Magic Patterns CEO Reviews Startup Websites
In a Nutshell
Magic Patterns CEO Alex Danalovich reveals their AI prototyping tool evolved from failed YC pivots, emphasizing that finding 10 users who truly love your product is the only validation that matters. The interview critiques multiple startup landing pages for common AI-generated design flaws: non-functional CTAs, vague value propositions, scroll-jacking animations, and overly generalized messaging that fails to convert. Key advice includes prioritizing user action over visual flair, testing with actual users rather than founders, and building brand systems early instead of defaulting to AI-generated trends.
These notes were generated by AI and may contain inaccuracies.
Magic Patterns is an AI design tool that takes users from an idea to a real interactive code-backed prototype in minutes. The tool serves product managers, founders, and professional designers to visually communicate their ideas through prototypes, interactive mockups, or code-ready exports for production.
Alex Danalovich and his co-founder are both front-end engineers by trade with no formal design background. When they entered their YC batch, their YC partner advised them to build something they knew really well. For them, that was translating Figma mockups into code. This insight became the foundation for Magic Patterns.
The company was one of the first prompt-to-design tools when it launched in October 2023. At that time, the only other similar tool on the market was Vzero. The models were so primitive that there was no chat functionality - it was literally just prompt, get a design, with no actual interaction with the agent.
The co-founders originally applied to YC with a text message analyzer that identified top words, funniest messages, and other insights across group chats. The product had revenue, which they believe is why they got into YC. However, their YC partner told them the idea was okay but recommended they pivot to something they knew really well - front-end engineering and implementing Figma mockups by code.
During their 9-month pivot journey, they explored several ideas including a component library editor, something described as "storybook on steroids," and a Chrome extension that allowed localhost to be instantly published on the internet as a feedback tool. The Chrome extension feature later became part of Magic Patterns.
The key learning from these pivots was Paul Graham's advice: find 10 users who really love you. None of the earlier pivots had 10 users who were paying and actively using the product. Once they launched what became Magic Patterns, those 10 users came very quickly.
The approach to staying ahead is described as making "customers our religion" - talking to them every single day and adapting to new workflows. An example given is that they released an MCP about a year ago as part of an internal hackathon project. They didn't think it would go anywhere, but MCP usage has now completely exploded and is one of the main ways people interact with Magic Patterns.
The philosophy is to constantly be "rebirthing your company" to adjust to new workflows, especially in the AI world.
The team is "obsessed with dog fooding" their product but never forces it, because forced dog fooding doesn't yield authentic usage patterns. They only use it when it's genuinely the best tool for the job.
They've been using Magic Patterns to create their own loading indicator, drawing inspiration from Wall-E's Eve character and Portal video game aesthetics. The system features global chat functionality that allows multiple updates to be made simultaneously.
Through the MCP integration with Cursor, users can prompt directly from their coding environment. Enterprise customers particularly value this because they can leverage extensive context from PRDs and meeting notes already present in their Cursor or Claude Code environments, then instantly put that context onto a visual canvas for better communication.
The canvas is described as an "artboard" where users can bring in context including images, PRDs, or documentation. A key differentiator is that users aren't blocked by a single thread - historically, design and prototyping tools required waiting for the AI agent to finish, but Magic Patterns supports many threads simultaneously.
The canvas is called "Magic OS" because it's positioned as an operating system for all of a user's context, not a canvas in the traditional Figma sense with tools on one side.
The team has had a canvas for roughly two years, with the demonstrated version being "canvas 2.0." One issue they've encountered is the initial onboarding experience and progressive disclosure around when it makes sense to use a canvas versus a single chat thread. The learning has been that a canvas is not necessarily for everyone.
The team hosts Magic Patterns meetups worldwide. Alex has been creating postcards from these events using the tool. The system can reference existing context on the canvas - images and descriptions from events in Colombia, Singapore, New York, and Toronto - and automatically generate designs based on that context.
