Lessons from real deployments
Articles on low-code, AI agents and running a business, drawing on published research and case studies.
12 results
Article list
AI Agents
CodeKasten: give the agent a branch, keep the merge for people
CodeKasten lets a team build its own internal tools, and pushes the branch–commit–merge model of source control all the way down to the data. What makes it worth analysing is how it answers the question every company using agents faces: how far does an agent's write access go.
AI Agents
Tech brief, 12–18 Oct 2026: AI and the verification problem
No major Mendix or Power Platform release last week. But Siemens, Meta, Grab and the latest Vietnamese market figures all point the same way: the question companies ask is moving from “what can AI do” to “what data does it use, which process does it run in, and how is the result verified”.
AI Agents
OpenAI Dots and the rules of engagement for enterprise agents
OpenAI has announced dots — always-on agents with their own cloud computer, connected to more than four thousand apps. For a company, the part worth reading is not what the agent can do, but the rules setting out when it may act alone, when it must ask, and when it must hand work back to a person.
AI Agents
Claude Sonnet 5.5: the price held, the cost per task fell
Anthropic has released Claude Sonnet 5.5 at the same token price as Sonnet 5, but it spends far fewer tokens per task and generates output more than 30% faster. For a company costing out an AI feature, the number to watch is tokens per task, not the benchmark score.
AI Agents
Jev and System One models: typed decisions instead of text
A new lab has released a model that writes no prose at all, returning typed decisions with calibrated probabilities. The part that matters for enterprises is not speed or price, but software knowing when to ask a human.
AI Agents
Designing UI with AI agents: pen.dev, Figma and Penpot
AI agents can write UI code, but the design usually lives in a tool they cannot reach. Three approaches compared — pen.dev, Figma and Penpot — and how Mendix developers and BAs can apply them.
AI Agents
Weekly brief 7–13 Sep: AI moves from answering to acting
A quiet week for Mendix releases, but one requirement is getting clearer: AI that takes part in critical processes needs an identity, a defined scope of authority, trustworthy data and decisions that can be traced.
AI Agents
Agent tooling has a short life: don't put business logic there
OpenAI shipped Agent Builder in November 2025, announced its retirement in June 2026 and shuts it down at the end of November. Custom GPTs are being replaced too. The lesson is about where business logic lives.
AI Agents
MCP becomes a vendor-neutral standard for the connection layer
The MCP protocol moved to a Linux Foundation body, and the July 2026 specification drops session state so it runs on ordinary HTTP infrastructure. The layer between AI and enterprise systems is settling down.