Press
What we will say on the record.
We are a small firm with firm opinions about how AI actually gets into companies. Below: the stories we are useful for, what we will say, who says it, and the files you will need. We reply to press within one business day.
Angles
Four stories we are good for.
The pilot graveyard
Most AI pilots die at the CRM, not at the model.
The demo works. Then it meets permissions, a data model nobody documented, and a dispatcher who has done it the other way for eleven years. We spend most of our hours on that half. It is the half nobody puts in a keynote.
Who is doing the assessing
Every platform vendor now has an AI consulting arm. Guess what it recommends.
If an AI readiness assessment ends with a quote for more of the assessor's own platform, it was a sales call. Vendor neutrality is a product, and mid-market companies are starting to ask for it by name.
Evaluation is the deliverable
If you cannot measure it, you did not ship it.
An agent that is right 94 percent of the time is a liability until you know which six percent it gets wrong. A 200-case evaluation suite built with the people who do the job is worth more than any slide of screenshots.
The two-page plan
What a CFO should get from an AI audit.
Not a wish list. A ranked shortlist of opportunities with effort, risk, and a number next to each one, and at least one line that says 'do not build this yet'. We can walk through a redacted example.
On the record
Quotable, attributed to Martin Gottweis.
“The model is the cheapest part of the project. The expensive part is the thirty years of process it has to fit into.”
“About a third of the people who call us, we tell not to build anything yet. That is the part of the job nobody puts in the brochure.”
“Ship it into the tool people already have open. Nobody in operations wants another tab.”
“An AI feature with no evaluation set is a demo with a logo on it.”
“If the vendor cannot tell you what happens to your data in one sentence, the answer is 'something you would not like'.”
Use these as written, with attribution to Martin Gottweis, Founder & Principal Engineer, Derkins Labs. For anything else, ask; we will put it in writing so it can be quoted.
Comment
Topics we can speak to on deadline.
- AI agents and copilots inside CRMs, ERPs, and back-office tools
- Evaluating and guardrailing LLM systems before rollout
- Build versus buy, and when the honest answer is 'not yet'
- Vendor lock-in and vendor-neutral AI strategy
- Document pipelines: intake, extraction, classification, routing
- AI in logistics, financial services back office, healthcare operations, and manufacturing
- What an AI audit should cost and what it should produce
Background conversations are fine and we will say so up front. On-the-record comment comes by email so there is no ambiguity about wording.
We do not comment on named clients, on companies we have not worked with, or on vendors' financials. We are happy to explain how a category of system works, what tends to break, and what a buyer should ask.
Based in New York. Available on US and European time.
Facts
At a glance.
- Founded
- 2026
- Headquarters
- New York, NY
- Founder
- Martin Gottweis, previously Meta, Booking.com
- Size
- Small and senior, by design
- Typical timeline
- Audit in two weeks. Pilot in four to six. Integration in eight to twelve.
- Pricing
- Fixed-fee audits. Scoped, fixed-price builds. Advisory by retainer.
- Models we ship with
- Claude, GPT, Gemini, and open-weight models on private infrastructure
- Sectors
- Logistics, B2B software, financial services, healthcare ops, manufacturing, professional services
- Markets
- North America and Europe
Who is speaking
Founder.
Martin Gottweis
Founder & Principal Engineer · previously Meta, Booking.com
Before Derkins Labs, Martin spent five years at Meta in New York as a technical lead on AI personalization and integrity systems: the graph-RAG infrastructure behind messaging search for roughly 500 million users, knowledge injection for WhatsApp AI, and enforcement infrastructure running at over eight billion queries per second.
Before that he was a senior developer at Booking.com in Amsterdam, where he led the hotel page, the core revenue funnel, and shipped 0-to-1 products in its new-ventures division. He has written production software since 2007 and holds a master's degree in aerospace engineering from the University of Zilina.
Logo
The mark, and how to use it.
Write the name as Derkins Labs, two words, capital D and L. The mark is an opened box: a business, taken apart and put back together with AI inside. Keep clear space around it at least equal to its height. Do not recolor it outside the cobalt family, add effects, or set it on busy imagery.