How it works
Every AgentWorks agent is built from the same six parts, in the same order. The parts are what keep an agent on your business and on its task. The order is what keeps it from going live before it has been proven.
The six parts
Part 1
The intelligence core
Models draft. Rules decide.
A language model is good at reading a message and writing a reply. It is not where a business should keep its decisions. So the decisions in each job are written as rules a person can read, in a fixed order, and the first rule that matches wins. The model drafts the words inside the decision the rules have already made.
On the platform today: When a call or message comes in, a fixed ladder of rules decides what it is and where it goes before any model is asked. The model's own opinion is recorded beside it and compared, not obeyed.
Part 2
The industry overlay
One engine, 27 overlays.
There is one platform. Each of the 27 industries is an overlay on it: the records and fields, the pipeline stages, the vocabulary, the ambassador who speaks it. A job proven in one industry can be laid over another by changing the overlay, not the engine.
On the platform today: All 27 overlays come from one registry, and the same registry sets up the CRM, the agents and this website. There is no second copy to drift.
Part 3
Lighthouse
What an agent knows about your business, in one place.
Lighthouse is the library each agent reads before it acts: your policies, your prices, your process, the documents your team has verified. Every document carries a status, a last-verified date and an expiry, so an agent works from what is current and can say when it is not.
On the platform today: Lighthouse is read on every agent call, scoped to your business. A shared core holds what every business needs; your own library sits on top of it and no other business can see it.
Part 4
Metadata as the connector
The record is the definition, and the agent reads the same one you do.
Your records are not hard-coded. The objects, fields and stages of your industry are definitions, and the CRM, the agents and the reports all read the same definitions. Change a field and everything that uses it sees the change. Connections to your other tools are accounts the agent works in, recorded against the same records.
On the platform today: EngageSuite360 builds every industry's screens from metadata, and an agent writing to a record is held to the same field definitions a person is.
Part 5
Workflows: policy and procedure
What an agent may do, task by task.
Each kind of work an agent can do carries a level: off, suggest, draft or autonomous. The level is granted per task by you, and some work has a ceiling nobody can raise: an outbound text, a deletion, a form submitted on another site never go past draft. Each agent's duties and guardrails are written down and compiled into it.
On the platform today: A task with no grant is treated as draft. A task set to off is refused and logged. Drafts wait in one approvals queue, and an agent cannot grant itself a higher level.
Part 6
Command oversight
Everything on the record, and a switch you can throw.
Every tool call an agent makes is logged with the agent, the target, the outcome and the time. Any agent can be paused from the console, and a paused agent is not woken by its schedule, a reminder or a heartbeat. An agent that causes two incidents in a week is stepped back to draft automatically.
On the platform today: The log is what the daily report and the activity feed are built from. The pause is checked at the door of every dispatch, so a paused agent takes no new work until you say so.
The six stages
An agent is one named job taken from a problem statement to a proven piece of work. It passes each stage before the next, and it does not go live until the fifth has been passed with you watching.
Stage 1
Frame
Name one job: what comes in, what is decided, what is recorded, what done looks like.
Stage 2
Model
Write the decisions in that job as rules a person can read. The model drafts inside them.
Stage 3
Bind
Bind the job to your records and tools through the metadata, so it reads what your team reads.
Stage 4
Proceduralize
Set the level per task, the ceilings, the approvals, the words it uses for your business.
Stage 5
Prove
Run it against real past cases, side by side with what your team did, before it goes live.
Stage 6
Deploy, then fork
Start at off or suggest, earn more task by task, and lay the proven job over the next industry.
This is the process behind every engagement. The parts themselves are listed on the facts page, and the controls on security.
Start with the job, not the software
The Industry Assessment frames the first job worth handing to an agent. It is free and arrives by email.
Take the Industry Assessment