Stop Designing Software for Humans: Design Software for AI (mrkelly.medium.com)

A provocative thesis: with capable large language models and agents now able to understand goals, plan actions and execute tasks, software should be redesigned not for humans as operators but for AIs as operators with humans as controllers. The article argues most legacy and enterprise software was built around clicking, menus and workflows because machines couldn’t interpret intent. That era is ending — email, CRMs, finance, HR and every vertical can be reconceived so AI triages, summarizes, reconciles, schedules and even negotiates on users’ behalf. This is both the biggest opportunity (new product categories, startups) and threat (incumbents that cling to UI-first designs may lose relevance). Technically, this shift implies new product primitives: goal-oriented APIs, tool invocation and execution layers, auditable action logs, interruptible control loops, and dynamically generated, context-aware UIs that prioritize decision-making over operation. Designers must build interfaces as safety and oversight cockpits — real-time dashboards for approval, correction and traceability — while engineers must enable secure, verifiable automation and human-in-the-loop governance. In short, UI won’t disappear; it will evolve into an adaptive mission dashboard that lets humans set intent and trust AI to act, making trust, observability and controllability the core technical challenges of the next decade.