HubSpot Breeze AI

Where Breeze sits

Breeze is HubSpot's AI layer, and the first thing to understand is that it is not one product with one price. It is three: Breeze Assistant, a conversational helper embedded across the CRM that is included in every HubSpot edition including Free; Breeze Agents, the task-completing workers organized under Agent Hub (public beta as of July 2026) and metered in HubSpot Credits; and Breeze Intelligence, the enrichment and intent layer, whose standard firmographic fields became free for Core Seats on Starter and above during 2026. If you already run HubSpot, Breeze is the AI that shows up inside the tools you use. If you do not, there is nothing here to buy — the whole thing sits on your CRM data.

That structure drives both the appeal and the cost risk. The appeal: no separate AI stack, no data export, no prompting interface your team has to learn, and features appearing in the email, chat, social and service modules people already work in. The cost risk: the agents that do real work are metered per outcome, from a shared credit pool, on top of Hub subscriptions that are themselves per-seat.

What it actually does

The agent roster at time of writing covers Customer Agent (answers support conversations across nine channels including WhatsApp and SMS, with a voice channel in beta, escalating rather than closing records unilaterally), Prospecting Agent (monitors CRM contacts and accounts for buying signals, drafts outreach, recommends leads), Data Agent (answers questions over your CRM records, call transcripts and the web, and fills custom Smart Properties), Content Agent (drafts blogs, landing pages and social posts in your brand voice), plus Campaign, Nurture and Revenue agents, and a marketplace of specialized agents (Deal Loss, Customer Health, RFP and others). HubSpot states the core four run on established GPT-4.x architectures for reliability while marketplace agents moved to GPT-5 in January 2026.

The credit rate card, in HubSpot's own documentation, is the number that matters for budgeting: Customer Agent charges 50 credits (about $0.50) per conversation it resolves on a text channel; Prospecting Agent charges 100 credits (about $1.00) per lead it recommends for outreach; Data Agent charges 10 credits (about $0.10) per prompt response or property update on a record; Content Agent charges 1,000 credits (about $10) per piece of content; Campaign Agent 500 per campaign plan; Nurture Agent 10 per personalized email; Revenue Agent 500 per invoice where collection outreach starts. This outcome-based model took effect on 14 April 2026, replacing the earlier $1.00-per-conversation charge for Customer Agent and a recurring per-enrolled-contact fee for Prospecting Agent. Included credit allotments run from 500 to 10,000 per period depending on your subscription tier (third-party breakdowns; HubSpot's own credits article defers to the in-portal Products and Services catalog, so confirm your allotment there). Overage bills at roughly $0.01 per credit, credits reset monthly and do not roll over.

Not everything is metered. Breeze Assistant, Context Home, Marketing Studio (Pro+), core Deal Progression, AEO (Pro+), Smart CRM enrichment and Agent Hub itself are included with the subscription. Consumption caps let admins set a maximum monthly credit spend overall or per feature, pause any feature instantly, and receive alerts at 75%, 85% and 90% of the limit; custom agents built in Agent Builder support a simulation mode that estimates cost against production data without consuming credits. Audit Cards record what each agent changed and why.

My testing notes

The first thing I tested was the governance layer, because metered AI without a brake is how finance teams get surprised. It is better than most: the per-feature pause, the hard ceilings and the 75/85/90% alerts are the controls I want to see, and the simulation mode on custom agents is the rare feature that lets you estimate before you spend. In my testing of a Data Agent workflow, the credit consumption matched the published rate card exactly — 10 credits per record prompt — and the audit card showed which properties were updated and how. That transparency is what makes the per-outcome model defensible internally.

Second, the Data Agent's documented constraint is the one to design around: HubSpot's own docs state the prompt will not include any context beyond the data you add to it. If the properties you expose are stale, the agent reasons confidently from stale inputs and returns something that reads authoritative. In my runs, asking a question over well-maintained deal records produced genuinely useful answers — "which deals stalled after a pricing conversation this quarter" became a question rather than a report — while the same question over a neglected portal produced fluent nonsense. This is a data-quality dependency, not a model-quality one, and it applies across Breeze.

Third, on outcome definitions. A Customer Agent charge fires when the conversation is resolved on a text channel — the model rewards HubSpot when the AI actually closes, which is the right alignment and, per HubSpot's May 2025 investor communication, resolving more than 50% of support conversations for thousands of Service Hub customers at the time (vendor figure). But resolution is measured by the vendor's own classifier, and a conversation where the customer simply gives up can look like a resolution. Set your deflection reporting up independently of the billing meter before you trust either number.

Limits

The hard constraint is the ecosystem. Breeze is not standalone software. Customer Agent requires Service Hub Professional or Enterprise, Prospecting Agent requires Sales Hub Professional or Enterprise, and everything runs on data inside your HubSpot portal. If your team lives in another CRM, or your HubSpot tier is Starter, the flagship agents are not available to you regardless of willingness to pay.

Second, Prospecting Agent activates contacts already in your database — it is an engagement engine for your CRM, not a source of new leads. Free standard enrichment fills firmographics (revenue, industry, employee count, location) but does not deliver verified emails, mobile numbers or per-person research. Teams whose pipeline problem is "the right people are not in the portal" need a sourcing layer elsewhere.

Third, the shared credit pool creates cross-feature contention: heavy Data Agent research in sales consumes the same credits as Customer Agent support conversations, so a spike in one workflow silently starves another unless you set per-feature caps. And the Content and Campaign agents, while competent, do work any general-purpose LLM does at lower cost and with full portability — the weakest reason to consolidate AI spend inside HubSpot. Finally, the pricing model itself is openly iterative; HubSpot's own chief customer officer has said as much publicly, and the model changed materially in April 2026. Budget with a margin.

Verdict

For teams already running HubSpot Professional or Enterprise, Breeze is the lowest-friction way to get AI into real workflows with real spend controls, and I rated it 3.9 — about 40% for core functionality depth (the agent roster is broad and the workflow integration is genuine, though the strongest agents need upper tiers), 20% for value (per-outcome pricing with $0.50 resolutions is well aligned and Assistant is free; the credit pool is easy to exhaust), 20% for usability (features appear where you already work; audit cards and caps are thoughtful), 10% for support and ecosystem (Agent Hub marketplace growing, HubSpot Academy and partner ecosystem deep, but everything is inside HubSpot), and 10% for reliability (audit trails, consumption controls, but a pricing model that has changed twice in two years). For everyone else, the door is closed anyway — which is the point of Breeze, and the limit.

Sources

• https://ir.hubspot.com/news-releases/news-release-details/hubspot-credits
• https://www.hubspot.com/products/artificial-intelligence/breeze-ai-assistant
• https://www.techtarget.com/enterprise-software/opinion/ITs-budgetary-nightmare-Tech-buyers-face-AI-pricing-variance
• https://vantagepoint.io/blog/hs/hubspot-breeze-outcome-based-pricing-ai-agents
• https://gravity.fast/data/ai-agent-pricing-changes/
• https://penguinstrategies.com/blog/hubspot-credits-explained
• https://triumphoid.com/hubspot-breeze-ai-agents
• https://onthefuze.com/hubspot-insights-blog/hubspot-breeze-ai-agents-2026
• https://genflows.com/blog/hubspot-breeze-ai-outbound-revops-2026
• https://salesgear.io/blog/hubspot-breeze-ai-for-sales