Key findings

• Breeze includes assistants, embedded AI, and agents with edition-dependent access.

• Workflow actions can analyze, research, summarize, categorize, and populate record data.

• Professional or Enterprise subscriptions are required for many workflow AI capabilities.

• All agents consume HubSpot Credits from July 23, 2026, and estimates may differ from actual use.

• CRM cleanliness, access settings, consent, and review gates determine production quality.

What Breeze includes

Breeze is HubSpot's AI platform rather than one feature. HubSpot describes three broad layers: Breeze Assistant for in-context help, embedded AI features throughout the customer platform, and Breeze Agents that perform higher-volume tasks. Current examples include prospecting, customer service, data research, and content-related agents. Breeze Studio is the environment for discovering, customizing, and managing agents and assistants, with some capabilities still labeled beta.

For an SMB, the important distinction is between assistance and automation. An assistant helps a person prepare, analyze, or draft. An embedded feature accelerates a task in the existing interface. An agent or workflow action can process records and pass outputs into later automation. These layers have different risk and credit profiles. A generated call summary is not the same operational commitment as an agent that researches an account and triggers outreach.

How Breeze works with workflows

HubSpot's workflow documentation shows several ways Breeze can be used. Data Agent actions can analyze, summarize, categorize, research, or fill supported properties for enrolled records. Outputs can feed normal actions and branches, allowing deterministic logic to control what happens next. HubSpot also documents a Run Agent workflow action in beta and Breeze Assistant support for building workflow triggers and actions.

A sound architecture keeps the agent's role narrow. For lead intake, a workflow can enroll a contact when required fields are present, ask the Data Agent to classify the inquiry against defined criteria, and branch on a controlled output. High-confidence routine records can receive a task or draft. Ambiguous, strategic, or sensitive records should go to a person. The workflow, not the model, should enforce consent, lifecycle-stage rules, owner assignment, and duplicate prevention.

Availability is subscription-dependent. Current knowledge-base pages list Professional and Enterprise editions across relevant Hubs for many AI workflow actions, and HubSpot Credits are required. Confirm eligibility in the exact portal because beta access and product packaging can vary.

Where Breeze creates SMB value

Breeze is strongest when the information required to do the work already exists in HubSpot. A sales manager can use CRM context to summarize a deal, identify missing fields, prepare meeting context, or research a company. A service team can use approved support content and customer history to resolve or route common issues. A marketing team can classify form responses, prepare segments, or create drafts that remain subject to brand and compliance review.

This reduces integration overhead compared with exporting data to a separate AI tool. Permissions, record relationships, workflow enrollment, and downstream actions are already part of the platform. It can also reduce context switching for users. The commercial caveat is platform concentration. If key operational data lives in an ERP, product database, or support system outside HubSpot, connectors and data synchronization still determine whether Breeze has complete context.

Do not automate around a dirty CRM. Duplicate companies, stale ownership, inconsistent lifecycle stages, and missing consent fields will be amplified. Data cleanup and process ownership are prerequisites, not post-launch optimizations.

Credits, editions, and total cost

HubSpot combines edition-based access with HubSpot Credits for selected AI work. Its current Breeze page describes included capabilities and credit allocations by edition, while individual knowledge-base pages specify subscriptions and credit requirements. HubSpot's July 2026 documentation states that all agents consume credits from July 23 and recommends reviewing estimated cost per execution and setting monthly run limits.

An estimate is not a fixed tariff for every run. HubSpot notes that actual credit cost can vary with run complexity. Forecast from production-like records, not a single demonstration. Calculate expected enrolled records, agent runs per record, retries, and seasonal peaks. Add core seats, the required Hub edition, implementation, data cleanup, review time, and any connected systems.

Plans, included credits, and agent pricing can change. Check the product page, pricing page, and in-portal estimates immediately before purchase or rollout. Configure alerts and monthly limits so a poorly scoped workflow cannot consume the entire shared credit pool.

Data controls and trust

HubSpot provides account-level AI settings that control access to generative features and CRM data. Relevant settings must be enabled for workflow AI actions. Administrators should decide which teams and roles can build, run, and change agents. A builder's ability to create a workflow should not imply permission to send sensitive fields to every AI feature or to activate customer-facing actions.

Review HubSpot's AI Trust documentation and model cards for the features being used. The Data Agent custom prompt documentation notes that its model is not connected to the internet, while research-oriented actions may have different data behavior. This is why controls must be evaluated feature by feature. Limit prompts to necessary data, avoid unnecessary sensitive information, and document which records and properties are eligible.

For outbound content, require review until acceptance data is strong. For record updates, constrain allowed properties and values. Keep an audit trail of the source record, generated output, workflow branch, resulting action, and correction.

A 30-day implementation plan

Week one should define one use case and baseline. Measure current handling time, queue delay, errors, and volume. Clean the minimum required properties and identify a process owner. Week two should build the workflow in a sandbox or controlled segment, define expected structured outputs, and create a test set with normal, incomplete, contradictory, and sensitive records.

Week three should run in review mode. The agent prepares research, classification, or a draft, but a person accepts or corrects it. Record acceptance, correction reasons, latency, and credit usage. Week four should automate only the low-risk branch that met the quality threshold. Keep exceptions manual, set a monthly run limit, and schedule a recurring review of prompts, properties, knowledge, and workflow history.

A useful launch gate is not 'the workflow ran.' It is that the business owner can explain the decision rules, the admin can stop it, users know how to report an issue, and the expected savings exceed platform and review cost.

When to choose Breeze or an external platform

Choose Breeze when HubSpot is the primary customer system and the use case depends on CRM records, conversations, HubSpot content, or HubSpot workflows. Choose an external orchestrator when the process spans many non-HubSpot systems, requires custom hosting, or needs technical transformations beyond the portal's practical limits. A hybrid is common: HubSpot remains the system of record while an external workflow handles specialized processing through narrowly scoped APIs.

Avoid duplicating logic across HubSpot and another platform without clear ownership. Decide where enrollment, consent, deduplication, and final state live. As of July 30, 2026, several Breeze Studio and workflow capabilities remain beta or plan-dependent, and credit treatment has just changed. Confirm the current product state and retest before moving a revenue-critical process from review to autonomy.

Sources and methodology

This article synthesizes the primary sources below as of the publication date. Forecasts and recommendations are directional scenarios, not guarantees; they should be tested against your workflow, data, risk tolerance, and current vendor documentation.

HubSpot: Breeze AI product overview (accessed 2026-07-30)

HubSpot: Use Breeze to manage data in workflows (accessed 2026-07-30)

HubSpot: Run agents using workflows (accessed 2026-07-30)

HubSpot: Build workflows using Breeze Assistant (accessed 2026-07-30)

HubSpot: Review estimated agent credit costs (accessed 2026-07-30)