Simon Willison

Highlight

The standout post today is Stop Making TUIs, where Simon echoes Thomas Ptacek’s argument that AI coding agents have effectively reduced the cost of building native GUIs to zero. It highlights a significant paradigm shift for developer tools, suggesting that the era of default Terminal User Interfaces (TUIs) for throwaway scripts is coming to an end in favor of “vibe-coded” native applications.

Posts

Stop Making TUIs · simonwillison.net Simon highlights a provocative piece by Thomas Ptacek arguing that developers should stop building Terminal User Interfaces (TUIs) and start building real native desktop applications. Thanks to coding agents, the cost of generating standard, functional GUIs has fallen to nearly zero. Simon reflects on his own experience with his SwiftUI-based bandwidth and GPU task bar monitors—which he “vibe-coded” in March 2026 and still uses daily—and admits that developers are running out of excuses to avoid building native apps.

llm-openrouter 0.7 · simonwillison.net Simon has released version 0.7 of his OpenRouter plugin for his LLM CLI tool, introducing compatibility with LLM 0.32. This upgrade enables the plugin to work much better with reasoning models on OpenRouter by leveraging OpenRouter’s native implementation of the Responses API. Excitingly for developer tool power users, the release adds three new server-side tools—Shell, WebFetch, and WebSearch—which can be easily enabled via command-line options like -T WebSearch.

llm 0.32.1 · simonwillison.net This urgent patch release addresses a sudden issue where fresh installs of Simon’s LLM CLI tool broke because the OpenAI Python library dropped its dependency on httpx. Since LLM relied on httpx but only installed it transitively via the OpenAI package, this dot-release temporarily fixes the problem by pinning openai<3. Simon notes that a major migration is on the horizon, with the upcoming 0.33 release planned to transition from httpx to Pydantic’s httpx2 library.

Quoting Matt Webb · simonwillison.net In this quick link post, Simon quotes Matt Webb’s reflections on using ChatGPT as an interactive tutor while developing Galactic Compass 2. Webb shares that instead of writing the code for him, the AI acted as a patient tutor to teach him how to use quaternions just enough to get the app’s new augmented reality mode working. The key takeaway is that outsourcing cognitive tasks to LLMs doesn’t stop our learning; rather, it often acts as a catalyst that pushes developers to learn even more deeply.

Project Pulse

Simon’s open-source updates today focus heavily on maintaining and expanding his LLM ecosystem, demonstrating active maintenance to handle breaking upstream dependencies while simultaneously introducing advanced integrations like server-side tools for reasoning models.

🎧 This digest would make a fantastic quick audio overview if you want a podcast-style brief to listen to on your commute!