What we do
Training and AI policy, integration and agents, and support that keeps it working — for every stage of your business.
Tier 1
Training & AI policy
Not another ChatGPT party-tricks seminar. We teach your team to use AI on their actual work — email, documents, operations — and we write the policy that says who may use what, on which data. The policy makes the training stick.
- — Claude á Íslandi: Iceland's first dedicated Claude seminar series. 11 public seminars sold out.
- — Private in-house workshops on your own tools, processes and files. We come to you.
- — AI policy and access matrix: permitted use with a defined gate, green/yellow/red data classification, a named policy owner, and a monthly usage audit.
- — Access is granted on proven competence, not job title.
- — A certificate documenting each participant's training — it documents the training, it does not certify compliance.
Tier 2
Integration & agents
Your filing is fine for people. It is not built for AI — we bridge that, then hand the repetitive work to agents that reach into the systems you already run. End to end through a whole process, not a demo of one step. No rip-and-replace, no migration, no eighteen-month roadmap.
Observe
We see the systems ourselves before proposing anything — in the room, or over a screen share. Anything only described and never seen is marked unverified until we have seen it.
Baseline
Which task, how often, who does it and how long it takes — with a named source and a date. We ask where a number proving it worked would come from: your time tracking, your invoices, your version control. A task we cannot measure is not a pilot, it is a risk.
Structure
We build the knowledge layer: what you know and how you decide, in plain text your own people can read and edit. We never copy your live systems into it — a copied customer table is a stale shadow of the real one by Tuesday. The agents read the real thing instead.
Build
Agents that pick the work up where it arrives and carry it to where it is finished, reading from your systems live. A human gate on anything irreversible or expensive to get wrong, with a named reviewer — a step with no named reviewer is an automated step.
Re-measure
The same tasks, the same source, before and after. If no baseline was captured we say so and claim nothing.
Ongoing
Ongoing partnership
Adoption falls off sharply 30 to 60 days after delivery unless somebody owns it. Every engagement ends with a named owner on your side and the documentation to run it without us. If you would rather we stayed on, that is available too.
- — Skills and evaluations kept current as the work changes
- — Model upgrades migrated for you — a new frontier model should be an improvement, not a rebuild
- — A monthly review of where the freed capacity actually went
- — New workflows as you find them
- — Available as a monthly subscription. Every engagement works without it.