Skip to content

Capabilities

AI systems: the model is the smallest part

A model can read, write and reason about language. Whether that helps your business depends on everything around it: what it's shown, what it's allowed to do, what checks its output, and who decides when it matters. These fourteen capabilities build that surrounding system, and we use the least AI that does the job.
ModelContextsources, permissionsChecksschema, rulesPeopleapproveRecord · every steprequestactionnot allowedto a person

Use the least AI that does the job

Each rung trusts the model with more. Climb only when the rung below can't do the work, because every step up adds behaviour to test, watch and explain.
  1. Rung 4

    Staged generation

    Climb when the next step depends on what the last one found.

On every rung

Not every AI feature is a chat box

The interface decides how much a person controls. We pick it on purpose.
  1. Tool

    You trigger each action

    A button that drafts a reply or fills a form. The person stays in charge of every step.

  2. Chat

    You ask, it answers

    Open questions over your sources. Useful when people don't know what they'll need.

  3. Agent

    You delegate, it reports

    Multi-step work you hand over, with a view of what it did and where it's waiting.

  4. Copilot

    It suggests, you decide

    Quiet suggestions inside the screen people already use.

Start with the symptom

“Staff can't find answers we've already written down”Internal Knowledge AssistantAnswers from your documents, with sources, inside each person's permissions.
“We retype the same paperwork every day”Document Intelligence SystemDocuments become records; people only see the exceptions.
“Calls go to voicemail when we're busy”AI ReceptionistAnswers, qualifies and books by your rules, with a person one transfer away.
“We tried a chatbot and stopped trusting it”AI Evaluation & GuardrailsDecide what a good answer is, test against it, and limit what it may do.