Amelia Wattenberger

Amelia Wattenberger builds and writes about AI interfaces, from writing tools with explicit context controls to workspaces for coordinating coding agents. Augment’s 2026 biography identifies her as Product Lead for Intent and a partner at Sutter Hill Ventures, following earlier work as a Principal Research Engineer at GitHub Next. Her public arguments emphasize making AI tools understandable and preserving the user’s ability to direct and evaluate the work.[4][3][6]

In her February 10, 2026 introduction to Intent, Wattenberger argues that the bottleneck in AI-assisted development has shifted toward tracking agents, plans, and changes ready for review. The workspace her team developed uses isolated git worktrees. A coordinator agent proposes a specification for human approval; implementor agents carry out the work; and a verifier checks results against the specification before returning them for human review.[4]

A central design decision is the jointly maintained specification. In “What spec-driven development gets wrong”, published February 20, 2026, she explains: “The spec isn't a human artifact or an agent artifact. Both sides read from it and write to it.” Agents update the plan with discoveries, changed assumptions, and constraints; developers can pause and revise it. She identifies update granularity as a design problem: reporting too much creates noise, while reporting too little hides consequential decisions.[5]

Wattenberger also cautions against letting faster execution dictate the pace of evaluation. In her March 9, 2026 essay “Building faster than you can think”, she writes, “Judgment doesn't scale with tooling.” She recommends writing specifications to expose assumptions and establish concrete review criteria, and comparing inexpensive prototypes to clarify what a developer actually wants.[6]

Before Intent, Wattenberger explored interfaces for AI-assisted writing, documentation, and developer learning. These earlier projects provide concrete context for her arguments about explicit controls, shared working material, and active participation.[4][3][1][7]

Wattenberger presented PenPal at the early-April 2023 San Francisco AI | UX meetup, documented in Latent Space’s April 26 recap. The writing app offered controls for feeling and inspirations alongside suggestions, summaries, and praise. In her personal essay on chatbot interfaces, she describes adding controls for writing type, writer identity, desired feeling, and inspirations to an AI writing tutor that suggests improvements such as smoothing transitions or adding tangible examples.[1][3]

In the undated essay “Why Chatbots Are Not the Future”, Wattenberger argues that a blank text box gives users little guidance about a tool’s capabilities or limits. She treats prompts as collections of context that interfaces could help users supply. She also criticizes isolated responses: comparing successive rewrites requires scrolling between versions rather than manipulating a persistent working buffer. For writing and coding, she argues, repeated prompting and reading forces users to switch too frequently between implementation and evaluation.[3]

Her team’s Copilot for Docs prototype explored personalization within a conversational interface. Wattenberger describes two stages: finding information relevant to a question, then synthesizing an answer from it. Sliders supplied context about the questioner’s experience and needs, allowing the interface to distinguish, for example, a developer new to a library from someone seeking a quick answer.[3]

GitHub Next credits Wattenberger and Eirini Kalliamvakou as the makers of Learning Sandbox, published in October 2024 as a research prototype. It uses LLMs to generate explanations and interactive code examples tailored to a developer’s task and existing knowledge. Developers can edit code, observe changes immediately, and ask questions that update the learning material. Its emphasis on hands-on experimentation gives users a way to test their understanding while making progress on a specific task.[7]