How to build a modern small-business tech stack
Email, invoicing, Excel, WhatsApp: your business already has a tech stack, it just grew by itself. What a stack is, why a few well-connected tools beat all-in-one, and where to start.
Blog
Reflections on data-driven marketing, AI orchestration, and process automation.
Back to ChatGPT. The method I wrote in Claude can come with me
ChatGPT and Codex now share a plugin directory, and the format those plugins store instructions in is the one Claude reads. What that changes for marketing work, with a demonstration still to be run.
Claude Code writes the code. The map of the environment stays with you
A one-line change left this site showing an old copy of itself for hours, and the code was correct. What was missing was a map of the environment, and you can learn to ask the AI for it.
What is Claude, and which model should you pick?
Updated
Two new models in one week, and Claude's menu has changed again. Version numbers come and go; the four tiers stay. A beginner's guide to picking the model by the task.
Where does a small business start with AI? With the people who do the work
The knowledge a small business needs to use AI already sits with its own people. Discover, plan, execute, update structures, in that order.
Why does the model need this data?
In an AI project that touches sensitive data, the question that shapes the architecture is why the model needs each piece of it. Privacy boundaries, least privilege and pseudonymisation versus anonymisation, with three real cases.
The source of truth: your company in a folder of markdown files
AI is only useful when it knows where the company's truth lives. Not a vector database, not a new standard: a folder of text files that someone has the authority to keep true.
Query fan-out: every question you ask an AI engine is dozens of searches
Updated
Every question you ask an AI search engine branches into dozens of parallel sub-searches. It is called query fan-out, and it changes the unit you optimise content for: no longer the keyword, the whole topic.
How to build an AI Council in Claude: 5 personas for better decisions
Updated
A practical prompt engineering guide to personas, multi-perspective prompting and synthesis in Claude, without turning any of it into a multi-agent system.
Agentic commerce: the storefront closed, and the real race has only just begun
Updated
The most hyped AI shopping feature of the year was switched off after five months. Understanding why tells you more about the future of commerce than any trillion-dollar forecast.
The standard that is just a folder (and why it should matter to you)
Updated
An organisation's knowledge lives spread across WhatsApp, spreadsheets and two people's heads. Google's new standard gives that problem a name, and the work it forces is worth more than the format.
Portugal's best customer experience belongs to the tax authority
Updated
The tax authority gets every invoice by the fifth of the month, machine-readable. The paying customer gets a PDF in a thread. That inversion is the diagnosis of Portuguese digitalisation, and it says what to fix before AI.
The AI Act: marking AI content is not your job
The transparency duties came into application on 2 August. The one everybody is bracing for sits with whoever supplies the model, not with the small firm using it.
The agent protocol wars are over. The map they left behind lies by symmetry.
MCP, A2A, AP2, x402: within eighteen months, the agent protocols were all donated to neutral foundations. The layer diagram that remains looks tidy, but it misleads on the essential point: the layers do not weigh the same.
I gave an agent the keys to my blog. These are the rules.
This article was uploaded and published by an agent, through an API and an MCP server built for this site. The rules that make it safe matter more than the code.