Insights
Notes from building AI systems
Research notes and working patterns from building agents, MCP servers, retrieval systems and AI-native products. Written by the people who build them, with the numbers behind each claim.
October 8, 2026 · 4 min read
Your website has a new visitor: AI agents
Forty-four tools already put an MCP server in front of a website. Who runs that server decides what an agent can see and do with your site, and who is in control. A summary of our founder's October 2026 survey, with what it means for a business.
October 8, 2026 · 4 min read
Should your product have an MCP server? A decision guide
Five signals that say build it now, three that say wait, and what a good first server looks like. Written for SaaS and platform teams deciding whether to let ChatGPT, Claude and other agents into their product.
October 8, 2026 · 4 min read
Keeping AI agents honest: the server-side grounding pattern
An agent that puts real items, prices or actions in front of a user cannot be allowed to invent any of them. The pattern we built for MayAI makes that impossible by construction rather than by prompt. Here is how it works and where it applies.
October 8, 2026 · 4 min read
What each AI host requires of an MCP server
A server that follows the MCP specification can still be rejected by the Claude directory, turned away from the ChatGPT app store, or have its tool names silently cut short by Gemini CLI. The rules of eight hosts, where they conflict, and what that means for how we build.
October 8, 2026 · 3 min read
Testing an MCP server: what exists and what is still missing
A server used by a language model needs more than a tool called by hand. Our founder reviewed about eighty tools that inspect, evaluate, audit, secure or load test an MCP server and compared thirty-one of them. Every capability exists somewhere; no open, local tool combines them.
October 8, 2026 · 3 min read
RAG for code and other structured material: why token-count chunking fails
Most retrieval systems split documents by token count, embed everything and hope. On code, contracts and catalogues that returns plausible fragments and misses the answer. What we built for DevGuardian AI instead, and how the same design applies beyond code.
October 8, 2026 · 4 min read
Multi-agent systems in .NET: what we learned building two of them
DevGuardian AI runs three role-based review agents; MayAI runs one agent with six tools. Both are built on Microsoft Agent Framework in .NET. Seven lessons about when to add agents, how to design tools, and what has to be in place before the first demo.
October 8, 2026 · 3 min read
From public data to market intelligence: how Valco AI was built
Dubai publishes its property market as open data, updated daily. It is public, and it is not ready to use. The pipeline, the data model, the forecasting models and the conversational layer we built for Valco Properties, and the order we built them in.
October 8, 2026 · 4 min read
AI readiness for Canadian businesses: a checklist of what has to be true first
Most AI projects fail on the data that was not ready, the workflow nobody mapped, or the guardrail nobody specified until the demo went wrong. The questions we ask in an AI readiness assessment, written as a checklist you can run yourself.
October 8, 2026 · 4 min read
An AI-first CMS: AI operates, people govern
Every content management system was designed for a person at a form, with AI added later as a button. What would we build if we started the other way round: a CMS for an agent to operate, with people in charge of what it may do? Our founder's design, and why we prototype it in FluentCMS.
October 8, 2026 · 4 min read
Documentation you can audit: observed, documented, inferred
AI-generated documentation reads beautifully, and nobody can tell which sentences came from the code and which the model dreamed up. Our founder compared seventeen tools that document a codebase, found the gap, and designed what he would build instead: facts with evidence and one of three labels.
Working on one of these problems?
The patterns in these notes are the ones we use on client work. Tell us what you are building.
Book an AI consultation