Support and operations agents
Read a request, look up the customer’s records across your tools, act within set limits and pass anything unusual to a person.
AI agent development for work that can’t be scripted
We build agents that use your tools to reach a goal you set, and we put the tests, limits and approval steps in place before they touch real customers or real money.
What is an AI agent?
An AI agent is a language model running in a loop: it plans a step, calls a tool such as a search, a database query or an API, looks at the result and decides what to do next, until the goal is met or it hits a limit. Unlike a workflow, the steps change from case to case. That makes agents more flexible, but also more expensive and harder to audit, so we only recommend one when a workflow can’t do the job.
- First release
- 6–12 weeks
- Discovery
- 1–2 weeks, fixed price
- Runs in
- Your cloud account
- Models
- Chosen per task
Agents we build
Research and analysis agents
Gather information from internal and public sources and write a brief with every claim linked to its source.
Agents that operate internal tools
Carry out multi-step tasks in your admin panels and APIs that would otherwise take someone an afternoon of clicking.
Casework agents
Work through a case file, ask for what’s missing and prepare a decision for a person to approve.
Assistants inside your product
An assistant your customers use, limited to the actions and data their own account allows.
Multi-agent systems
A coordinator that splits a job between specialist agents and checks their work before anything is returned.
How an agent runs
Every agent we build runs inside the same five-step loop. The approval step is what turns a demo into something you can leave running.
Goal
A request or an event sets the goal and the tools the agent may use.
Plan
The model picks the next step and the tool to call.
Act
The tool runs with the agent’s permissions, never more.
Check
Results are checked against rules, and step and spending limits are enforced.
Approve
Risky actions wait for a person, with the full trace attached.
Included in every build
- Evaluation setReal cases from your data, with the right answer, that every change is tested against before release.
- Step-by-step logsEvery plan, tool call, result and cost recorded, so you can see why the agent did what it did.
- LimitsCaps on steps, spend and time per run, and a list of actions the agent can never take.
- Approval stepAnything with financial, legal or customer impact waits for a person until the data says it doesn’t need to.
- Runbook and handoverWritten for the engineer who maintains it after us.
AI agent development FAQ
Do we need an AI agent or an AI workflow?
If you can draw the process as a flowchart, you need a workflow: it’s cheaper, faster and easier to audit. You need an agent when each case needs different steps across several tools. We’ll tell you which during discovery, and most of the time it’s a workflow.
How do you stop an agent doing something it shouldn’t?
It only gets the tools and permissions the task needs. Limits on steps, spend and time end runaway runs, and anything risky waits for a person to approve it.
How do you test an AI agent?
Against an evaluation set built from your real cases, before every release. We check whether it reached the right result, how many steps and how much money it took, and what it did when it couldn’t finish.
How much does an agent cost to run?
It depends on how many steps each run takes and which model it uses. We estimate a per-run cost during discovery, set a hard limit per run and track spend live after launch.
Is our data used to train AI models?
No. We use model providers and settings that do not train on your data, and we can run on models hosted in your own cloud when required.