Gamaro news and resources.
RSS feedPublished on October 10, 2026
A personal agent and a company's agent can read the same data and run on comparable models. What separates them is their principal: the objective they serve, the rules they apply, the trail they can prove. Agentic commerce protocols are starting to formalize this mandate without aligning objectives, and one promise too many can bind the company.
Read morePublished on September 10, 2026
An April 2026 study shows that a model steered by its prompt alone beats a LangGraph orchestration on procedural tasks. But an average score does not guarantee that an anomaly already spotted will not come back. A field report on the code harness around the model, and on the deterministic guards born, one by one, from anomalies spotted while building the product.
Read morePublished on September 2, 2026
Classic B2B marketing profiling, built on long forms and silent behavioral tracking, can enter into tension with GDPR and ePrivacy requirements. Stopgaps such as asymmetric cookie banners or background data enrichment have already cost several major publishers significant fines, without restoring user trust. Progressive profiling through a conversational agent does not automatically make a process compliant: done well, it can enable a collection architecture more compatible with several GDPR principles, by supporting purpose-specific consent and better-managed data minimization. It also introduces its own risks, starting with the surplus of information a natural-language conversation can surface beyond what was actually asked. The article also covers the legal limits to respect: the restrictions on fully automated decisions (Article 22 GDPR) and the new transparency obligations imposed by the AI Act.
Read morePublished on August 31, 2026
Since February 2026, Leboncoin's catalogue can be searched in natural language directly inside ChatGPT — with no inference bill for Leboncoin, since OpenAI foots it. The real price is paid elsewhere: the conversational interface and intentional data handed over to a third party for free. And if a platform chooses instead to host its own agent, the calculation becomes real again: more than $3 million a year at proprietary API rates. The architecture built to last then combines a compliance router, locally hosted models, and cross-provider fallback reserved for complex cases only.
Read morePublished on August 29, 2026
In 2026, conversion relies less and less on static forms or complex filters. The Model Context Protocol (MCP) connects LLMs to proprietary inventories in real time, while the Agent2Agent (A2A) protocol allows, when needed, a task to be split across specialized agents. Combined with progressive Zero-Party Data collection, this conversational approach is recomposing online search, as illustrated by Leboncoin's integration into ChatGPT.
Read more