Compared to other forms of knowledge work, coding offers one of the least hostile environments for agents to work in. Agents have frictionless access to the core data they need (the codebase) to work from, there are limited access control complexities, there are generally good practices around documentation (compared to other forms of work), and it’s an entirely text-based medium. Kind of the perfect setup for agents. Other areas of knowledge work will require a bunch of similar practices to emerge to enable agents to be effective. We’ll need better documentation around our processes, knowledge will need to be available and formatted in a way for agents to easily work with, access controls will have to be well-maintained, we’ll need clear sources of truth for the work that agents are doing, and so on. All of this will happen, but it remains one of the gaps between the outcomes that we see with coding agents today and the rest of knowledge work. AI coding is providing the roadmap, but it will take some time.