RAG that respects access control
Permissions are the hard part of enterprise retrieval, and the part most designs defer. Four approaches, and why the index has to know who is asking.
Read insightinsights
Perspectives on building AI and systems that hold up in production.
Practical thinking on operational AI, architecture, integration, and platform engineering for organizations with real complexity and regulatory weight.
AllAI governanceAI implementationAI-assisted developmentData platformsData residencyDiscussionEngineering deliveryIntegrationKnowledge retrievalOperational AIProduct deliveryPublic sectorSovereign AISystems architectureWeb platform
AI Act deployer readiness: a working checklist
Twenty questions per system, grouped by what they imply for the architecture. Use it to find the gaps before someone else does.
Read insightWhat an authority must now be able to show
Three Danish public authorities, three levels of competence, the same failure: none of them can produce a record of what an AI system did with a citizen's data. That record is no longer optional.
Read insightAI in your editor, or AI in their platform
Two things get called the same name, and only one of them puts an AI system into the customer's estate. Which one you mean decides who has to govern it.
Read insightBugs fixed overnight, reviewed in the morning
We built an in-house service that works the backlog of small, deferred defects while the office is empty, using computing capacity the organisation has already paid for. What makes it usable is not the model. It is the narrow scope, the team's own tests, and the review queue waiting at 08:00. The more interesting property is that the loop from report to released fix can close — which makes autonomy a decision about which changes you trust, rather than an engineering leap.
Read insightOKF and RAG in the same system
The Open Knowledge Format describes a corpus. Retrieval finds things in one. What each contributes when they run together, what the pair does that neither does alone, and how to tell which of the three shapes your problem needs.
Read insightYour AI needs playback—not a blockchain
A trustworthy AI system should not merely retain an answer. It should let you return to that interaction and inspect the evidence, decisions, configuration, and controls behind it.
Read insightWhat the Open Knowledge Format does
A directory of markdown files with YAML frontmatter, published by Google Cloud as an open specification. What OKF requires, what its trust fields record, and which of your problems it leaves untouched.
Read insightForms an agent can fill: what WebMCP changes
An agent filling a form today reads the rendered page and infers. WebMCP lets the page declare the same capability as a named tool with a schema. Two attributes on the form, one per field, and no script. The demonstration below runs in this article.
Read insightWebMCP and the dual-mode website: what leadership actually decides
There is a widely held expectation that web pages end up as a prompt box. WebMCP is the more modest version of that future: the application stays visible, and a prompt is added as a second way to operate it. Two things have to exist for that to work. Only one of them is yours to build, and whether you build the second one as well is the decision about who controls the agent.
Read insightWebMCP security: the backend does not change, but intent does
A form completed through WebMCP should run the same authentication, authorization and business rules as one filled in by hand. The security question WebMCP actually raises is not whether the backend needs to change. It is whether the action a valid, authorized agent just took is the action the user meant.
Read insightWhat an AI gateway does, and when you need one
One internal interface between your applications and whichever model serves them. What belongs in that layer, and the point at which not having one starts to cost you.
Read insightThe AI code trust gap: adoption is settled, ownership is not
Around 90% of developers use AI daily, more distrust its accuracy than trust it, and its security pass rate has not moved in a year. Read together, the 2025-2026 evidence says the constraint has shifted from writing software to owning it, and that is a specification and accountability problem rather than a tooling one.
Read insightThe EU AI Act: what it asks of deployers
Most organizations buying or building on AI are deployers rather than providers. That distinction decides which obligations land on you, and most of them are architectural.
Read insightOn-premises vs hosted LLM: how to choose
The choice is usually settled by data residency and contracts, not by cost or model quality. Here is what each option costs you in practice.
Read insightWhat sovereign AI actually means
The term covers four separable properties, and vendors tend to sell the cheapest one. Which of them you actually need depends on what you are protecting against.
Read insightAI-assisted implementation with frontier models
Frontier models can accelerate implementation, but only when they are used inside a disciplined delivery method: clear architecture, review, testing, security, and production ownership.
Read insightHow to turn a business-built AI prototype into production software
A business-built prototype proves intent and interaction. It does not prove architecture, security, data integrity or operational readiness. Treat it as an executable specification: preserve intent by default, and preserve generated code only where evidence justifies it.
Read insightWhy AI pilots stall before production
A working demo is not a working system. The difference between AI that ships and AI that stalls is operational integration, not model quality.
Read insightSovereign AI in regulated EU enterprises
Data residency and modern AI are not mutually exclusive. A governed model gateway and on-premises deployment let regulated organizations use AI on their terms.
Read insight
Newsletter
Occasional notes on building systems that hold up
A short email when we publish something worth your time. Architecture, integration, and operational AI in regulated organizations. No cadence promises, no forwarding your address.