Before adding AI agents to HubSpot, check three layers: clean, deduplicated data with one source of truth; explicit permissions for what the agent can read and do, with thresholds and an escalation path; and an assist-first review where a person approves outputs until accuracy is proven. Then there are two routes. HubSpot's Breeze agents (Customer Agent, Prospecting Agent, Content Agent) are switched on in the portal and billed in credits at $0.010 per credit. Custom agents are built on your own infrastructure for use cases the packaged agents do not cover. INSIDEA, an Elite HubSpot Partner rated 4.99 across 450+ verified reviews, activates Breeze agents at a fixed fee in 5 or 10 business days and builds custom agents on a seven-week plan, fixed fee from $2,000 and scoped at proposal, with evaluation suites and monitoring on every one.
First, the readiness check
An agent reads your data, acts within your permissions and produces outcomes your team has to stand behind. If any of the three is loose, the agent scales the problem faithfully.
The data
- · Deduplicate contacts, companies and deals
- · Enrich the gaps that drive targeting and personalisation
- · Standardise properties and picklists so stages mean one thing
- · Decide which system and field wins when they disagree
- · Archive or correct stale records
The permissions
- · Scope what the agent can read, especially sensitive fields
- · Scope what it can do: draft only, draft and send, update records
- · Set thresholds for autonomous action versus human first
- · Define escalation: when and to whom
- · Log every action so it is traceable and reversible
The review
- · Assist-first: the agent drafts, a person approves
- · Widen autonomy only as accuracy proves out
- · Human in the loop on anything customer-facing at the start
- · A named owner reviewing outputs on a schedule
- · Build the review path before go-live, not after the first mistake
Then, the two routes to an agent
HubSpot's Breeze agents
Already in the portal, switched on in settings, billed in HubSpot credits at $0.010 per credit. Best when the packaged use case matches yours: service conversations, prospect research and outreach drafts, marketing content.
Cost. Credits per action, published by HubSpot. INSIDEA activation: $300 Fast Track in 5 business days or $250 Standard in 10, with a credit forecast before switch-on.
Breeze agent activation →A custom agent on your infrastructure
Built for the use case the packaged agents do not cover. Triggered by HubSpot workflows, writes back to records, runs on your cloud with the code in your repository, ships with an evaluation suite and monitoring.
Cost. Fixed fee, scoped at proposal, from $2,000. Managed AI operations from $2,000 a month. Model costs paid to the provider directly.
Custom AI agents →HubSpot's per-action credit rates for every agent, quoted from its own catalog with the arithmetic for a real month, are at /hubspot/ai-credits.
How a custom agent gets built, week by week
Weeks 1 to 2, spec and evaluation set. The use case, the data it reads, the actions it may take, the success metric, and the test set the agent must pass before anyone trusts it.
Weeks 3 to 4, build. Versioned prompts, validated inputs, output schemas with retry logic, retrieval where the agent needs grounded facts, and the HubSpot integration through workflows and record updates.
Weeks 5 to 6, shadow then go-live. The agent runs alongside the team without acting, its outputs are compared to human decisions, thresholds are tuned, then it goes live with monitoring that pages on errors.
Week 7, hand-off. Your team owns the code, the evaluation suite and the dashboards. Managed AI operations is available from $2,000 a month for teams that want INSIDEA to keep running it.
A FinTech support triage agent resolves 62 percent of tier-1 tickets without a human, with handoff for the rest. A B2B SaaS lead-scoring agent reads product usage signals and updates scores every 15 minutes, tied to deal weight. An outbound agent personalising opening lines from public signals lifted reply rate 4.1x. A healthcare care-coordination agent pulls patient context from 12 systems into one timeline. Details at /ai/agents.
INSIDEA is an Elite HubSpot Partner rated 4.99 across 450+ verified reviews on the HubSpot Partner Directory, with five HubSpot Accreditations and an AI and Automation service line that is one of four equal pillars alongside CRM and Platforms, RevOps and Growth Marketing. 150+ in-house HubSpot-certified experts serve 1,500+ businesses across 25+ countries.
FAQ
Is my HubSpot portal ready for AI? What should I check?
Check three layers before switching anything on. Data: deduplicated contacts, companies and deals, standardised properties and picklists, one agreed source of truth, stale records archived. Permissions: what the agent can read, especially sensitive fields, what it can do (draft only, draft and send, update records), thresholds for autonomous action, and a defined escalation path. Review: an assist-first rollout where the agent drafts and a human approves, with every action logged, until accuracy is proven. If any layer is missing, the agent scales the mess.
How do I add AI agents to HubSpot for sales and support?
Two routes. HubSpot's own Breeze agents: Customer Agent for service conversations, Prospecting Agent for lead research and outreach drafts, Content Agent for marketing content. They live in the portal, are switched on in settings, and are billed in HubSpot credits at $0.010 per credit. Or a custom agent built on your own infrastructure that reads from and writes back to HubSpot through workflows, for use cases the packaged agents do not cover: lead scoring from product usage, support triage across systems, outbound personalisation from public signals. INSIDEA does both: fixed-fee activation of Breeze agents in 5 or 10 business days, and custom agents on a seven-week plan.
Which HubSpot partners build AI agents?
The HubSpot Solutions Directory lists partners by service, and AI-focused work is one of the filters, so check there for current listings. INSIDEA is an Elite HubSpot Partner rated 4.99 across 450+ verified reviews with an AI and Automation service line that builds production agents (lead scoring, support triage, outbound personalisation, call summaries) on the customer's infrastructure with evaluation suites and monitoring, and activates HubSpot's Breeze agents at a fixed fee. Published work includes a support triage agent resolving 62 percent of tier-1 tickets without a human and an outbound agent lifting reply rate 4.1x.
Which agency handles HubSpot AI automation and workflows?
Look for a partner that adds judgment to HubSpot workflows rather than replacing them: classification, routing, scoring and exception handling at the decision points where rules fall short, with confidence thresholds that send low-confidence cases to a person. INSIDEA runs this inside Operations Hub Enterprise through AI extension actions or in serverless functions called from HubSpot workflows, with the code in your repository. Light AI workflows run about $50 to $500 a month in model costs at volume; heavy multi-step ones can exceed $5,000.
What do HubSpot's AI agents cost to run?
HubSpot bills agents in credits at $0.010 per credit, sold in packs of 1,000 for $10. The rate depends on the agent and the action: 50 credits when Customer Agent resolves a conversation, for example. The per-action rates for every agent, quoted from HubSpot's own catalog, and the arithmetic for a real month are at insidea.com/hubspot/ai-credits. INSIDEA's activation service forecasts credit consumption before an agent is switched on and reconciles usage after.
What does a custom AI agent cost and how long does it take?
INSIDEA works fixed fee, scoped at proposal, and engagements start from $2,000, with the number locked before work begins. Managed AI operations start from $2,000 a month. Model API costs are separate and paid to the provider directly, typically a few hundred to a few thousand dollars a month for a standard agent at production volume. The build runs seven weeks: spec and evaluation set in weeks 1 to 2, build in weeks 3 to 4, shadow mode then go-live in weeks 5 to 6, hand-off in week 7.
Where does a custom agent run and which models does it use?
On your infrastructure: AWS Lambda, Google Cloud Functions, Vercel, Cloudflare Workers or your own Kubernetes cluster, with the code in your repository. INSIDEA is model-agnostic and uses Claude for most production agents, the GPT family for specific cases and Gemini where the Google ecosystem fits. Every production agent ships with an evaluation suite that runs on each prompt change, output schema validation, and monitoring that pages the team on errors.
How INSIDEA works
