Nad Chishtie: Lovable's Design System For Agents
TL;DR
Lovable’s design system is now partly written for agents, not humans — Nad Chishtie says their new system is “50% components” and “50% skill banks,” with AI skills acting more like linters for things like keyboard focus, duplicate components, PR checks, and visual diffing.
In enterprise, the real owners of vibe coding are becoming security and design — not engineering or product, because once lots of teams can ship prototypes, the immediate problems are data safety and whether everything suddenly looks off-brand or like “spaghetti.”
Lovable runs designers as true product owners inside small teams — embedded designers work directly with engineers, sometimes without PMs, and are expected to help choose problems, ship, iterate, and remove features so each five-person team feels like the company when it was five people.
The company’s growth forced a shift from YOLO shipping to controlled rollouts — Lovable now leans on beta channels including a private Slack of roughly 1,000 testers, Discord opt-ins, and an enterprise customer advisory board after seeing enterprise users put the product on personal cards before security and admin features were ready.
Nad thinks the winners in AI-native work are end-to-end generalists with gumption — the people who can go from a napkin sketch or user insight all the way to production, rather than waiting for research, data, or stakeholder validation at every step.
For hiring, Nad values self-learning and intrinsic drive more than polished tool mastery — he looks for people who teach themselves, tinker obsessively, and treat AI like a material to shape, whether that shows up in design work, side projects, or even hobbies like woodworking.
The Breakdown
The accidental designer who nearly told Lovable to become a browser
Nad opens with a great almost-disaster: before his first day, he emailed CEO Fabian a whole thesis arguing that Lovable should be a desktop-style web browser that “creates the web instead of browsing it.” Fabian basically replied, “that’s so dumb,” and Nad laughs that he’s glad he did. It sets up Nad’s whole personality: opinionated, systems-level, and always thinking a few steps ahead.
Why being a generalist finally became a superpower
He describes falling into design because programming “sucks the energy from my soul,” not because he lacked ambition but because he wanted to build without years of delayed gratification. A formative conversation with Steven Olmstead helped him realize he wasn’t an A+ specialist — he was a “B+ at everything” who excelled at zero-to-one work, balancing craft with speed, and sitting comfortably in deeply technical rooms where many designers can’t.
The AI-native worker he’s betting on
Asked who thrives now, Nad points to people with “gumption” — the ones who hit an obstacle, figure it out, and keep moving all the way from idea to production. He’s blunt that people who need every move validated by research, data, or a stakeholder chain may struggle as orgs get faster and flatter.
Democracy creates prototypes — and also spaghetti
When everyone can make things with tools like Lovable, the upside is obvious: more people can express ideas that previously died in their heads. The downside is just as obvious: people anchor to prototypes quickly, and products can turn into “literal spaghetti.” Nad says that means design can’t stay a gatekeeper; its role shifts toward enabling others while preserving comprehension so the experience still feels singular.
How Lovable actually ships ideas now
Inside Lovable, EPD is split across AI-forward work, core product experience, and growth, with designers embedded in teams and often operating without PMs. If something is reversible, they’ll usually ship once it clears the quality bar, often through partial rollouts in production because “pictures of an AI engine don’t cut it.” But they’ve moved away from chaotic daily UI changes and now use controlled channels: a private beta Slack with about 1,000 users, Discord opt-ins, and an enterprise advisory board.
The design system reboot: agents as first-class collaborators
Nad says he did a complete 180 on design systems after going the first 18 months of Lovable without one. The turning point was “agent maxing” — pushing AI hard enough to see what actually works. Background agents felt like fake productivity, but skill banks worked: AI checks that behave like linters, catching existing components, fixing keyboard focus across the app, and plugging directly into CI/CD workflows.
Lovable is building itself a stack of internal apps
The team is increasingly “eating our SaaS stack.” Nad mentions an internal agent built with Lovable that replaces chunks of product analytics and the data warehouse, and a custom org-planning tool layered on top of HR software because normal HR tools can’t model real team structures. His point is simple: problems that used to require whiteboards and sticky notes now get solved by building exactly the software they need.
What enterprises fight about — and what Nad looks for in talent
One of his sharpest observations is that when vibe coding lands in a big company, everyone fights over who owns it — engineering, marketing, product, everyone. But the two stakeholders that actually matter most are security and design. On hiring, he ends with a similarly grounded take: AI fluency matters a lot, but the deeper signal is self-learning, intrinsic drive, and people who can’t stop tinkering — the ones who love the game enough to keep teaching themselves as the landscape changes.