252 artigos por Marcelo Santiago
Microsoft is bringing the Copilot harness to Foundry Agent Service. Understand the role of this orchestration layer, how it relates to models and tools, and what builders still need to validate.
Version 2.2.5 lets you reserve a percentage of the context window in compaction.buffer while keeping compatibility with fixed token values.
Version 2.1.316 controls subagent effort, fixes orphaned shell processes and schedules the removal of fast mode in Opus 4.8.
A practical workflow for reviewing diffs, asking an agent for feedback, and deciding what goes into the pull request without blindly accepting suggestions.
Learn how to organize project instructions for Codex, code review and other agents without repeating rules in every prompt.
The partnership with G42 connects AI agents to ADNOC operations, from wells to market, with human oversight and infrastructure in the United Arab Emirates.
Announced on November 2, LongCat-2.5 combines multimodal input, agentic coding and API access. See the published numbers and how to test it.
The GDPx benchmark evaluates agents across 11 high value domains. The best system scored 12%, with results varying by task.
In December, OpenAI will share the ChatGPT and Codex usage quota across Plus, Pro and Business plans. Understand what has already been confirmed, what is still to be defined and why automatic credit top-up is a separate mechanism.
Playwright CLI lets programming agents interact with a browser via commands and skills. See how to test a web app flow on TodoMVC, install the tool, and protect sessions and login data.
Learn how to add public videos with captions to Gemini Notebook and query the transcript, without confusing the imported text with the whole video.
Use llm-d 1.0 on Kubernetes with five checks for architecture, performance, tenants, observability and safe updates.
In-depth technical curation: agentic architectures, real token benchmarks, MCP servers and open-source tools in your inbox and on the blog.
