A revops manager wants to ship a custom prospecting agent tuned to their ICP and the engineering team is two quarters out on the roadmap. Breeze Studio is HubSpot’s answer to that gap — a no-code builder that turns ops people into agent operators without filing a ticket. Useful and bounded, like every no-code platform. Knowing where the boundary lives keeps you out of trouble.
What Breeze Studio is
Breeze Studio lets you configure pre-built Breeze agents for your workflow or build entirely new agents from a base template. Beta in 2026. Sits alongside the 20+ agents and assistants in the broader Breeze ecosystem.
You can:
- Configure existing agents (prompts, escalation rules, channels)
- Build new agents from templates
- Set up triggers and routing
- Define structured outputs
- Pick which actions the agent can take
- Configure approval workflows for high-impact actions
You cannot:
- Write arbitrary code in the agent loop
- Add custom integrations not in the action library
- Override platform safety rails
- Deploy outside the Breeze runtime
Who it serves well
Good fit:
- RevOps customizing the Prospecting Agent to ICP signals
- CS leads tuning Customer Agent for support playbook
- Marketing ops shaping Content Agent for brand voice
- Sales ops adjusting agent tone per region
Poor fit:
- Custom integrations to internal systems
- Complex multi-agent orchestration
- Real-time dependencies on external data sources
- Use cases requiring deterministic output (compliance)
The 80 percent of agent customization work fits within the no-code ceiling. The remaining 20 percent still needs developers.
What you can build
Common patterns:
- Industry-specific Customer Agent
Trained on your KB
Routes by product area
Escalates with playbook context
- ICP-tuned Prospecting Agent
Filters by your firmographic criteria
Scores by engagement signals you define
Surfaces with personalized outreach draft
- Brand-voice Content Agent
Style guide loaded
Approved phrasing patterns
Banned terms enforced
- Data hygiene Data Agent
Per-property cleansing rules
Standardization patterns
Confidence threshold tuning
The agent marketplace and templates accelerate common patterns — start there before building from scratch.
Configuration anatomy
Agent definition:
Name: "Acme Support Agent"
Description: brief description
Base agent: Customer Agent
Knowledge sources: KB articles, product docs
Tone: friendly-formal
Languages: en, es, fr
Triggers:
Channel: chat, WhatsApp, email
Filter: ticket pipeline = tier 1
Available actions:
Search KB
Update ticket properties
Escalate to human
Send templated reply
Approval required:
Refund > $500
Account closure
Custom legal language
Escalation:
Confidence < 0.7
Customer requests human
3+ failed attempts
Save versions as you iterate. Roll back to a known-good version when a tuning change regresses.
Beta caveats and production posture
Recommended use during beta:
- Internal tools and ops automation
- Low-stakes workflows
- Channels with human review queue downstream
Not recommended during beta:
- Customer-facing without extensive review
- Decisions with financial impact
- Compliance-sensitive interactions
- Sole-arbiter use (no human in loop)
Beta features can change. Build with the assumption that contracts may evolve and your prompts may need re-tuning.
Iteration cadence
Day 1-7: daily review of every action, manual approval queue
Week 2-4: daily review of exceptions, structured feedback
Month 2: weekly review, prompt updates batched
Month 3+: monthly review, automated alerting on regressions
Calibration requires real traffic. Synthetic test cases catch obvious bugs but miss the long tail of edge cases that appear only at volume.
Governance for citizen builders
When ops people can build agents, you need governance to prevent agent sprawl:
- Registry: every agent in a shared sheet with owner, purpose, status
- Naming convention: standardized prefixes
- Approval: agent goes live after revops sign-off
- Review cadence: quarterly audit per agent
- Retirement: archive agents with low usage or low quality
Without this, you wake up to 40 agents nobody can explain.
What to do this week
Pick one agent your team would build today, prototype it in Breeze Studio with a 5-day approval-required period, document the configuration in a shared registry, and define the success metric before promoting it past pilot.