Review and preparation
Read approved documents and records, organise the information and prepare drafts for review. Define what a useful result must include.
AI Agents and Employees
We build AI agents and AI employees for defined roles in your business. They review information, prepare work and use approved tools, with access and human approval requirements agreed with your team.
Give the Agent a Defined Role
An AI employee is a software agent configured for a business role. Its responsibilities and limits should be as clear as the work you ask it to handle.
Actions in your tools
Connect the agent to the systems needed for its role. Separate reading, drafting and making changes so permissions match the task.
Specialist delegation
Where the workflow benefits, delegate bounded tasks to specialist agents. Keep responsibility, shared state and verification in the surrounding system.
Example Workflow
A Business Request, Prepared for Approval
Follow a request through an agent’s work, with checks before any approved change.
Receive the task
A person submits a request within the agent’s defined role.
Gather information
The agent reads permitted sources and identifies missing material.
Prepare the action
A draft or proposed change is presented with supporting information.
Review and record
A person approves where required; the system records the action.
Decide What the Agent May Change
Tool access is scoped to the role. Customer contact, record changes and other consequential actions require explicit permission in the workflow. An AI employee does not remove the need for accountable people.
Systems and demonstrations
Explore the workflow, the controls and the decisions people retain.
Demonstration
Finance Agents
Six specialist AI agents prepare accounting work, apply written policies and route consequential changes for approval. Evaluation records distinguish a correct result from a missed exception or an escalation that should have happened.
Prepare the work
Agents organise the records and perform their defined review or matching task against the written policy.
Escalate and approve
Uncertain cases are escalated. Correcting journal entries remain drafts until a person approves them.
Evaluate changes
Expected outcomes distinguish correct answers, correct escalations and missed exceptions before a changed agent is released.
Controls and responsibility
Agents prepare; people post. Written policies and intake redaction are part of the workflow. Graded test cases expose errors and missed escalations rather than treating model confidence as proof.
The six roles
- Categorisation
- Proposes account coding and escalates uncertain charges.
- Reconciliation
- Matches book and bank records.
- Close and variance
- Flags differences against the materiality policy.
- Three-way match
- Checks purchase order, invoice and receipt.
- Client intake
- Checks required documents and redacts personal information.
- Action preparation
- Drafts correcting entries for approval.
Demonstration
Marketing Operations
A cross-brand operations view connects specialist marketing agents, a lead pipeline and an outreach console. Proposed outbound messages pass a safety check and a person before sending.
Bring the activity together
A shared view organises performance across brands and channels, with personal data masked in the lead pipeline.
Run specialist checks
Agents handle content routing, campaign go-live checks, campaign variance, sponsorship review and cross-channel reconciliation.
Review outreach
Every proposed outbound message passes the safety gate and human approval.
Controls and responsibility
Outbound messages do not send themselves. The lead pipeline masks personal information and the person reviewing outreach retains the sending decision.
Questions About the Service
What can an AI employee do?
Its role can include document review, draft preparation, record lookup and agreed actions in connected software. The actual scope depends on your process, tools and approval requirements.
Does it replace a member of staff?
We scope work rather than promise a headcount replacement. Your team remains responsible for customer relationships, important decisions and the work that needs professional judgment.
How do you check that the agent works?
We agree acceptance criteria, test representative requests and exceptions, and review tool access and approval behaviour before release. The evaluation reflects the task being built.
How is an AI harness different from a model or agent framework?
The model generates outputs. An agent framework provides building blocks for an application. The harness is the surrounding runtime and controls for the actual workflow, including tool access, state and verification. A framework can help build it, but the application still needs to enforce its own boundaries.
Tell Us What You Want to Build
Bring a workflow, a requirement or a problem your team wants to solve.