Over the last few days, I have been tracking one question: where is AI getting more practical, not just more impressive?
My answer is this: the most interesting shift right now is not a single benchmark jump. It is AI moving into interfaces and control layers that people already use every day: the browser, the spreadsheet, the enterprise workspace, and the security review process.
1) Google is turning the browser into an AI surface
On March 11, 2026, Google announced that its AI features in Chrome are expanding to India, New Zealand, and Canada. The company also said the new experiences are built on Gemini 3.1.
My take (inference): this matters because the browser is still one of the most durable software surfaces in the world. If AI becomes native to browsing, product discovery, comparison, and research all get pulled closer to the interface people already trust.
2) OpenAI is buying evaluation muscle
On March 9, 2026, OpenAI announced that it has agreed to acquire Promptfoo, a company known for AI evaluations and security testing. The stated goal is to make evals and red teaming more accessible to developers building with OpenAI models.
My take (inference): this is a meaningful signal. AI quality is no longer just about model output. It is about whether teams can systematically test prompts, agents, tools, and failure modes before shipping them.
3) Microsoft is selling governed model choice, not just one assistant
On March 9, 2026, Microsoft introduced Frontier and Frontier Plus for Microsoft 365 Copilot. The notable shift is that enterprise customers can access newer OpenAI models as well as Anthropic's Claude Cowork inside Microsoft's managed productivity environment.
My take (inference): enterprise buyers increasingly want optionality without operational chaos. The winning platform is the one that offers multiple strong models while keeping identity, policy, and workflow controls stable.
4) OpenAI is pushing directly into existing work surfaces
On March 10, 2026, OpenAI announced ChatGPT for Excel with live financial data support. The release is aimed at turning spreadsheet work into something AI can actively accelerate instead of just comment on from another tab.
My take (inference): this is how AI becomes durable in business. It does not win by asking users to change their habits first. It wins by showing up inside the tool where the real work is already happening.
5) Anthropic is trying to normalize security disclosure for the AI era
On March 6, 2026, Anthropic published its proposal for coordinated disclosure around AI-discovered vulnerabilities. The argument is simple: if AI changes vulnerability discovery throughput, disclosure practices need to adapt too.
My take (inference): this is one of the clearest signs that the AI market is maturing. Once systems become useful enough to materially speed up security research, process design becomes as important as model capability.
Why today's mix is interesting
Taken together, these updates point in the same direction. AI is becoming less of a standalone destination and more of an embedded layer across everyday software.
That changes what matters. Distribution matters. Evaluation matters. Governance matters. And product teams that can integrate AI into already-familiar surfaces will probably move faster than teams that only chase frontier demos.
Sources
- Google Chrome AI expansion (March 11, 2026): https://blog.google/intl/en-in/company-news/technology/chrome-ai-experiences-india-new-zealand-canada/
- OpenAI to acquire Promptfoo (March 9, 2026): https://openai.com/index/openai-to-acquire-promptfoo/
- Microsoft 365 Copilot Frontier and Frontier Plus (March 9, 2026): https://blogs.microsoft.com/blog/2026/03/09/introducing-frontier-and-frontier-plus-in-microsoft-365-copilot/
- OpenAI ChatGPT for Excel (March 10, 2026): https://openai.com/index/introducing-chatgpt-for-excel/
- Anthropic on coordinated disclosure and AI-discovered vulnerabilities (March 6, 2026): https://www.anthropic.com/news/coordinated-disclosure-and-ai-discovered-vulnerabilities