Method · Understand to decide

Guides

Clear guides on AI and the tools I use: what they are, what they do, what they change. Written for merchants and business owners who want to get up to speed fast — but readable at any level. Three reading depths on every page: in 30 seconds, the visual body, and going deeper.

★ · Most recent

Latest guides

Case study · September 4, 2026 Agent Security: a multi-GitHub security cockpit Four open-source scanners, sandboxed access, a risk memory, and multiple LLMs to review uncertain cases — without giving them the final word. Read the case study → Case study · July 13, 2026 A competitive pricing dashboard, built in-house The competitor data (Wiser) was already in prod, delivered by the integrator. The management layer — quoted at several weeks — built internally with Claude Code in a matter of days. The most clear-cut build vs. buy of the year. Read the case study → Case study · June 17, 2026 An agent that processes supplier invoices It reads emails, extracts each invoice, matches it to the right Odoo order, and pushes it into Teams. ~€2M in historical drift identified — and a workflow that can't drift anymore. Read the case study →

01 · Understand

The basics

What we're actually talking about, jargon-free: AI, LLMs, Claude, and what it all changes.

Guide 01 · May 31, 2026 What is AI, and what is an "LLM"? Generative AI and LLMs without the jargon: what they do, who the major players are (OpenAI, Anthropic, Google, Mistral…), and what they can't do. Read the guide → Guide 02 · May 31, 2026 What is Claude, and why I built everything on it Anthropic's assistant: the Opus/Sonnet/Haiku tiers, why it's in the top tier, and everything I run on it. Read the guide → Guide 03 · May 31, 2026 Why AI genuinely changes how we work A shift on the same scale as electricity: what it flips for a business owner (doing → directing), and what it doesn't change. Read the guide → Guide 07 · May 31, 2026 What is an "AI agent"? An assistant responds, an agent acts — in a loop, until the job is done. What that enables, and the guardrail without which it goes off the rails. Read the guide → Guide 16 · June 10, 2026 Fable, Mythos, Glasswing: why Anthropic released two versions of the same model On June 9, Anthropic launched Fable 5 — the first building block of a class above Opus — and reserved the guardrail-free version, Mythos 5, for vetted cyber-defenders. The story, jargon-free. Read the guide →

02 · Tooling

Getting set up

The concrete building blocks for going from "chatting" to "making AI act" on your tools.

Guide 04 · May 31, 2026 Chat, Cowork, or Code: which one to use Three ways to use Claude depending on the task, the truth about the "terminal," and how to install Claude Code in a single command. Read the guide → Guide 05 · May 31, 2026 What is GitHub, and why it's free and magical The vault that backs up your work, keeps the full history, and once connected, deploys itself automatically. For non-developers. Read the guide → Guide 06 · May 31, 2026 What is an "MCP," and how to connect Claude to your tools Copy-paste, browser, API, MCP: the 4 ways to connect Claude to Shopify, Gmail, Drive… and why MCP changes everything. Read the guide → Guide 17 · June 10, 2026 What is a "loop"? Recurring agent tasks An agent that repeats the same task on a regular schedule: monitor, check, alert — without you. Claude Code's /loop, how it differs from a scheduled task, and the framework. Read the guide →

03 · Method

Working well with AI

The habits that make the difference between a "gimmick" and a real time-saver.

Guide 13 · May 31, 2026 How to talk to AI effectively: writing a good prompt A good prompt = 80% of the result. The recipe (context, task, format, constraints), the key habits, and why iterating beats the "magic phrase." Read the guide → Guide 08 · May 31, 2026 Writing a great CLAUDE.md (the highest-ROI file you'll create) The file that tells Claude how to work with you: tone, conventions, pitfalls. How to write it and grow it over time. Read the guide → Guide 09 · May 31, 2026 The 4 principles I apply with AI (Karpathy-style) Who Karpathy is, and 4 habits that make AI save time: think first, keep it simple, stay surgical, steer by objective. Read the guide → Guide 10 · May 31, 2026 Compound Engineering: your method improves itself Every lesson becomes a reusable rule: the compound interest effect applied to how you work with AI. Read the guide →

04 · Decide

For business owners and merchants

The strategic questions: high-ROI use cases, cost, data, visibility.

Guide 11 · May 31, 2026 AI in practice, for a merchant or an SMB The use cases that pay off quickly, without a technical team, seen from my store — and where to start. Read the guide → Guide 15 · May 31, 2026 What AI costs: subscriptions, API, and keeping the bill under control Everything is counted in tokens. The ways to pay (free, subscription, API, enterprise), which one to choose, and how to keep costs in check. Read the guide → Guide 12 · May 31, 2026 Is my data safe? Where your data goes, who sees it, whether it's used to train models — and the simple habits (plan, settings, sovereignty) to stay worry-free. Read the guide → Guide 14 · May 31, 2026 Getting visible in AI: GEO explained The SEO of the AI era: getting cited in ChatGPT, Perplexity, or Claude's answers. Why it matters and how to approach it. Read the guide →

05 · Build in public

Case studies

Concrete, with real time and cost figures: how, without knowing how to code, I shipped a full site and an AI Watch bot in a weekend with Claude Code.

Deep dive · June 14, 2026 Does buying an ERP still make sense when AI lets you build your own? HF-OS: a press ERP built from scratch in 5 days / 96 PRs with Claude Code, plus the extended Odoo at Cobra. CRM, ad management, invoicing, follow-ups — and the real build vs. buy question. Read the deep dive → Comparison · June 16, 2026 Odoo or custom ERP: which to choose in 2026? A criterion-by-criterion decision comparison (cost, timeline, flexibility, dependency, risk) from someone who does both: extended Odoo at Cobra, custom ERP built in 5 days. Read the comparison → Guide · June 1, 2026 A site + an AI bot in a weekend? The big question, in 3 reading depths: start with the need, pick the dumbest possible stack, separate responsibilities, put guardrails in place. Read the guide → Case study · June 1, 2026 Knowledge Hub: from idea to production A complete static site (70+ pages, dark mode, GEO layer) in vanilla HTML/CSS/JS on GitHub Pages. Decisions, roadblocks, real time. Token cost: zero. Read the case study → Case study · June 1, 2026 AI Watch bot: from idea to production A serverless agent: X/Twitter → Claude API → Telegram + site. 2-repo architecture, cross-repo bridge, CI headaches. Real cost ~a few cents/day. Read the case study → Case study · July 13, 2026 A competitive pricing dashboard, built in-house The competitor data (Wiser) was already in prod, delivered by the integrator. The management layer — quoted at several weeks, sitting in the backlog — built internally with Claude Code in a matter of days. The most clear-cut build vs. buy of the year. Read the case study → Case study · June 17, 2026 An agent that processes supplier invoices It reads emails, extracts each invoice, matches it to the right Odoo order, and pushes it into Teams. Hosted on Scaleway, 15-min cron. ~€2M in historical drift identified — and a workflow that can't drift anymore. Read the case study → Making-of · June 1, 2026 Shipped in a weekend The story, from a dinner with friends to "a beautiful thing shipped in 2 days." Without being a developer, working side by side with Claude Code. Read the story →

The series keeps growing. For a dated log of what I'm building, head to the journal.