正文 Markdown
One of the finite limits to automation is the issue of accountability in a workflow in an organization. In theory the agent can do anything, but whatever you have the agent do yourself, you should expect to be accountable for that work at some level. There are lots of conversations around an agents compressing job boundaries in an organization. This will certainly happen, but the limit of this becomes our own ability to understand the domains that we’re compressing. You have to take on the context of what that other job has to think about. It’s now easier for a product manager to write software, and it’s easy for an engineer to get product specs now. But those are only tiny parts of the total context of either of those jobs. The product manager should be spending their time talking to customers, studying the market, studying competition, thinking through design and usability, and so on. And the engineer should be thinking through the entire system design, making sure they build secure software, handling site issues, and so on. Merging the context of these two entirely different tasks eventually means you get worse results from both. The same is true for lots of other parts of an organization. Companies didn’t specialize roles over time just for fun. They specialized because the market caused the need for specialization as expectations changed. Some boundaries will disappear between roles because of AI, but equally agents will become force multipliers of specialization as well in other areas.