How long does it take to build an AI agent?

Two to four weeks from kickoff to handover for a scoped single-workflow agent, plus a one-to-two week feasibility review that includes the revFADP and FINMA scoping. Swiss engagements carry more up-front compliance mapping than EU ones, and that is real work rather than paperwork.

How much does AI agent development cost in Switzerland?

A scoped build is the usual entry point. Swiss projects carry additional compliance mapping, which shows in the band. What else moves the number is systems touched, data usability, cost of a wrong action, and who operates it afterwards. The pricing page publishes the ranges.

How do you handle revFADP and Swiss data residency?

With a written map, before any customer data moves: each processing step, the named provider, the region the data sits in, and the legal basis under revFADP and under GDPR where EU residents' data is involved. Data can stay in Switzerland or the EU depending on the capability. You keep the document.

Will the agent take actions autonomously?

In Swiss regulated work, generally no, and deliberately so. It drafts and a human approves. That is easier to make reliable, faster to deploy, and aligned with FINMA's expectations around reversibility. It also captures most of the value, since the expensive part of these workflows is the assembly rather than the final decision.

What does FINMA mean for the architecture?

It determines what can be delegated, what stays auditable, and what must remain reversible. Practically: every tool call logged with its inputs, role-scoped access, a defined rollback for anything touching a system of record, and a documented boundary between agent action and human approval. Retrofitting this means rebuilding, which is the most common reason Swiss pilots stall before production.

Does the EU AI Act apply to us?

Not directly, since Switzerland is outside the EU and the EEA, but it does if you place an AI system on the EU market, which covers much of Zurich's pharma and financial sector. Most Swiss clients are in scope for revFADP and the EU AI Act at once. Pharma workflows touching clinical or safety data can also classify a tier higher than expected, so we resolve that at scoping.

Do you work in German?

No. Our working language is English and documentation is delivered in English. Most Zurich technical and financial teams work comfortably in English, but if you need German-language facilitation and deliverables, a local agency will serve you better and we will say so on the first call.

Why hire a Stockholm team instead of a Zurich one?

For some projects you should hire locally, particularly if you need people in the building several days a week. Where we fit is when you want the regulatory literacy without the local overhead: identical time zone, Nordic senior engineering rates rather than Swiss ones, and travel for the sessions that genuinely need a room.

What do we own at the end?

Code and configuration in your own repository, on infrastructure you control, plus the evaluation set, the audit trail and a runbook. There is no platform of ours you must keep paying for to keep your workflow running.

Built for Switzerland, from Stockholm

AI agent development in Zurich

We build AI agents for Swiss companies where a wrong action is expensive: draft-and-approve by design, every tool call logged, and the data residency answer written down before anything moves.

Swiss buyers have met the vendors who demo well and cannot deploy. In regulated Zurich work the constraint is rarely model capability, it is auditability and reversibility, so that is where our engineering effort goes.

Talk about a workflow See pricing

Draft-and-approve is not a limitation here, it is the architecture

In most Zurich engagements the client tells us early that nothing may reach a customer, a regulator or a ledger without a human signing it off. Vendors sometimes treat that as a constraint to be negotiated down. We treat it as a gift.

A draft-and-approve agent is dramatically easier to make genuinely reliable than an autonomous one. The failure mode is a human rejecting a bad draft, which costs seconds, rather than a wrong action propagating into a system of record. And it still captures most of the value, because in these workflows the expensive part is the assembly and cross-referencing, not the final click.

It also means the system can go live sooner. There is no long tail of guardrail engineering required to make autonomous action safe, because the autonomy is not there. If accuracy on your evaluation set later justifies loosening a specific gate, that is a decision made with evidence rather than at launch.

What FINMA actually changes about the engineering

FINMA's outsourcing expectations are frequently treated as a legal annex. They are architecture constraints. They determine what may be delegated to a third party, what has to remain auditable, and what must stay reversible, and each of those is a decision that has to be made before the first line of code rather than after.

Concretely: every tool call the agent makes is logged with its inputs and the reasoning behind it, so an incident is investigable. Access is role-scoped. Any action affecting a system of record has a defined rollback. And the boundary between what the agent may do directly and what requires human approval is documented as part of the design, not discovered during a review.

Retrofitting this into a system built without it means rebuilding it. That is the single most common reason we see Swiss AI pilots fail to reach production, and it is entirely avoidable.

The data residency map, before anything moves

Every Swiss engagement starts with a written map: each processing step, the named provider handling it, the region the data sits in, and the legal basis under revFADP and, where EU residents' data is involved, GDPR. That document exists before any customer data moves, and you keep it.

Where a capability is only available from a provider that cannot offer Swiss or EU processing, we put the trade-off in front of you as an explicit scoping decision. Sometimes the answer is a smaller model in an acceptable region and a slightly worse result. Sometimes it is that the workflow is not yet a good candidate. Both beat discovering the problem in an audit.

Because Switzerland sits outside the EU and the EEA while most Zurich companies still process EU residents' data, the common situation is being in scope for revFADP and GDPR simultaneously. That dual scope is the actual complexity, and a vendor who has not noticed it is a risk.

Use cases

What we are asked to build in Zurich

Real workflow shapes from this market, with the point where a human stays in the loop stated for each one.

Private banking and wealth management

Client documentation drafting

Portfolio commentary, client reporting and research synthesis, drafted by the agent and signed off by a relationship manager. Nothing reaches a client without a human approving it, which is both a regulatory constraint and the design we would choose anyway.

