The hype around AI agents is giving way to harder questions. How do you actually manage agents in production? What happens when they break? And why are the biggest wins coming from teams that throw out their old processes entirely?
01
Boris Cherny: Stop speeding up old work, redesign everything for agents
Boris Cherny, who works in AI product development, made a sharp observation about teams getting the biggest wins from AI: they're not using it to speed up existing workflows, they're completely changing how work gets done. "What steps can you delete, what handoffs go away, what can an agent just own end to end," he wrote, citing Salesforce's deep approach to AI transformation.
Why it matters: Your company is probably trying to make AI fit into your current org chart. The teams actually winning are asking which parts of that org chart can disappear entirely. That's a much scarier but more profitable question.
Onyx Security CEO: Agents need watchers because they're already breaking things
Onyx Security CEO Maxim Bar Kogan appeared on No Priors to discuss his company's approach to agent oversight. His thesis: "As you're exponentially doing more things with AI, you're going to start having really bad actions happen. We've seen agents accidentally publishing code and tokens that they weren't supposed to." Enterprises realize the risk has grown exponentially but can't stop agent adoption, so they need guardrails instead.
Why it matters: The AutoGPT demo that inspired Onyx in 2024 looked like science fiction. Now agents are accidentally leaking credentials in production, and someone has to build the systems that watch the watchers.
Box CEO Aaron Levie thinks $500M infrastructure spending proves the app layer's value
Box CEO Aaron Levie responded to news of a major company spending $500 million to build their own application infrastructure, calling it "the app layer couldn't get a better advertisement than a company spending $500M to build their own version of it." He sees this as validation that software companies have tremendous value.
Why it matters: When enterprises are willing to spend half a billion dollars to avoid depending on third-party software, that tells you exactly how valuable controlling the application layer has become. Every SaaS company should be taking notes.
Cursor's Thibault Sottiaux teases big adoption numbers
Cursor co-founder Thibault Sottiaux posted cryptically about "a number today on a codex dashboard" that made him happy, promising more details soon. He thanked users for adopting Codex and emphasized "we are still early. So early."
Josh Woodward says multilingual AI is now "ridiculously easy"
Developer Josh Woodward shared a brief note that building multilingual AI applications has become "ridiculously easy," though he didn't elaborate on the specific tools or approaches making this possible.