KENNY CHIEN / PRODUCTS / DEEP AGENTS
Deep Agents
Long-horizon agents that decompose goals, spawn sub-agents, and persist state across days of work — built for enterprise workflows.
What is Deep Agents?
Deep Agents is a platform for long-horizon AI agents — agents that take a business goal, decompose it into a plan, spawn scoped sub-agents for the parts, and persist their state across sessions and days of work. It is built for enterprise workflows that do not fit inside a chat window: multi-step operations, research that spans a week, processes with human checkpoints in the middle.
The design premise is that autonomy without structure is a liability. Deep agents operate through a defined action interface — the entities, operations, and permissions of your business — so every step they take is one your organization explicitly allowed, and every decision leaves a trail. The unit of engineering here is not the prompt; it is the agent loop, governed.
A chat answers a question. A deep agent owns a goal — and keeps owning it after everyone has closed the tab.
How it works
Goal decomposition
A deep agent starts from a business goal, not a prompt. It breaks the goal into a plan of verifiable sub-tasks, each with its own definition of done, and revises the plan as reality pushes back.
Scoped sub-agents
For each sub-task, the agent spawns a sub-agent with exactly the permissions and context that task needs — nothing more. Failures stay contained; capability is granted, never improvised.
Persistent state
Work survives the session. Plans, progress, and intermediate artifacts persist across days, so an agent can pause for a human approval on Tuesday and resume exactly where it stopped on Wednesday.
Enterprise guardrails
Agents act through your ontology's action interface: typed operations with preconditions, permission checks, and a full decision trail. Human checkpoints sit wherever your risk profile demands them.
Who it's for
Enterprises with long-running workflows — operations, compliance, research, reporting — that outlast any single session
Platform teams that have outgrown single-prompt agents and are hand-rolling orchestration
Organizations that already model their business as an ontology and want agents that respect it
Leaders who want agent autonomy with auditability, not autonomy instead of it
Status and availability
Deep Agents is in development and available on request. Demos are arranged individually against scenarios close to yours, and early design partners shape what gets built next. Email me a description of the workflow you want to hand to an agent, and I will tell you plainly what works today and what does not exist yet.
Frequently asked questions
How is a deep agent different from a chatbot or a copilot?
A chatbot answers within a conversation and a copilot assists within a session. A deep agent owns a goal across sessions: it decomposes the goal into a plan, spawns scoped sub-agents for the parts, persists its state, and reports progress until the work is done or a human stops it.
What does my organization need before deploying deep agents?
An honest answer: an ontology helps enormously. Deep agents act through a defined interface of entities, actions, and permissions, and organizations that have mapped that layer get value fastest. The Ontology Canvas and Ontology Generator exist precisely to build it, and my agentic ontology consulting covers the same ground with me in the room.
Can I see it running today?
The platform is in development and demos are arranged individually. Email me with the workflow you have in mind, and I will show you the current state against a scenario close to yours — and tell you plainly what is not built yet.
Go deeper
Have a workflow in mind?
Describe the work you want an agent to own — the goal, the systems it touches, and where a human must stay in the loop. I will reply with an honest read and a demo when the fit is real.