Insurance and reinsurance

Claims file assembly and triage

The agent gathers the file, cross-references policy terms, flags the discrepancies and ranks the queue. The decision stays with an underwriter, who now receives a complete file instead of assembling one.

Reinsurance

Submission review support

High-volume submission intake where most cases follow the pattern and the exceptions genuinely need judgement. The agent handles the assembly and the pattern-matching; the exceptions escalate with context attached.

Pharma and med-tech

Regulated documentation workflows

Documentation review against internal standards and applicable requirements. The EU AI Act classification question is real here, because a workflow touching clinical or safety data can sit a tier higher than assumed, so we resolve it at scoping.

Commodities trading

Internal research synthesis

Pulling together market and counterparty information from internal and licensed sources into a briefing, with every claim traceable back to its source document rather than asserted by the model.

How we build

A Zurich agent build, week by week

Two to four weeks from kickoff to handover. The order matters more than the tooling: measure first, prototype on real data second, close the loop third.

01
Watch the workflow being done
2 to 4 days
We sit with the people who run the process today, and we measure it: how many cases, how long each takes, where they stall, and which exceptions actually recur. Most projects that fail do so because this step was skipped and the brief described the process as management believes it works rather than as it runs.
You end up with: A measured baseline you can hold the finished system against, and a written list of the exceptions nobody had documented.
02
Build the thin version on your real data
3 to 5 days
Not a demo on a curated sample. Your records, including the ones with missing fields and inconsistent formatting. This is where you discover that a third of the source rows lack something the workflow depends on, and it is much better to discover that in week one than in month three.
You end up with: A narrow tool running on production-shaped data, and an honest assessment of whether the rest is worth building.
03
Close the agent loop
1 to 2 weeks
Now the agent plans across steps, calls the tools it needs, and handles the cases the thin version could not. Human review gates go on every action that is expensive to undo. We build the evaluation set from your real cases at the same time, including the failures, because an agent with no evaluation set is an agent nobody can safely change later.
You end up with: A working agent, an evaluation suite built from your own cases, and audit logging on every tool call.
04
Hand it over properly
2 to 4 days
A runbook, a training session with the people who will operate it, and a documented path for what to do when it breaks. We do not make handover deliberately incomplete to keep you dependent on us. If you want us to keep operating it, that is a separate retainer you choose, not a trap you fall into.
You end up with: Runbook, handover session, and the code and configuration in your own repository.

How we build

Positions we hold on every build

These are decisions we make the same way every time, because each one is a reason agent projects fail when it goes the other way.

Bounded autonomy by default

An agent starts read-only and draft-first. Actions that are expensive or awkward to reverse stay behind a human approval gate, permanently if that is the right answer. Full autonomy is something a system earns by demonstrating accuracy on your evaluation set, not a launch feature.

An evaluation set from your real cases

Built from your actual records, including the ones the agent gets wrong. Without it, nobody can safely change a prompt or swap a model six months later, which is how working systems quietly rot.

Every tool call logged

Each action, its inputs, and the decision behind it are recorded. This is what makes an incident investigable, and under FINMA, NIS2 or medical-device rules it is a requirement rather than a nicety.

Retrieval quality over model size

Most disappointing agents are not under-powered, they are under-informed. Getting the right context in front of the model reliably matters more than which model it is, and it is where the engineering effort usually belongs.

Your repository, your infrastructure

Code and configuration live in your repository and run on infrastructure you control. There is no VLIRTZ platform you have to keep paying for to keep your own workflow running.

One workflow before three

We decline company-wide assistant scopes. A single workflow, measured and shipped, tells you more about whether this approach works for you than any roadmap, and it is recoverable if the answer is no.

Pricing

What an AI agent costs in Zurich

We quote in CHF for this market. Where you land depends on how many systems the agent touches, how usable your data already is, and how expensive a wrong action would be. Ask and you get a range on the first call, not the third.

Agent feasibility review

On request

Workflow measurement, data assessment, and a revFADP and FINMA scoping note. Includes a scoped build proposal.

Timeline: 1 to 2 weeks

Scoped agent build

On request

One workflow end to end, human approval gates, full audit trail, evaluation set, documented rollback, runbook and handover.

Timeline: 2 to 4 weeks

Additional workflow

On request

A second or third agent reusing the orchestration, retrieval and audit infrastructure from the first.

Timeline: 2 to 3 weeks each

Sustain retainer

On request

Monitoring, drift checks, model and prompt updates, and a defined response time on failures.

Timeline: Rolling

Full pricing breakdown Get a range for your workflow

Straight answers

What an AI agent will not do for you

Every one of these has ended a project somewhere. We would rather raise them before you sign than explain them in month two.

It will not fix a process nobody has agreed on: If two departments genuinely disagree about how a case should be handled, an agent forces that disagreement into the open rather than resolving it. That is useful, but it is a management outcome, not a technical one.
It will not rescue unusable source data: Retrieval over clean, structured records is straightforward. Retrieval over scanned documents, three competing sources of truth, and a field that has been wrong since a migration is where budgets disappear. Sometimes the honest recommendation is a data project first.
It will not be right every time: The question is never whether it makes mistakes, it is whether the mistakes are caught before they cost anything. That is what the review gates and the evaluation set are for, and it is why we measure the baseline first.
It will not maintain itself: Models change and your source systems change. An unmaintained agent degrades quietly rather than failing loudly, which is worse. Budget for maintenance or plan to retire it.

More about working with us in Zurich

This page covers how we build agents. The Zurich market page covers the rest: the regulators that shape a project there, the sectors we see most, and when we are the wrong partner.

AI Software Agency for Zurich Companies

FAQ