Users can select specific elements and apply changes across all instances at once. For example, after generating postcards for multiple cities, changes made to the Singapore postcard can be applied to all cities simultaneously.
The tool is used by everyone from the ops team to the sales team. The go-to-market person creates postcards in Magic Patterns to send to users - someone who historically didn't know how to use Figma and would have had to outsource this type of work.
Magic Patterns is model agnostic and feature agnostic. Users are encouraged to export to Figma for high-fidelity precision or export to whichever coding agent they prefer. This differs from competitors like Claude Design, which only works with Anthropic models and only exports to Claude Code with no Figma integration.
The advice for improving prompts is to leverage the features tools offer. AI enablement teams at large companies are using features like skills agents and bringing in unique company context. Magic Patterns has a skills feature where users can add their skills or design system components. The recommendation is to spend 10 minutes bringing in existing context, which saves hours in the long run.
One notable creation is a Roller Coaster Tycoon-style chocolate theme park built entirely in the browser using WebGL rendering. The system used a design system called "3D theme park" with components for tracks, rides, and paths. When given proper guardrails through components, the AI can output a theme park in one or two shots.
This demonstrates that design has historically been 2D, but the future involves designing full-on interactions that are high fidelity using components.
The first website reviewed is for Ego, described as "an applied AI research lab building personal agents indistinguishable from humans." The tagline reads "Agents are machines. We make them people."
The main issue identified is that the real metric for landing pages is conversion, and the site lacks clear direction on what action to take. The CTA appears to be "Ronin" but it's unclear what Ronin is - whether it's a product, model, or company feature. The site mentions "Ronin launching September" but doesn't clarify the relationship to Ego.
The reviewers note that having an input form visible and expanded directly on the page yields much higher conversion rates than requiring users to click through to access a form. This was validated through testing at Creative Market.
The core value proposition - "a voice agent with a personality, a memory, and a mind of its own" - appears halfway down the page when it should be at the top. The site requires significant scrolling before users understand what the product is and who it's for.
The site features overly animated elements including blinking dots and what appears to be music faders. There's a non-functional input box that users can interact with visually but cannot actually type into, creating a misleading experience.
The "scroll" indicator on the page is identified as a vibe coding smell, as humans naturally scroll without prompting. The eyeball graphic at the top is provocative but the overall experience leaves users confused about the product's purpose and the required action.
The non-functional input creates distraction from the core CTA of joining the waitlist. The site's dark mode red aesthetic evokes 1980s video games, which could be memorable, but the execution undermines the user experience.
Sentient OS is described as "Your Mac sentient" - an on-device LLM that understands your entire life and proactively offers to get your work done through computer use. The site features a laptop that opens and closes based on scroll position. The animation shows a laptop screen open with elements flowing into it, then the lid closes with text stating "While you sleep, your Mac understands your entire life." When scrolling further, the lid opens again to reveal completed work.
The reviewer notes that the animation effectively reinforces the product concept - the computer is asleep when the lid closes, work happens during sleep, and the laptop opens to show finished work when the user wakes. This is considered effective animation because it communicates something that static content could not convey.
The reviewer strongly advises against scroll jacking, noting that decacorn companies never use scroll tracking because it does not convert and is frustrating for users. When scrolling precision differs from user expectations, it creates an unnatural experience where scrolling continues after the user stops touching the mouse or trackpad.
The reviewer emphasizes that the primary goal of a landing page should be getting users to click the main CTA button, such as "Download for Mac OS." The glowing gradient and massive drop shadow on the download button are identified as "vibe coding smells" - attention-grabbing but potentially overdone. The homebrew installation command is noted as being too small to read easily at 100% zoom.
For products that run locally on user machines, privacy and trust are critical. The reviewer notes that AI apps and LLMs need significant data and context to be valuable, creating a new standard for trust, privacy, and security. Products that proactively address data access, usage, retention policies, and security measures are more likely to succeed in getting users to download and recommend them.
