streugut.ch
Archive

The direction

Where the AI race is heading

AI competition is moving away from a single frontier race and toward control of the path from model to deployed workflow. That shift is being carried by agents that can act inside existing software, by partner-led enterprise rollouts that package adoption and governance, and by an infrastructure contest widening from chips to capacity, financing, and power. What is particular now is how directly vendors are tying these layers together into managed systems for real work.

The move

What actors actually did

Microsoft marks new era at Build, OpenAI and xAI debut fresh coding models

Microsoft’s annual Build conference opened in San Francisco on May 31, the company’s first major developer event since separating from OpenAI. In the lead-up, Microsoft previewed a suite of homegrown AI models, including a new coding model intended to strengthen GitHub Copilot. During the week, Microsoft also released Windows 365 for Agents, allowing Copilot agents to run workflows in Cloud PCs across enterprise environments, and launched SRE Agent tools in Azure MCP Server for developer access via IDEs, terminals, and AI assistants.

Several competitors launched new agentic coding models. OpenAI updated GPT-5.5 Instant in ChatGPT and the API, improving response quality and retiring older models like GPT-4.5 and o3. Meanwhile, xAI released Grok Build 0.1, a model designed for agentic coding tasks, now available in public beta via its API. Anthropic rolled out Claude Opus 4.8 with enhanced performance in coding, agentic tasks, and reasoning benchmarks, along with new features and reduced pricing.

In parallel, Nvidia’s Vera CPU, engineered for high agentic AI throughput, debuted in benchmarks, and SoftBank pledged up to €75 billion to develop Europe’s largest AI data center project in France. These moves underline the increasing focus on infrastructure and computing power as the agentic AI landscape evolves.

Sources