HubSpot Breeze vs. custom Claude agent: which should you build?
The two routes to a customer agent, side by side. We build both, so this is the comparison we use on scoping calls.
| HubSpot Breeze Customer Agent | Custom Claude agent | |
|---|---|---|
| What it is | HubSpot's native AI agent, under Service > Customer Agent whichever Hub's Professional or Enterprise license unlocked it | A bespoke agent built on the Claude Agent SDK or API, connected to HubSpot through MCP or custom integrations |
| Subscription | Professional or Enterprise on any one Hub (Marketing, Sales, Service, Data, Content, Smart CRM or Commerce), plus HubSpot Credits for usage | No HubSpot tier requirement; cost is Claude API usage plus the engineering to connect it |
| Setup effort | No-code and live in about 15 minutes; real value depends on knowledge sources and escalation setup | An engineering build; 2 to 3 weeks at minimum |
| Channels | 9+ native: web chat, email, WhatsApp, SMS, Instagram, Telegram, LINE, Slack, Facebook Messenger, voice (beta) | Whatever you build connectors for; no pre-built marketplace |
| Knowledge | Knowledge base articles, website and landing pages, blogs, uploaded files, public URLs | Anything you connect: HubSpot, data warehouse, billing, internal APIs, docs |
| Governance | Seat and permission gated; CRM read and write toggled per agent, with approval gates on write actions | You grant tool access explicitly in code: more granular, more DIY |
| Pricing model | Outcome-based since April 14, 2026: $0.50 per resolved conversation, credits consumed only on resolution, free 28-day trial | Token-based usage, tied to consumption rather than outcomes |
| Model control | Abstracted away, no visibility or choice | You pick and route models per task; routing can cut cost 40 to 70% |
| Reach beyond HubSpot | Actions make live, scoped API calls to outside systems (order status, pricing, inventory) with no code | Core strength: reasons across CRM, finance, product database and more in one multi-step agent |
| Maintenance | Managed and updated by HubSpot | Owned by you or a partner; needs ongoing engineering |
| Track record | HubSpot reports 8,000+ activated customers, 65% average resolution and 90%+ for top performers | No standard benchmark; results depend entirely on build quality |
| Best fit | Support grounded in your CRM and knowledge base, with fast time to value | Work across several systems, bespoke multi-step logic, strict compliance or data residency |
| Weak spot | Actions are single-endpoint lookups, not multi-step orchestration across systems | No marketplace; needs sustained engineering and costs are less predictable |
Why customer agents fail
The agent is rarely the weak link. Six failure modes kill these programs, and the first two do most of the damage.
No human escalation pathThe big one
About 90 percent of companies will not let an agent close every loop, and they are right. Your team owns escalations, refunds and edge cases, and feeds the corrections back into the agent. We build and tune the system; your team keeps the escalation seat, or trust dies with the first bad answer.
Empty knowledge and contextOur specialty
An agent with no knowledge sources, guidelines or AI context deflects nothing and makes up the rest. What the agent knows, how it speaks and where its authority ends are engineered into every build.
No data access
If the agent cannot see orders, billing or account status, every second question becomes "let me transfer you". Actions into your outside systems are part of the build, not an afterthought.
Wrong channel rollout
Launching every channel at full coverage on day one is how agents fail in public. We start on one channel at partial coverage, watch resolution quality, then expand channel by channel.
One-agent thinking
A customer agent, a knowledge base agent and the workflows around them beat one agent trying to do everything. Every agent in the build has one job.
No feedback loop
Unresolved conversations are the training data. Without a weekly review that feeds them back into knowledge sources and guidelines, the agent never improves past launch quality.
The fuel: credits and cost
HubSpot's Breeze Customer Agent is outcome-based: since April 14, 2026 it costs 50 HubSpot Credits ($0.50) per resolved conversation, and credits are only used when a conversation actually resolves. There is a free 28-day trial and every plan includes a starting credit allowance.
Every build sets a credit budget so spend never surprises you, and we negotiate credit packages with HubSpot where we can (final pricing is always HubSpot's call). The full math is in our Breeze Customer Agent cost guide.