[AINews] All Model Labs are now Agent Labs
The article from Latent Space highlights a quiet but significant trend: every major AI lab is now positioning itself as an agent lab. OpenAI's recent GPT-5 preview emphasized agentic capabilities, Anthropic's Claude 4 is being built with tool use as a first-class feature, and Google's Gemini 2.0 is designed for autonomous task completion. This convergence means that future model releases will prioritize reliability in multi-step reasoning, function calling, and long-horizon planning over raw language generation. For developers, this translates to APIs that are increasingly agent-first, with built-in support for loops, memory, and external tool integration. The consequence is that building agents will become more about orchestration and less about prompt engineering, as models handle more of the reasoning internally. The article notes that this shift is already affecting pricing and availability, with agent-optimized models commanding higher per-token costs but offering lower latency for complex tasks.
Agent-first models will shift developer focus from prompt engineering to orchestration.