标签 #ai-spend
5 篇文章。
-
· Paul Lukic
找到评审漏掉的东西的那次评审：面向大改动的可选深度评审
一个评审者看一遍 diff 会放行的东西，几个对抗式评审者加反驳者不会放行。这篇文章讲一次多代理评审在常规评审已经通过的 176 次提交上找到了什么、它花多少钱、为什么它是可选的，以及哪种循环形状能防止它吃光你的预算。
-
· Paul Lukic
One Model for Six Jobs: Why We Made Per-Task Model Selection Opt-In
Coding agents delegate to sub-agents with wildly different demands, and almost everyone runs all of them on one model. Here is the cost of that, why a shipped default would be worse, and how we built model selection that asks once and always reports what it did.
-
· Paul Lukic
AI 编程代理的递归自我改进：哪些是真的，哪些还是理论，以及我们实际交付了什么
递归自我改进，是指一个人工智能系统改进自己“改进自己”的能力。这篇文章讲清楚理论到底说了什么，为什么模型层面的循环至今仍是理论，以及我们如何为编程代理做出了指令层面的版本，并让人守在闸门前。
-
· Paul Lukic
AI Tool Sprawl: Your Devs Are Drowning in Inefficient Agents
Running five AI coding tools doesn't give you five times the leverage. It gives you five agents grepping the same repo, and a token bill that quietly compounds.
-
· Paul Lukic
Build vs. Buy Code Intelligence: The Real ROI
Building in-house code graphs costs $250k+ and 6 months. Adopting Coograph cuts AI spend 80% while freeing engineers to ship product.