Surface governance worked for 30 years because a human's blast radius is bounded. People are slow, interruptible, and can only do so much damage in a day.
An agent removes the bound. N tool calls a second, across systems, every call individually legitimate.
So the boundary has to be structural, not bolted on.
probabilistic -> plan, suggest, draft
deterministic -> write, commit, pay, approve
Anything added after the architecture is chosen gets routed around. Not maliciously — the agent finds the path of least resistance.
最近大家都在說 Agent First System
近期灣區一堆優退、裁員或是自己開一人公司的案例
AI 可以讓員工第一次感覺自己像一人公司,接模型、跑 agent、做決策,
同時公司看到的不一定是效率神話,而是是可能失控的效能
Agent First,是員工看到的降落傘,還是老闆看到的裁員名單
https://t.co/DoXzIdfWQB
Yes, I actually used @InterviewCoder to get an Amazon offer.
Yes, I used to also run an account that taught people how to solve Leetcode questions.
Enjoy.
https://t.co/cgzrMDBE2Q
Introducing Willow, our new state-of-the-art quantum computing chip with a breakthrough that can reduce errors exponentially as we scale up using more qubits, cracking a 30-year challenge in the field. In benchmark tests, Willow solved a standard computation in <5 mins that would take a leading supercomputer over 10^25 years, far beyond the age of the universe(!).