Explore how HubSpot's AI agents enhance productivity while understanding their limitations. Learn what you need for successful deployment.
.png)
There is a gap between how HubSpot’s AI agents are marketed and how they behave in practice. There, we said it. But that gap is not a defect, rather it’s more where a lot of businesses currently sit; either disappointed because agents didn’t deliver on expectations, or holding back from deployment because they’re not sure what they’re actually buying.
This article is an attempt to close that gap. Not to dampen the case for agents, because the case IS strong, but to be precise about what they currently* do, what they currently* depend on, and what currently* still requires human judgement.
HubSpot’s agent layer sits within its Breeze AI suite. Where Breeze Copilot assists; drafting emails, summarising calls, suggesting next steps, the agents are designed to act. They run autonomously through multi-step tasks without requiring human input at each stage.
HubSpot now offers over 20 agents and assistants through the Breeze Marketplace, with seven core agents in general availability across marketing, sales, and service. We’re going to focus on the four with the clearest production track records: the Prospecting Agent, the Customer Agent, the Content Agent, and the Knowledge Base Agent. Each one operates in a defined domain with none of them operating outside it.
That is worth just sitting and considering before we go any further - these are not general-purpose AI systems. They are specialists. And like humans in specialist fields, what they do within their domain, they do REALLY well, but what sits outside of that, currently* remains your team’s responsibility.
The Prospecting Agent is designed to handle the top of the sales funnel: identifying target accounts, monitoring buying signals, researching contacts, and drafting personalised outreach. HubSpot describes it as a 24/7 business development representative.
In practice, it works well when your CRM is in good shape. It uses your existing contact and company data to identify patterns, flags funding rounds and other trigger events, and generates outreach drafts grounded in what HubSpot knows about the prospect.
Where it falls short is, to be fair, predictable. If your data is incomplete, ie. missing job titles, stale contact records, inconsistent segmentation, then the agent’s outputs reflect that directly. Poor inputs produce poor outreach. Early adopters report early users saw outreach response rates at double the industry benchmark under clean data conditions. Which is amazing.
“Prospecting Agent is only truly effective if your CRM is exceptionally clean and well-tagged.”
The agent is also NOT a relationship builder. It handles volume and timing well, but it doesn’t handle nuance, negotiation, or the kind of contextual reading that experienced sales people bring to complex accounts.
The Customer Agent is the most mature of HubSpot’s current agent suite, and the one with the clearest deployment evidence. It is designed to resolve customer and prospect queries autonomously, drawing on your knowledge base, CRM history, and conversation context.
We have seen this agent perform in production across several client implementations, and the results are really consistent. In one case - a two-sided research platform managing participant support queries - the agent reached a 70-80% autonomous resolution rate from the moment it went live. Across the first nine months of operation, it handled over 400 conversations, operated 24 hours a day, and responded in whatever language participants wrote in, despite the team only providing knowledge in English. 24/7 response times and responding in any language is something they would never have been able to achieve using humans, unless they invested A LOT of money and resources.
What made it work was not sophistication, but clarity. The implementation mapped the 11 most common participant queries to structured short-answer responses within HubSpot. It wasn’t super complex integrations or elaborate configuration. The agent handled what it knew, escalated what it did not, and the team stopped thinking about it, which is exactly the outcome you want!
In a separate implementation, a business processing 700-800 customer sessions weekly deployed the agent to handle its highest-volume repetitive queries. The alternative was doubling the customer care team. The agent removed that pressure entirely.
The dependency worth noting here though is that the agent is only as good as the knowledge you give it. If your documentation is incomplete, inconsistent, or not maintained, the agent will either fail to resolve tickets or resolve them incorrectly. Knowledge governance is not optional, it is the very foundations the agent runs on.
The Content Agent generates marketing content at scale; blog posts, landing pages, case studies, social posts, all working from your brand voice settings and CRM data. HubSpot enhanced it significantly in 2025: it now uses uploaded reference files, suggests topics based on top-performing content, and automates pre-publish tasks including meta descriptions and internal linking.
