Jensen Huang has joined Donald Trump’s China trip as a late addition. That does not confirm any specific AI deal is being negotiated, but it is a strong enough signal that chips, AI infrastructure, and market access are now live topics at the state level.

SAP has made its thesis unusually explicit: enterprise AI will not win by stapling chatbots onto workflows. It will win when agents are embedded directly into ERP, procurement, supply chain, and the governance layer companies already trust. Autonomous Enterprise and the SAP Business AI Platform are really a push to turn AI into an execution layer inside the system of record.

Perplexity is positioning Qwen3 235B on GB200 NVL72 as a production inference story, not just a training headline. The more important signal is that Blackwell is being framed as an advantage in throughput and cost-to-serve for large MoE models.

Reimagining the mouse pointer for the AI era: Google DeepMind is rethinking the mouse pointer itself as a way to express intent to AI, instead of treating the chat box as the default entry point. If this direction holds, desktop interaction starts to look less like manual clicking and more like a control surface for agents.

SAP and Anthropic bring Claude into SAP Business AI Platform: Claude is becoming a reasoning and agent layer inside SAP’s AI stack. The broader signal is familiar but important: enterprise AI stacks are increasingly being decided by workflow embedding and governance fit, not raw benchmark performance.

Googlebook turns the AI work surface thesis into a commercial product move: If AI Pointer is the interface demo, Googlebook is the product attempt. Google is clearly trying to turn the idea of an AI-native work surface into a commercial computing category, not just a research concept.

Fast mode for Claude Opus 4.7 is an interesting latency and cost signal: According to ClaudeDevs, fast mode for Claude Opus 4.7 is now in research preview on the API and in Claude Code. The sourcing here is still thinner than the rest of the issue, but if accurate, it is a useful sign that model vendors are optimizing for workflow fit and latency, not just quality.

/goal looks like a small but meaningful primitive for anyone using coding agents every day. The core idea is simple: instead of letting the agent finish one response and stop, you give it a concrete objective and make it keep going until that objective is actually complete.

That solves a familiar problem. Agents often stop too early, declare victory when only part of the work is done, or complete a prompt without completing the operational job. /goal turns persistence into a first-class control instead of forcing the user to rebuild it with longer prompts.

The best use case is work with a clear definition of done: fix a bug until tests pass, finish a refactor flow, or ship an artifact to the correct approval gate. For fuzzier tasks, it can still help, but it needs tighter scope boundaries so the agent does not wander.

The practical impact is straightforward: you get a persistence layer at the workflow level instead of the prompt level. For heavy agent users, that is the kind of small change that compounds into real time savings.

Read yesterday’s One Practical Use of AI: Use GPT-Realtime-Translate and GPT-Realtime-Whisper to build multilingual voice workflows

If you’re optimizing for growth, you need ecomm tactics that actually work. Not mushy strategies.

Go-to-Millions is the ecommerce growth newsletter from Ari Murray, packed with tactical insights, smart creative, and marketing that drives revenue.

Every issue is built for operators: clear, punchy, and grounded in what’s working, from product strategy to paid media to conversion lifts.

Subscribe for free and get your next growth unlock delivered weekly.

Sign up now

Brookings makes the point many teams still avoid: model benchmarks are not enough to tell you whether an agent will perform well in the real world. Agent performance depends on domain, environment, user behavior, and the way the task itself is structured.

OpenAI is telling a story that stretches beyond the startup ecosystem. AI adoption is now spreading into smaller businesses and everyday operators, from retail to services to trades. That matters because it pulls AI back out of the “frontier labs versus big tech” narrative and back into a simpler question: does this tool help people work faster, cheaper, and with more autonomy?