How we build AI agents

This is a 30–45 minute conversation. We ask about your current workflow in detail: what triggers it, what data moves through it, where humans are involved and why, where things break down.

The output isn't a proposal — it's shared understanding. We'll know whether your process is a good candidate for automation. You'll have a realistic sense of what building it involves.

If we think something isn't automatable, or isn't worth it at your current scale, we'll say so here. Better to know early.

After the discovery call, we write a technical scope document. This specifies what the agent will do, what it won't do, which systems it integrates with, how data flows, and how edge cases are handled.

We attach a fixed price and a delivery estimate. These don't change once signed — unless you add scope, in which case we agree a change order first.

You review it, ask questions, and decide whether to proceed. No pressure on this timeline from our side.

We build the agent against the agreed spec. During this phase you'll see iterative demos — not necessarily polished, but functional enough to give meaningful feedback.

Testing is done against real scenarios from your business: normal cases, edge cases, failure modes. We want to know how the agent behaves when something unexpected comes through, not just when it goes right.

GDPR-relevant decisions — what data the agent processes, how long it's retained, who has access — get documented during this phase.

Deployment means the agent runs in your environment, connected to your actual systems, processing real data. We handle the deployment and stay available for the first week post-launch.

After that, ongoing support is available through our monthly retainer. The retainer covers monitoring, bug fixes, minor adjustments, and keeps the agent working as underlying systems or processes evolve.

You're not locked in — retainers are cancellable with 30 days notice. If the agent is stable and your processes aren't changing, you may not need one.