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
An MCP server for a website lets AI agents search the site, ask it questions and act on it for a person, and who runs that server decides what agents see. A summary of founder Amir Pournasserian's October 2026 survey of 44 tools, and what it means for a business.
October 8, 2026 · 4 min read
Should your product have an MCP server? A decision guide
An MCP server lets AI agents in ChatGPT, Claude and other hosts use your product through a small set of tools. This decision guide for SaaS and platform teams gives five signals to build one now, three to wait, and what a good first server looks like.
October 8, 2026 · 4 min read
Keeping AI agents honest: the server-side grounding pattern
Server-side grounding is a pattern in which an AI agent returns only identifiers and the server builds everything the user sees from real tool results, so the agent cannot show an invented item. How we built it for MayAI, the six rules around it, and where it applies.
October 8, 2026 · 4 min read
What each AI host requires of an MCP server
Each AI host adds its own rules to the MCP specification, so a compliant server can still be rejected by the Claude directory or the ChatGPT plugin directory (shared with Codex). The rules of eight hosts as of 1 October 2026, where they conflict, and how we build to them.
October 8, 2026 · 4 min read
Testing an MCP server: what exists and what is still missing
Testing an MCP server covers six jobs: inspection, evaluations of tool choice, design audits, host-rule checks, security scans and load tests. Founder Amir Pournasserian reviewed about eighty tools and compared thirty-one on 1 October 2026: every capability exists somewhere, but 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-augmented generation (RAG) systems split documents by token count, which cuts code, contracts and catalogues into fragments that look right and miss the answer. What we built for DevGuardian AI instead, from code-aware chunks to a dependency graph and incremental refresh, and how it applies beyond code.
October 8, 2026 · 4 min read
Multi-agent systems in .NET: what we learned building two of them
A multi-agent system is one in which several AI agents, each with its own role and tools, work on the same task. Seven lessons from building two agent systems on Microsoft Agent Framework in .NET: DevGuardian AI, with three review agents, and MayAI, with one agent and six tools.
October 8, 2026 · 3 min read
From public data to market intelligence: how Valco AI was built
Valco AI is the platform we built for Valco Properties to turn Dubai's daily public property records into market intelligence and a conversational assistant for its staff. The pipeline, the data model, the forecasting models and the conversational layer, 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
An AI readiness assessment checks whether the workflow, data, privacy rules, guardrails, platform and people around a process are ready before anyone builds AI. These are the questions we ask in ours, written as a checklist you can run yourself, with a way to score the answers.
October 8, 2026 · 4 min read
An AI-first CMS: AI operates, people govern
An AI-first CMS is a content management system built for an AI agent to operate, with people deciding what it may publish. Founder Amir Pournasserian's October 2026 design: tools over MCP, rules that become checks, facts stored once and a review desk, and why we plan to prototype it in FluentCMS.
October 8, 2026 · 4 min read
Documentation you can audit: observed, documented, inferred
Auditable documentation labels every statement with where it came from: observed in the code, documented by a person or inferred by a model. Founder Amir Pournasserian compared seventeen AI documentation tools on 1 October 2026, found that none shows this, and designed a fact base that does.
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