts or rejects each output, and review the acceptance rate weekly. Once acceptance sits comfortably above 85 per cent, let it act directly on the low-risk portion of the volume and keep human review for the rest.
Two habits make the difference between a pilot that expands and one that stalls. The first is naming a single owner per agent, someone accountable for its acceptance rate the way a manager is accountable for a metric. The second is publishing the results where the affected team can see them, ideally the same Agent Hub screen the agent runs on. Adoption follows visible evidence far more reliably than it follows training sessions, and teams who watch an agent quietly clear 400 routing decisions a week tend to arrive with their own suggestions for the next one.
The Sirocco perspective
What we find encouraging about this release is how well it suits organisations that were previously priced out of custom agent work. As an independent CRM partner working across HubSpot, Salesforce and Dynamics 365, we spend a lot of time helping teams decide where automation genuinely pays, and the honest answer used to be that bespoke agents only made sense above a certain deal size. Agent Builder moves that line down substantially. A mid-sized manufacturer or a growing services business can now run a real portfolio of agents against its own processes, and measure each one on published rates rather than on faith.
Our advice to clients this month is straightforward: use the beta window. Pick two workflows, build them, measure acceptance and credit consumption for four weeks, and go into your next planning cycle with your own numbers rather than a vendor benchmark. The teams who do this will be the ones setting the internal standard for what a good agent looks like, and that standard is a genuinely valuable thing to own before the rest of the organisation starts asking. The tooling has arrived, the pricing is legible, and the runway to a measurable result is now short enough to fit inside a single quarter.
If you are weighing up where agents fit in your HubSpot portfolio, or how the same approach compares across Salesforce and Dynamics 365, schedule a consultation and we will walk through it with you.
Get in Touch
Tell us which HubSpot workflow you are thinking of handing to an agent first, and we will help you scope it, price it in credits, and set the acceptance target it needs to hit.
