News.

Practical writing on AI, software and EU-compliant building — from the team that builds it. New articles weekly.

15 July 2026 · Erwin Laravel + AI: the patterns we reuse on every SaaS build. Laravel 13 shipped an official AI SDK, so "how do we add AI" is a solved question. The interesting part is the wiring around the model call. The six patterns we reach for every time — the model layer behind an interface, AI on its own queue, streaming that survives a refresh, structured output, cost logging and a fake provider in every test. 13 July 2026 · Jeffrey Voice AI for support flows: Hume, ElevenLabs and when to use which. Your client wants "a voice bot for support" and expects a Hume-vs-ElevenLabs answer. Wrong first question. Pick the architecture — cascading (STT→LLM→TTS) or speech-to-speech — and the platform half-chooses itself. Where each one wins, and the EU emotion-recognition rule that can decide it for you. 6 July 2026 · Erwin EU AI Act risk classification in practice: one real use case, walked through. "Is our AI high-risk?" has a procedure, not a gut answer. We take one real build — an AI that reads CVs and shortlists candidates — and run it through the Act's four checks: prohibited practices, the Annex III trap, the exception that can save you, and the one design choice that removes it. 1 July 2026 · Jeffrey The real cost of an AI integration in 2026 — and what drives it up. Every quote opens with "what does the AI itself cost?" — the wrong number to fixate on. With real 2026 model prices: why the token bill is the smallest, most predictable line, and where the money actually goes — retrieval, evaluation, retries, guardrails and maintenance. 29 June 2026 · Erwin Copilots on company data: RAG without the buzzwords. Your client wants an AI that answers questions on their own documents. That's RAG — and the weekend demo always works. Production is where it breaks. The retrieval stack we actually ship: chunking, hybrid search, reranking, and the evaluation step everyone skips. 22 June 2026 · Jeffrey GDPR for AI features: the 8 questions your client's DPO will ask. The AI Act gets the headlines, but the GDPR is what gets your AI feature blocked the week before launch. The eight questions a data protection officer puts to you — lawful basis, automated decisions, DPIA, data transfers — and the answers that keep the project moving. 17 June 2026 · Erwin White-label development, week to week — what it actually looks like. Most white-label means one rigid thing: coders who stay hidden. Ours flexes — invisible when you want it, or fronting your client under your brand when that's easier, because we're lead developers, not just a pair of hands. 15 June 2026 · Jeffrey Mistral, Bedrock or Anthropic direct: how we pick the model stack for EU clients. They get framed as three competitors. They're three layers of one decision — which model, and where it runs. The order we actually decide in when the client is in Europe and the data has rules. 11 June 2026 · Erwin EU AI Act: what actually applies from 2 August 2026 — and what just moved. Transparency duties land in August; the high-risk obligations just moved to December 2027 under the Digital Omnibus. The timeline as it stands — and the build checklist that doesn't change. 11 June 2026 · Jeffrey The future of digital belongs to small, experienced teams — empowered by AI. Headcount used to be a proxy for capability. AI broke that proxy. Why a senior team of three now outbuilds a feature factory of thirty — and what we changed in our own workflow.