In two weeks of August, Anthropic made Sonnet 5's price permanent at 2 and 10 dollars per million tokens, updated Opus 5, OpenAI restructured its lineup into durable tiers and Google shipped Gemini 3.7 Flash. If you run an AI feature in production, your cost tiering is already stale.
, 2 min read
In July it was a deadline to prepare for, now it is applicable law: from 2 August chatbots, generated content and deepfakes must be disclosed, and on 31 July the Commission confirmed enforcement would begin. What we fixed across client projects in half a day.
, 3 min read
On 12 June a US export-control order switched off a frontier model for everyone, everywhere, for eighteen days. What that means for a European SME with a feature in production, and how you actually defend against it.
, 3 min read
The Digital Omnibus took effect on 27 July and pushes the high-risk rules out to December 2027. Article 50 did not move: from 2 August 2026 chatbots, generated content and deepfakes must be disclosed, and in Italy ACN does the enforcing.
, 3 min read
The EU AI regulation phases in over time. Many SMEs assume it doesn't touch them. Often it does. Here's what matters if you build AI features.
, 3 min read
It's no longer autocomplete. Agents read the repo, open PRs, fix tests. But the bottleneck moved, it didn't vanish.
, 2 min read
A year after the autonomous-agents season, the operational level has risen. But winning use cases are few and well-defined.
, 2 min read
Discover how GSD can help you manage context and streamline complex project workflows in software development
, 45 min read
Discover how OpenAI Codex uses artificial intelligence to generate high-quality code and improve developer productivity
, 30 min read
Vercel revamped the Marketplace integrating AI assistants for setup, debug, deploy. What really works.
, 2 min read
Microsoft built native AI into VS Code. For Copilot subscribers, is it still worth $19/month? Field test.
, 2 min read
After a year of agents on tickets, lead routing and qualification, we know where they pay off and where they ruin things. Three patterns that work, three that don't.
, 2 min read
Anthropic shipped Claude 4.5 with more efficient prompt caching. For teams running AI assistants in production, savings are real.
, 2 min read
Mature AI video models in 2025. For marketing, training or product: where they pay off and where not yet.
, 2 min read
Two years of AI assistance in our daily work. Which tool we use for what, where each one shines, and why the future is not single-tool.
, 3 min read
Pinecone, Qdrant, Weaviate. Or just pgvector. The real thresholds where dedicated databases win.
, 2 min read
Built to handle whole tasks, not single lines. After a month on real projects, where it shines and where it doesn't.
, 2 min read
In November Anthropic announced the Model Context Protocol. It looks like a technical detail; it is a paradigm shift for anyone coding with AI assistants.
, 2 min read
The first rules on foundation models kick in May 2025. What providers, integrators and end-customers must do.
, 2 min read
Semantic document search across 8,000 internal documents. Postgres + pgvector + embeddings: lessons learned.
, 2 min read
After 38 hours of trilogue, on December 8 2023 the political deal landed. What changes for AI builders and users in Europe.
, 2 min read
Tell us what you need. We reply within one working day.
