01 · AI integrations
We help teams move from AI experiments to a few agent workflows that are useful, measurable and safe to run.
Why now
Agents are now a production pattern, not a demo. The open standards are settled enough to build on, and the EU rules are in force. The hard part is the boring part: access, evaluation and ownership.
Most organisations already have access to capable models. The harder work is to decide where an agent belongs, what it may touch, who owns it and how you know it works.
We build on open standards such as the Model Context Protocol, so your tools and data stay portable between model vendors. Every workflow ships with evaluations, tracing and a named human owner.
Field notes
The rules, standards and platforms we design around today. Each note links to its source.
The new Model Context Protocol spec removes the session handshake. MCP servers become simpler to host and scale behind ordinary load balancers.
Our advice: Build new MCP servers on the current spec, and plan a short upgrade for servers that still depend on sessions.
Under Article 50 of the EU AI Act, people must know when they talk to an AI system. The digital omnibus moves most high risk duties to December 2027.
Our advice: Label AI chat and generated content now. Use the extra time on high risk systems to set up documentation, not to wait.
Nkom is the coordinating AI authority. The government has the AI Act simplifications on public consultation until 11 December 2026.
Our advice: Treat the EU dates as your dates. If AI is core to your business, answer the consultation.
The A2A protocol reached version 1.0 and has support from more than 150 organisations. Agents from different vendors can now hand work to each other.
Our advice: Use A2A when agents cross vendor or team lines. Inside one system, plain tool calls are simpler and easier to secure.
Related work
An AI agent workflow that handles routine support cases end to end, with human approval for anything that touches money.
Positioning, launch content and an accessible docs site for a European software product entering Norway, Sweden and Denmark.
Ways to work together
Most engagements begin with a short, focused piece of work.
A short assessment of where agents create real value, where they add risk and where a plain script is enough.
We build one agent workflow end to end, with MCP connectors, evaluations and agreed success measures.
We roll out, set permissions, keep evaluations running and prepare the documentation the AI Act asks for.
Questions
Yes. We work with the cloud, models and SaaS tools you already use. Open standards such as MCP let us connect them without lock in.
Only if a chatbot is the right answer. We start from the workflow and the decision behind it, then choose the interface.
The Act comes into Norwegian law through the EEA. Nkom is the coordinating authority, and the national law is in preparation. We design for the EU rules now, so you are ready when they apply here.
Every agent gets scoped permissions, an approval step for actions that matter, evaluations that run on each change and a named owner.