How AI Tools Are Changing Design in 2026

Key Takeaways

• Designers are moving away from traditional tools like Figma toward AI-powered platforms like Conductor, Paper Design, and Claude
• Voice-to-code workflows (using tools like Aqua) replace typing, enabling faster ideation and feature building
• "Disposable design" — rapidly generating multiple design iterations and selecting the best elements — is becoming standard practice
• Custom internal tools built with Claude enable fine-tuning of details like shader effects without requiring traditional graphics software
• Dual-purpose design (human vs. machine-readable versions) is emerging as websites serve both users and AI agents

The New AI-First Design Workflow

Eve Bouffard, Head of Design at Y Combinator, now works almost exclusively in Conductor and Paper Design for end-to-end projects. Rather than typing, she uses Aqua, a voice-to-code tool, to describe features in a stream of consciousness — letting AI handle the implementation. This shift fundamentally changes how designers think about iteration and problem-solving. Instead of manually building mockups, designers can now generate dozens of variations in minutes and cherry-pick the best elements.

Building with Context: The "soul.md" Approach

For the Sodazine project, Eve recorded every meeting and dumped all transcripts into a single soul.md file — treating it as the definitive source of truth for the project. This exhaustive glossary of context enables Claude and other AI agents to make better decisions across all future design work. Rather than writing high-level takeaways, capturing everything and letting AI extract patterns is proving more effective. The same applies to manifestos, design guidelines, and written content — all consolidated into one file that feeds every subsequent creative decision.

Disposable Design and Rapid Iteration

When designing the Sodazine website, Eve created a mood board on Pinterest, then fed those images plus the project context into Claude and asked for 16 one-shot website variations. She even built a personal glossary tool to bookmark and organize her favorite iterations — a custom internal tool that would have taken hours in traditional design software. This "disposable design" approach prioritizes speed and exploration over polish in early stages. Designers can now generate, review, and refine hundreds of ideas in the time it used to take to create one high-fidelity mockup.

Human vs. Machine Content Design

A notable innovation on the Paxel landing page is the toggle between "human" and "machine" versions of the website. The machine version is a distilled markdown file optimized for AI agents to consume quickly, with a disclaimer instructing agents not to run commands. This reflects an emerging design challenge: optimizing not just for human visual experience, but for how AI agents parse and understand content. It's content design, not visual design — ensuring agents get exactly what they need to be effective.

Custom Tools for Fine-Tuning

Rather than accepting AI's first output, Eve builds micro-tools for herself. For the dithering shader effect on Paxel, she created a modal to tweak every parameter until the feel was perfect — then discarded it. For Startup School speaker cards, she built a tool that automatically generates personalized cards with speaker images pulled from her inbox. For shader-based animations, she created a screen-recording timer to capture perfect four-second loops that seamlessly repeat on social media.

This "tool-building muscle" is a new skill emerging in design: recognizing that you can build anything for yourself to solve a specific problem, then discard it once done. Everything becomes editable, movable, and changeable — limited only by creativity and imagination.

Practical Applications: Paxel, Sodazine, and Startup School

Paxel analyzes coding transcripts to show developers insights about their patterns — inspired by Spotify Wrapped. The landing page explicitly states its purpose upfront, using interactive cards with hover effects. A form at the bottom allows users to submit feature requests and bug reports as natural language prompts; these fire off an agent in the backend that opens pull requests for the team to review.

Sodazine — short for "State of the Art" — is a physical zine celebrating San Francisco, created collaboratively with artists and writers. For the promotional website, Eve deliberately avoided AI for the graphic design and cover art, choosing Illustrator instead to maintain intentionality and craft. However, she used Claude extensively to generate website layout variations based on the mood board and project context.

Startup School at Chase Center features personalized tickets with shader-based visual effects. Each ticket renders the attendee's name, city, and event information, reusing the same shader parameters across all marketing assets — ensuring consistency from acceptance notification through in-venue branding.

Conclusion

AI tools aren't replacing designers — they're fundamentally changing the design process. The bottleneck has shifted from execution to imagination. Designers who master voice-to-code workflows, context management (via files like soul.md), rapid iteration, and custom tool-building will define the next era of digital design. The future isn't about photorealism or pixel-perfect mockups; it's about capturing intent, enabling fast exploration, and building the right creative muscle.

---

Original source: New Ways To Design With AI Tools

powered by osmu.app