Atlas InnovationsSoftware for lab & analytics

Service · AI development

AI-Native Development Organization

AI agents accelerate your development — or produce undocumented sprawl. The difference is the system around them.

Typical starting point

  • Individual developers use AI tools, but nobody knows what ended up in the product.
  • In regulated projects, AI adoption fails on one question: “Who decided this, and where is the evidence?”
  • Pilot projects remain a playground because governance, documentation, and testing don't grow with them.

What you get

  • A working model for AI agents under governance: defined guardrails, documented decisions, anchored test runs — every change stays attributable to a trigger and a piece of evidence.
  • A toolchain from lived practice: we build our own project and evidence system and our agent-session control with these very methods, every day — you buy experience, not slides.
  • A fit for regulated environments: the model is built for audit situations and withstands the question of origin and justification for every change.

How we proceed

  1. Ways-of-working workshop. One day with your team: where AI agents can carry weight in your development process — and where they can't (yet). With a live look at our own working mode.
  2. Pilot in a real project. One well-bounded initiative runs in the new working model — with governance and evidence from day one.
  3. Anchoring. The pilot becomes the standard: guardrails, roles, and tools transfer into your organization; we stay until it runs without us.

One workshop day, in concrete terms

Six hours, on site or remote. The agenda is tailored to your team — this is how the day can run:

  1. 10:00–10:30

    Who is at the table, what you expect from the day — and a first snapshot: where AI tools already live in your development today, officially or unofficially.

  2. 10:30–12:00

    We map your development process from the trigger of a change to its evidence. Together we mark where context gets lost today and where an auditor would ask first.

  3. 12:00–12:45
    Lunch break
  4. 12:45–14:15

    Evidence-led AI development on a real project, not a slide deck: documented trigger, guardrails, an agent run under supervision, anchored test evidence — including a look into our tools Atlas and Tower.

  5. 14:15–14:30
    Coffee break
  6. 14:30–15:30

    Where should a pilot start in your house? We assess candidates from your portfolio, sketch the first guardrails, and define what success would look like after four weeks.

  7. 15:30–16:00

    The day's results at a glance, open points, feedback round. Within a few days you receive the written results note with a concrete pilot proposal.

10:00–10:30

Arrival & introductions

The start is deliberately short and slide-free: a round of introductions with role picture — who develops, who owns quality and compliance, who decides — and your expectations for the day, captured visibly.

How we work

  • Three baseline questions to the room: Which AI tools are in use today — officially or unofficially? Where does their output end up? Who knows about it?
  • The answers stay on the board as a first picture of opportunities and risks, and we return to them throughout the day.

What you take away

  • An honest, shared picture of the status quo — often the first occasion where “unofficial” AI use comes onto the table openly.

No preparation is required — on purpose: we want to see the everyday reality, not the rehearsed version of it.

10:30–12:00 · Session 1

Where you stand

We draw your development process as a map on the whiteboard: trigger → requirement → implementation → test → release → evidence. Then we take one or two real, recently shipped changes and trace them backwards: Why does this change exist? Who decided it? Where is the proof?

How we work

  • We mark directly on the map: where context gets lost, where media breaks occur, where an auditor would ask first.
  • Ideal cast: development plus quality/regulatory, up to about ten people — the friction between the two perspectives is intended and productive.

What you take away

  • The annotated process map — photographed and included in the written results note.
  • A first jointly prioritized list of the places where evidence breaks down today.

12:45–14:15 · Session 2

Live demonstration

The core of the day, without a single slide deck: a real agent run on a genuine piece of work from our own project routine — start to finish.

What you see live

  • The documented trigger: the task originates from a linked requirement, not from an empty prompt.
  • Guardrails in action: binding rules the agent works against — including the case where one is violated: the run is blocked and the violation logged.
  • The agent run under supervision in our control center: progress, budget, intermediate results.
  • The completion with anchored evidence: test run and result are linked to requirement and change — still queryable weeks later.

What you take away

  • A concrete, reliable picture of what AI governance looks like in daily work — a benchmark for every tool decision you make afterwards.

Interrupting is explicitly encouraged: every step is explored until the room has questioned it.

14:30–15:30 · Session 3

Transfer to your organization

Now it gets concrete for you: from your portfolio we collect well-bounded initiatives that qualify as a pilot — and assess them together.

How we work

  • Candidates are scored on four criteria: risk, evidence obligations, team maturity, expected benefit.
  • A first guardrail sketch for your organization: what may an agent do in your house — and what must it never do?
  • Success criteria measurable after four weeks: for instance the share of documented triggers, test anchoring of every change, or lead time from trigger to evidence.

What you take away

  • One or two prioritized pilot candidates with a guardrail sketch and measurable success criteria — the basis of the pilot proposal in the results note.

15:30–16:00

Wrap-up & feedback

The day ends with commitment instead of a sales slide: we summarize the results of all three sessions, note open questions, and ask for your feedback — including what was missing.

Afterwards, at no extra cost

  • Within a few days you receive the written results note: the annotated process map, the assessed pilot candidates, the guardrail sketch, and a concrete proposal for how the pilot starts.
  • The note is written to work without us — whether you run the pilot with us is your decision afterwards.

Entry product

Ways-of-Working Workshop

One day, fixed price, on site or remote: an assessment of where you stand, a live demonstration of evidence-led AI development, and a concrete proposal for where a pilot in your organization should start.

Fixed price €2,900 plus VAT · results note with pilot proposal

Request workshop

Related services

Instrument Software & Control · Compliance & Security by Design · Lab IT & Integration