Agents.
Agents are the payoff of the stack below them. On a dirty kernel they automate confusion.
Governed agents and MCP servers that run on a clean kernel. Child services live under this family slug: agents/breeze-agents, agents/custom-mcp-servers, and the rest.
Agents is Layer 04 of four.
Every service we offer maps to one of four layers. The Agents owns ai consulting & co-pilots. Each service in this family uses the slug agents, then its own name.
Teammates, not demos.
Agents are Layer 04, the top of the stack and the reason the other three exist. An agent is only as good as what it can read: the Surface it cites, the Voice it speaks in, the Kernel it queries.
Companies that skip straight to agents get demos. Companies that build the stack underneath get teammates.
The process.
Discover · maturity and blast radius
Where the company sits on the climb from individual tools to governed agents. What an agent is allowed to touch.
Expose · tools and resources
Company knowledge as MCP tools. Scoped questions. Logged answers. No pasted context.
Wire · Breeze plus custom agents
HubSpot native agents configured and supervised. Custom agents where Breeze cannot reach.
Operate · owner, log, review
Every agent has an owner, a scope, logging on, and a review cadence sized to its blast radius.
Agents that survive legal.
Things people ask before booking.
What is the difference between a copilot and an agent?
A copilot assists a person who is doing the work. An agent is assigned the work. They fail differently. A copilot's mistake gets caught by the human in the loop. An agent's mistake gets caught by governance, or it does not get caught.
Why do agent pilots fail?
The pilot ran on curated context. Production runs on your real data. The fix is not a better model. It is the three layers underneath.
What has to exist before agents work?
A kernel the agent can trust. A codified voice. Governance: an owner, a log, and a review loop. What is usually missing is the wiring between them.
Who supervises them?
A named person at your company. Every agent we ship has an owner, a scope, logging on, and a review cadence sized to its blast radius.
See how agents fits into The Agents.
Before anyone deploys an agent, the diagnostic answers the only question that matters: is the layer underneath ready? Ninety minutes, your stack on the whiteboard, and a straight call on whether agents come next or kernel work comes first.