Your Team's First Draft Is Now a Prompt

AI stopped being the thing you bolt on at the end. It moved into the first ten minutes of the work: the ideation, the rough spec, the logo sketch, the interview loop. The tools are named, they are shipping, and your team is probably already testing them without telling you. Let me catch you up on what actually changed.

Starting from what already works

When your designers open a blank canvas, they burn time reinventing patterns that shipped years ago. Mobbin now has an MCP that plugs your AI agents into more than 600,000 real product screens. So what an agent builds starts from proven layouts, not from a guess.

That is a real shift in where ideation begins. Instead of prompting a model cold and hoping the output looks right, you point it at a catalog of shipped work. The tradeoff is obvious once you say it out loud: your team leans on what exists, which is faster, but it also nudges everyone toward the same familiar patterns. Speed up front, sameness if you are not careful.

Claude as the art director, not the artist

The smart move with AI visuals is to treat the model as the brief writer, not the image maker. Faheem Munshi frames Claude as the creative director and Canva as the production team. Claude plans the strategy, writes the brief, sets the color codes and copy. The image tools just execute.

His old cost was $350 a month for a part-time designer. Now he runs a fixed brand brief through every asset, so consistency is a system output instead of a discipline problem. Nick Babich took the same approach for a logo, importing Microsoft's Fluent design system into Claude Design first, then asking for the mark. Both start with rules, then generate. The lesson for your team: the quality lives in the brief, not the prompt.

The interview now asks candidates to build

Hiring is catching up fast. Aakash Gupta tracked vibe coding rounds spreading from a few first movers last year to Figma, Shopify, Perplexity, Adobe, Microsoft, and Meta. In one reported case, Meta added a Product Sense with AI round: 30 minutes of product sense, then 30 minutes of vibe coding. Candidates open Cursor and build the idea they just pitched.

What the hiring managers watch for is not the code. Jaclyn Konzelmann at Google AI said jumping straight into execution worries her, because knowing what to build matters more. Jyothi Nookula at DataRobot said she watches how a candidate shepherds the AI, whether they catch edge cases or just wrap up the first thing the model spits out. The screen share strips away the LinkedIn résumé and shows the real level.

AI moved into the doc and the desktop

The ideation is bleeding into tools your team already opens every day. Google Docs now lets Gemini generate and edit images inside the document, pulling diagrams and infographics from your own text. Ask for a diagram at the top of the proposal, and it reads the proposal to build one. That is for AI Pro and Ultra subscribers and Workspace plans.

Gemini also reaches past the app now. Screen-aware reasoning lets it see your Mac's display and act on what is open, turning notes into a summary or drafting an email from files you name. Handy, and risky. The article flags prompt injection, where a hidden command on a website reads like it came from you. If your team turns this on, they are handing an AI a view of everything on screen.

The deep cut

The fastest way in is not a bigger model. It is a reusable brief. Munshi's brand brief and Babich's imported design system do the same job the vibe coding interviewers are grading for: they define what good looks like before the AI runs. Nookula watches whether a PM catches the edge case the model missed. Konzelmann worries about people who skip straight to building. Same skill, two sides of the desk.

So the thing to build this quarter is not an AI tool license. It is the brief your team runs everything through: the design system, the brand rules, the product judgment about what to make. The models are cheap and getting cheaper. The judgment that steers them is what you are actually hiring and training for. Ship the brief first, then let the tools execute.

Three questions for your team

• Do we have a written brief, design system, or brand rulebook that an AI agent could start from today, or would every prompt start cold?
• If we added a build-it-live step to our PM or design interviews, what would we actually grade: the output, or how the person steers the AI and catches what it misses?
• Before anyone turns on screen-aware AI features, who owns the call on the prompt injection risk of letting a model see everything on a work machine?