Notes from the studio.
Design, AI and building, as we do it.
- AI coding agents for designers who build: Claude Code, OpenAI Codex, Google Antigravity and Cursor compared for designers who ship real apps: where each runs, what it costs and which models it uses. (Models and tools, 2026-11-12)
- AI design and prototyping tools compared: Figma Make, Claude Design, Google Stitch, v0, Lovable and Bolt compared: what each does well, what it costs and where it fits in a real product design workflow. (Models and tools, 2026-11-12)
- Claude vs GPT vs Gemini for product work: Claude Opus 5.5, GPT-6 Astra and Gemini 4 Carbon compared on price, context, benchmarks and real product tasks, from a designer who ships apps with AI daily. (Models and tools, 2026-11-12)
- Design for dark mode from day one: Dark mode added late means a rebuild. Start with semantic color tokens, check contrast in both themes, and let elevation do the work of shadows. (Design, 2026-11-12)
- Design systems that live in code: A design system is only real when it ships. Why we keep tokens and components in the repo, and how that works for a designer and a developer. (Product design, 2026-11-12)
- Designers are becoming design builders: A design builder is a designer who ships working software with AI agents. Why the role is forming now, and what changes for the people who take it on. (Design builders, 2026-11-12)
- Designing AI features people can trust: Trust in AI features comes from small design choices: showing sources, admitting uncertainty, refusing well, and keeping people in control of the result. (AI in practice, 2026-11-12)
- Designing for the time between sessions: Most of an app’s life happens while it is closed. Learn how to design return moments, notifications, widgets and re-entry so the next session starts well. (Product design, 2026-11-12)
- Designing the empty state of a chat product: The empty chat screen is the first thing people see. Design it to set expectations, offer a starting point and build trust from the first message. (AI in practice, 2026-11-12)
- Fast AI models compared: Flash, Haiku and GPT: Gemini Flash, Claude Haiku 5.5 and GPT-6 Luna and Terra compared on price, speed and quality for in-app AI features and high-volume work, as of November 2026. (Models and tools, 2026-11-12)
- How a small studio builds a brand system: A brand system for a small team should fit on a few pages and live in code. Build one from a clear idea, a handful of tokens and a few rules. (Design, 2026-11-12)
- How to hire and structure around design builders: How teams can hire design builders, set up review and ownership, and decide what still needs a dedicated human designer as AI agents change the work. (Design builders, 2026-11-12)
- How to review AI-generated UI for slop: AI-generated interfaces share the same flaws: generic gradients, emoji labels and glow. How to spot slop in review and fix it with real design choices. (AI in practice, 2026-11-12)
- Motion that explains instead of decorates: Good motion answers a question: where did that come from, what changed, what happens next. A practical test for which animations earn their place. (Design, 2026-11-12)
- Onboarding that earns the paywall: A paywall works best after a person has felt the value of your app. Shape onboarding so the ask arrives once the product has proved itself. (Product design, 2026-11-12)
- Prompting is product design: A system prompt shapes tone, limits and behavior for every user. Treat it as product design: write it with intent, test it with real cases and revise it. (AI in practice, 2026-11-12)
- Shipping small with a two-week loop: A two-week loop keeps a small product moving: pick one bet, build it, ship it, look at what happened. How we run it, day by day. (Product design, 2026-11-12)
- The AI model metrics that actually matter: Benchmarks, latency, cost per million tokens, context window and tool use: what each AI model metric tells a product team, and how to run your own small eval. (Models and tools, 2026-11-12)
- The first skills a design builder should learn: Four skills give designers the fastest start building with AI agents: reading code, using git, prompting well, and training taste so it stays the bottleneck. (Design builders, 2026-11-12)
- Type and restraint on a marketing site: A marketing site lives or dies on its type. Here are the rules we use: one family, a short scale, a fixed measure, and the discipline to stop adding things. (Design, 2026-11-12)
- What a design builder’s week looks like: A realistic look at how a design builder spends the week: shaping problems, building in the repo, testing on devices and reviewing with engineers. (Design builders, 2026-11-12)