Eve Bouffard

person · updated Jul 13, 2026

person concept tool org talk claim — click a node to jump to its page; hover an arrow for the relation

Eve Bouffard is the Head of Design at Y Combinator (YC), where she applies AI-assisted workflows to design projects ranging from digital branding to physical installations. She is a prominent practitioner-advocate for coding-agent-first design, arguing that traditional tools like Figma can be largely replaced by coding agents for most of a project end-to-end.

Toolchain and Workflow

Bouffard has shifted her practice away from conventional design software, working "almost exclusively nowadays in Conductor and Paper Design" for coding-agent-driven design. (0:31) She also uses Aqua, a YC company, to interact with her computer via voice. (0:22) For visual inspiration she still returns to Pinterest to build mood boards, which she argues are a critical input for breaking generic AI output. (0:41, 21:11)

For code generation and implementation she uses Claude, including asking it to implement paper shaders — specifically the dithering shader from paper.design — directly into her projects. (8:20, 8:28)

Core Design Philosophy

The bottleneck has shifted

Bouffard claims that AI capability is no longer the limiting factor in design: "The biggest bottleneck in AI-assisted design is no longer software capability but the designer's creativity and imagination." (6:33) This frames the designer's role as primarily one of ideation and taste rather than execution.

Disposable / one-shot prototyping

She advocates strongly for disposable design — building custom single-use fine-tuning tools and then discarding them — describing this as "a key muscle to develop when working with coding agents." (6:21) Her practice of one-shot prototyping includes generating the same design prompt 16 different times to explore variation space. (16:57, 18:47)

Context saturation as a design method

Bouffard's strongest methodological claim is that flooding an agent with rich context produces better output than constrained prompting: "Capturing as much information as possible and sharing it with an agent is the best way to build software going forward." (16:16) She advocates for soul.md — a file populated by dumping full meeting transcripts rather than high-level notes — as a "source of truth" that the agent can draw on continuously. (14:18) A well-fed agent, she argues, will "surprise designers by including details and ideas they would not have thought of themselves." (19:52)

Visual references over text prompts

To escape generic AI aesthetic output, she recommends providing mood boards and real website examples rather than relying on text descriptions alone. (21:11)

Predictions and Forward-Looking Claims

Bouffard makes several structural predictions about how software and design will evolve:

  • Dual-layer websites: Future sites will need both a human-facing visual version and a separate machine-readable markdown version for AI agents to consume. (7:07)
  • Agent-triggered PRs: Feature request forms that automatically fire off agents to open pull requests represent "the future of how software will be built." (15:16)
  • Scalable shader-driven branding: Shader-based visual identities produced through coding agents can scale consistently from social media cards to arena-scale physical screens. (30:03)
  • Designing for agents is a content problem: Creating interfaces or assets for AI agents is fundamentally an exercise in distilling and structuring information, not a visual design challenge. (8:13)