Custom AI agent development, built and run for you
We take one repeated, document-heavy process and build an agent that does it the way your best person would: reading, checking against your rules, drafting, and handing the decision to a named person. It is tested on your cases before anyone relies on it, and it keeps the records your auditors will ask for.
What the engagement covers
- One process with a named owner: intake and triage, case preparation, document review against a checklist, answers from your policies, reconciliation or exception handling.
- The knowledge sources the agent reads (your documents, policies and templates) and the systems it connects to, each connection built and tested in the project.
- Which actions wait for a person, the AI notice for customer-facing agents, and the cost limits.
- A pilot on live cases with your team checking the output, then production on your approval.
- Good fitsHigh-volume, rule-bound work with documents and a clear owner: intake and triage, case preparation, document review against a checklist, answers from your policies, reconciliation and exception handling.
- Poor fitsDecisions that should not be automated at all, processes with no written rules and no owner, or a one-off analysis. We say so in the scoping call, and we will tell you when an off-the-shelf tool is enough.
What you get, in writing
- A written brief and a risk classification for the use case, with the reasoning (AI Act Art. 6 and Annex III).
- A pass bar you sign before the build: the cases, the score it must reach, and the safety cases that must all pass.
- The agent, built on the Wise Agents platform: your knowledge sources, the tools it may use, and the actions that wait for a person.
- A test report on your cases, re-run before every new version.
- A documentation pack: what the agent does, its limits, its model route and region, its oversight design and its logs (Art. 11 and 13 as a guide).
- Go-live on your approval, then monitoring, cost limits and a stop control.
How long it takes
- Sign the pass bar 1 week
- Build in weekly rounds 2–5 weeks
- Test and document 1–2 weeks
- Approve, then a supervised pilot about 2 weeks
A pilot usually takes 6–10 weeks from the signed brief, after a scoping step of one to two weeks. A production agent, with integrations and the full documentation pack, usually takes 3–5 months.
What this service doesn’t do
- Decisions with legal or similar effects taken by the agent alone. A person decides those.
- Legal advice. Your counsel and DPO confirm the classification and the contracts.
- Changes to your own systems beyond the agent’s connection, such as a new API on a core system.
- Model usage, which is billed at the provider’s price without mark-up and capped by your limits.
- Hosting on your own infrastructure. It is possible, and priced separately after scoping.
- A certificate. No certificate exists for being “AI Act compliant”, and we don’t sell one.
How the rules apply here
Each control points to the article it answers. We read them on 9 November 2026; your counsel confirms how they apply to you.
Tested before anyone relies on it
AI Act Art. 15Art. 10Your cases, plus safety cases we write with you (wrong customer, missing document, a prompt that tries to override the rules). The release is blocked below the bar.
A person decides what matters
Art. 14GDPR Art. 22Payouts, refusals, anything with a legal or similar effect waits for a named person. The agent prepares; it does not decide alone.
Everything is recorded
Art. 12Art. 26(6)Each run keeps its input, the sources and tools used, the model, the cost and who approved what.
People know it is an AI
Art. 50Customer-facing agents say so at the first message.
Questions about this service
How long does a first agent take?
A pilot on one process usually takes six to ten weeks from the signed brief: the pass bar, the build and test rounds, then a supervised pilot. A larger process, or one with many integrations, takes longer, and we say so in the proposal.
Do you build on our cloud or yours?
By default on the Wise Agents platform, hosted in Nuremberg, Germany. If your policy requires your own environment, we discuss it in scoping; it changes the price and the timeline.
Which models do you use?
We choose per agent, by testing candidates on your cases. Each deployment names its model provider, route and region in your contract. We never switch the model without re-running your tests.
Who owns the agent?
You own your data, your cases and the agent’s configuration. The platform stays ours; the contract says what you can export and how. Export is manual today, on request.
What is the difference between a pilot and production?
A pilot proves one process on live cases with your team checking every output. Production adds the integrations, the full test set, the documentation pack and training for your reviewers, then the agent runs with monitoring and cost limits.
Bring one process. Leave with a risk class, a pass bar and a plan.
A 45-minute scoping call with the people who would build it. You get a first view of the AI Act risk class, the controls it needs, and whether an agent is the right tool at all.