Peter Yang 与纳德拉对谈:微软如何围绕智能体重组
13 years after working at Microsoft, I had the pleasure to talk to @satyanadella at Ignite. My top 5 takeaways, lightly edited from his answers:
Peter Yang 在 Ignite 与微软 CEO 纳德拉对谈,整理出五点要点。纳德拉称微软要像一家新公司那样重组,每位个人贡献者都成为带 token 预算的智能体管理者,Copilot Cowork 由 11 名开发者构建,GitHub Copilot 运行时由一名开发者维护。他还表示对齐未必随规模自然出现,需要独立于模型的信任架构,并建议企业记录智能体轨迹用于微调,避免被单一模型锁定。
正文
13 years after working at Microsoft, I had the pleasure to talk to @satyanadella at Ignite. My top 5 takeaways, lightly edited from his answers:
1. Can Microsoft catch up to the frontier labs?
"This reminds me of a question when I was in Azure. People asked, 'Can you ever catch up with Amazon? Does anyone need a cloud other than Amazon?'
My answer is the world will demand more than one cloud and model.
What customers expect from [Microsoft] is flexibility and optionality, while delivering the tokens they need for their task with the economics in mind."
2. Do you have any advice for people who are anxious about how AI is changing work?
From Satya: "My job is to turn that anxiety into restructuring Microsoft so we look like any new company. The new company looks like this:
There's a founder. The founder has a token budget. Then the founder hires somebody, and they get a token budget too. All of Microsoft should look like that.
All ICs are now managers of agents with a token budget. When people see that, I hope they say, oh, there's a path.
The Copilot Cowork product is built by just 11 devs. And one developer builds and maintains the entire GitHub Copilot runtime."
3. What happens to software if it's primarily built for agents?
"All of our business apps are now a headless plugin to Copilot. And they're used more than they ever were, because agents go in 24/7.
I'm doing more CRM lookups than I ever did because I'm not doing it, my Autopilot is.
Every day it looks at my OneNote and my meetings, pulls everything from Dynamics CRM for that customer, and checks the supply chain system. This 360-degree view was never possible before."
4. Do models get safer as they get smarter?
"I used to think alignment would emerge as you scale. Based on empirical results, maybe not.
So you need a trust architecture that's independent of the model, because the liability is all with us.
That's why I like the metaphor of treating it like insider risk."
5. How should companies avoid getting locked into one model?
"Assume the next model drop causes you to replan, so you don't build this massive thing around a model that becomes a boat anchor.
What I'm trying to do is turn the model business into a database business.
Use the model, but what you build on top is yours. Log what your agents do, use those traces to fine-tune, and when the next upgrade ships, fine-tune again and keep going."
@satyanadella if you're reading this would love to go deeper on these topics sometime 🙂
AI 热点