It is really useful for volume. A team that needs to maintain consistent content output across multiple channels without the headcount to support it will find genuine value here.
It is not a substitute for editorial judgement, though. Content that builds authority, as in the kind that ranks, gets shared, and generates trust with mid-market buyers still requires perspective, specificity, and original thinking. The Content Agent can scaffold and produce, but it cannot originate a point of view that your audience has not read before.
Used well, it handles the operational side of content production while your team handles the intellectual side. However, if it is used poorly, it fills a website with content that says nothing distinctive. Not ideal when you’re trying to stand out against competitors.
The Knowledge Base Agent is specifically designed to identify gaps in your support documentation and fill them in real time, using incoming ticket data to understand what customers are asking that your knowledge base does not currently cover.
It works in conjunction with the Customer Agent; so when the Customer Agent cannot resolve a query because the answer does not exist in your documentation, the Knowledge Base Agent flags the gap and can draft the missing article. This is what HubSpot calls multi-agent orchestration: agents working together to improve each other’s performance over time. Beautiful.
The practical implication is that your knowledge base becomes self-improving within defined boundaries. It will not write policy, nor will it decide what your product should do. But it will reduce the manual overhead of maintaining support documentation as your product and customer base evolve.
The pattern across every HubSpot agent is the same. Performance is directly proportional to data quality and configuration precision.
This isn’t one of those caveats buried in small print. It is the central variable. An agent deployed on clean, well-structured data with clearly defined knowledge or outreach parameters will consistently outperform a manual process on volume, speed, and availability. An agent deployed on fragmented CRM data, incomplete documentation, or vague configuration will either underperform or produce outputs that require more human correction than the manual process it replaced.
The businesses seeing genuine returns from HubSpot agents are not the ones who deployed fastest but the ones who had the foundations in place: consolidated data, maintained records, documented knowledge, and clear definitions of what the agent was and was not responsible for.
To be direct about the current limits:
Before deploying any of these agents, there are four questions you need to ask yourself:
If you cannot answer yes to all four, the deployment work is not the first step. The infrastructure work is.
Done right, the commercial logic for HubSpot agents is straightforward. The businesses we have seen deploy them successfully are not processing fewer inquiries or producing less content. They are processing more, at the same team size, without the cost base that volume would previously have required.
That is the real argument for agents: not that they replace capability, but that they absorb scale. The research platform handling a growing volume of participant queries at the same headcount as a year ago. The customer operations team managing 700-800 sessions a week without doubling in size. In both cases, the agent absorbed load that the business couldn’t have sustained manually without significant additional cost.
HubSpot shifted its two flagship agents to outcome-based pricing in April 2026 to $0.50 per resolved customer conversation for the Customer Agent, and $1 per lead recommended for outreach for the Prospecting Agent, both with a free 28-day trial. That changes the risk calculation for early deployment considerably. You are not paying for potential, instead you’re paying for outcomes.
HubSpot’s AI agents are genuinely useful, but sadly, they’re not magic. They do a defined set of things really well, within a set of conditions that require real work to establish.
And like we said at the start of this article, the businesses that will see returns from this technology are not the ones moving fastest, they’re the ones building correctly with clean data, maintained knowledge, clear scope, and the discipline to treat agents as infrastructure rather than shortcuts.
If you are trying to work out where you sit on that spectrum, it is worth starting with an honest assessment of your current CRM and knowledge base health before committing to a deployment plan.
Centralise works with B2B businesses to build the revenue infrastructure that makes HubSpot and its AI layer, actually perform. If you’re considering agent deployment and want to understand what readiness looks like for your business, get in touch by filling out the form below.
Thinking about moving to HubSpot? Here's what to expect, from timelines and team involvement to how we handle remote teams and post-launch support.
Centralise began with co-founders Daniel O'Reilly and Calvin O'Callaghan, whose journey to launch a leading HubSpot agency started in the heart of...
We’ll show you what HubSpot results can look like over time, at 8 weeks, 6 months, and 12 months. Whether you’ve just started or are already a few...
Be the first to know about our new blog content!