AI products have a unique design problem: the more capable they are, the easier it is to overwhelm the people using them. Here's how we think about clarity as a design constraint, not an afterthought.
Most software has a relatively fixed shape: a CRM has contacts and deals, a calendar has events and dates. AI products don't have that luxury — the same chat box might write code, summarize a document, or plan a birthday party. That flexibility is the whole point, and it's also a design trap.
Our approach starts from a simple constraint: at any given moment, a user should be able to answer "what can I do here, and what will happen if I do it?" without guessing. That sounds obvious, but it shapes everything from how we label AI modes to how we preview the consequences of an action before it runs.
We lean heavily on progressive disclosure — showing the simplest version of a capability first, and revealing depth only as someone reaches for it. Forge's three modes (Ask, Agent, Plan) are a good example: a newcomer can start with simple Q&A, and only encounter "autonomous coding" once they're ready to trust it.
The hardest part of this work isn't coming up with the simple version — it's having the discipline to keep it simple as the product grows. Every new capability is a small argument for one more toggle, one more mode, one more thing to explain. Our job is to keep losing those arguments on behalf of the people who'll actually use the product.