AI agents & chat assistants

Useful agents for routine conversations, with people still in control.

We build narrow, practical agents that collect information, answer routine questions, prompt next steps and hand sensitive work back to a person. The aim is not to replace judgement, but to remove avoidable chasing and make the conversation easier to manage.

Where agents help

Good agents handle one useful job clearly.

Intake and first contact

Collect the right information, ask sensible follow-up questions and turn a conversation into a structured record for your team to review.

Routine questions

Answer common questions from your own policies, service notes or knowledge base, with a clear route back to a person when needed.

Reminders and next steps

Prompt people to complete forms, book appointments, provide missing information or understand what happens next.

Internal team support

Help staff find process guidance, summarise case context, prepare draft responses or understand where a piece of work stands.

Guardrails

Built for useful support, not unchecked automation.

For health, rehabilitation and other case-based services, the most important design work is deciding when the agent should answer, when it should ask for more information and when it should stop.

  • Start with one narrow job, not a general-purpose chatbot
  • Use source material and make limits clear
  • Hand sensitive, uncertain or high-risk matters back to a person
  • Keep logs, ownership and review points visible
  • Test with real examples before widening the scope

Already in use

Cassie already uses this pattern for real operational work.

In Cassie, agent-assisted handling helps monitor incoming messages, prepare replies for review, support routine prompts and keep admin moving without removing human oversight.

See Cassie

How we build

Start narrow, test properly, then decide whether to extend.

01

Choose the job

We define one practical conversation the agent should handle: intake, questions, reminders, triage or internal support.

02

Build the guardrails

We connect the right source material, shape the prompts, define escalation rules and decide what the agent must not do.

03

Test and extend

We test with realistic examples, review the outputs with your team and only then decide whether to widen the agent's role.

Human handover

Any agent should make the next step clearer, not hide uncertainty.

Start

Tell us the conversation you want to improve.

We can help decide whether an agent is useful, what it should handle and where human review needs to remain in the loop.

Enquiry

Start a scoping conversation.

An initial conversation, without charge, to understand your process, current constraints and the most appropriate next step.

We respond within two working days.