Insights

Notes from the wrong side of the data-residency wall.

Practical writing on sovereign AI, retrieval that holds up under audit, marketplace procurement and what actually stops enterprise AI reaching production. No thought leadership for its own sake.

Latest

Start here

Looking for the detail a security review needs? Read the white papers.

AI costResearch note

What an answer costs: how K-Lake reduces the tokens an assistant spends

Four retrieval approaches, the same files and the same questions. How many tokens each one put into the conversation, what that means for an AI budget, and the limits of the test.

12 October 2026 · 6 min read

Read the research note

"The assistant searched widely and read narrowly. 99.3% of what K-Lake found stayed on disk."

From the research note

Procurement

What a CISO actually asks in an AI security review

The questions that stall AI pilots at the security review, from how many copies of your documents will exist to whether you can switch it off, and how K-Lake answers each one.

12 October 2026 · 5 min read · Read

Marketplace

Using MACC to fund an AI deployment you already need

Committed Azure spend is budget that already exists and has an expiry date. How Marketplace procurement works, and the condition that decides whether it counts.

12 October 2026 · 4 min read · Read

Retrieval

Citations are the audit trail

Why grounded retrieval with source links is what makes an AI answer defensible to a regulator, and what to ask any vendor about it.

12 October 2026 · 4 min read · Read

Sovereign AI

Why enterprise AI stalls at one question, and what to do about it

Most stalled AI rollouts are boundary problems, not model problems. The three architectural answers to "where does our data go?"

14 August 2026 · 6 min read · Read

What we write about

Four themes, and nothing else

We write where we have something specific to say. If a topic does not sit under one of these, it does not belong here.

Sovereign and regulated AI

Data residency, air-gapped deployment, and building AI into environments that cannot send data out.

Retrieval that holds up

Grounding, permission-aware access, citations, and why keyword search alone keeps failing enterprise users.

Procurement and security review

What stalls enterprise AI deals, how to prepare for the questionnaire, and marketplace and MACC mechanics.

Regulation in practice

DORA, the EU AI Act, NHS data governance and what they mean operationally rather than theoretically.

Follow along

New pieces are published here first, then syndicated to LinkedIn. Subscribe by RSS, or follow the company page to get them in your feed.

Follow on LinkedIn RSS feed