Designing for developers means designing for LLMs too
encore.dev
A new piece argues dev tools need to work for AI too, not just humans. Docs and APIs that only make sense to people are already behind.
Based on reporting by encore.dev — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
There's a shift happening in how software gets built, and it's easy to miss if you're still thinking of AI coding assistants as autocomplete with extra steps. The argument from Encore's piece is blunt: if you design a developer tool, an API, or a set of docs purely for a human reading a screen, you're already building for yesterday's user. Increasingly, the thing parsing your documentation, generating a client, or debugging an error message isn't a person at 2am with a coffee. It's an LLM acting on a person's behalf.
That changes what "good design" even means. A well-written README with clear prose and a friendly tone is great for a human skimming on GitHub. But an LLM doesn't skim — it processes structure, consistency, and predictable patterns. Ambiguous naming, inconsistent error formats, or documentation scattered across five different tone-of-voice styles doesn't just annoy a human engineer; it actively confuses a model trying to reason about your API and produce correct code. The tools that will feel effortless in this new era are the ones that are legible to both audiences at once.
This isn't really a hypothetical concern anymore. Developers are pasting error messages, schema definitions, and whole codebases into ChatGPT, Claude, and Copilot every day, expecting a coherent answer back. When the underlying tool's documentation is messy or self-contradictory, the LLM's output degrades too — hallucinated parameters, wrong endpoints, confidently wrong advice. The blame often lands on the model, but the root cause can just as easily be a poorly structured spec or an inconsistent naming convention on the tool-maker's side.
The practical takeaway is that things like structured error codes, machine-readable API specs, and consistent terminology aren't just nice-to-haves for automation pipelines anymore. They're now part of the core UX, because a growing share of your "users" are language models relaying information to humans who never touch your raw docs at all. Companies that treat this as an afterthought will end up with tools that quietly work worse inside AI workflows, even if the human-facing docs look polished.
My take — AI-written commentary, not fact-checked reporting
This is one of those obvious-in-hindsight shifts that most tool builders are still ignoring, and it's going to bite the slow movers hard. If your API only makes sense to a human reading top to bottom, you're optimizing for a shrinking slice of your actual traffic — the rest is going through a model first. Treat LLM-legibility as a first-class design constraint now, not a 2027 problem.
Read more about this at: encore.dev