When building a text message analyzer that reads messages locally, the key trust-building elements included prominently featuring "open source" and "completely on device" messaging on the hero section, emphasizing that the internet does not touch the data in any way.
Random bolding of words is identified as a sign of AI-generated writing. The reviewer also notes that fonts like "Instrument Serif" and "Space Grotesque" are commonly used in AI-generated designs, which may eventually date sites as characteristic of mid-2020s design.
Founders are advised to establish a brand book early, including decisions on fonts and design systems, rather than defaulting to AI-generated outputs. This creates more timeless design rather than following trends that may quickly become outdated.
Saranovas is described as "your AI operating partner" that "reads every record your company produces, finds the operational inefficiencies nobody's tracking, and warns you while problems are still small." The reviewer finds the copy too generalized and vague, making it unclear what the product actually does and who should use it.
The hero section contains excessive white space and small all-caps text listing claims like "every claim cited," "years of history," "live from day one," "nothing installed," and "no interviews." The reviewer notes that users still don't understand what the product does after reading these claims.
The site mentions targeting "enterprise, middle market, private equity" which focuses on who the company targets rather than the user's problems. Users shouldn't have to identify which category they belong to; instead, messaging should focus on specific problems and solutions.
Vibe-coded sites often read like investor pitch decks rather than user-focused landing pages. The recommended approach is to speak directly to users about who should use the product, why it's valuable, what problem it solves, and include social proof from similar users.
The presence of a "trust and security" link in the main navigation is noted as unusual but potentially becoming more common for AI sites handling sensitive information.
Founders should test websites with actual users without revealing it's their site, as users tend to be too polite when they know the owner. User feedback is described as "the ultimate test" for website effectiveness.
Slashy is an intelligent inbox product that "drafts replies in your voice, schedules meetings, and handles follow-ups." The logo is noted as potentially confusing, as it might be read as "Sashy" without context.
The demo is large but contains too much text, making it difficult to understand what to take away. The reviewer suggests an animation might work better than a screen-recorded video for demonstrating the product.
Tabs that cycle automatically on a timer are problematic because users may miss content behind other tabs. The reviewer notes that timers might start before users actually see the content, affecting what users experience.
Social proof featuring investors and founders may inadvertently signal that the product is only for VCs rather than general consumers, even if the product is intended for broader use.
The product demonstrates the ability to make edits that match the user's writing style, which the reviewer finds impressive and resonant.
The ability to handle multiple inboxes (text messages, WhatsApp, Gmail, Slack) from one interface is identified as a particularly compelling feature that resonates with users.
Long landing pages can communicate that a company has been around for a while and is substantial, similar to how PostHog uses a massive footer to communicate company size.
To improve design taste, the reviewer recommends looking at beautiful things, visiting exhibits like the Matisse exhibit in San Francisco, consuming aesthetically pleasing content, and reading "The Design of Everyday Things." Design is defined as thinking through user experience rather than just pixel-perfect border radius.
At Magic Patterns, every team member handles customer support daily, experiencing user frustrations firsthand. This leads to immediate fixes when engineers encounter problems. The team maintains a high bar for quality, encouraging fixes for anything that "feels off" and spending extra time beyond AI-generated outputs.
While taking time for excellent design may slow shipping, the alternative risks users leaving due to confusion or blocked workflows. The reviewer notes that with modern development speed, the historical trade-off between quality and speed has diminished.
Magic OS was released as a new canvas version described as "Pinterest and Figma had a baby," allowing users to chat and prompt designs while incorporating PRDs and context. The future of design involves code as a medium, with frontier labs focusing on coding agents as the ultimate production medium for users.
Magic Patterns enables new tools to build things that weren't possible before. Everything is being taken to a new level with new things being possible that almost seemed like they never would be.
The conversation wraps up with appreciation for having the Magic Patterns team on the show to demonstrate how they are enabling this advancement.
The host thanks the guest for joining the discussion.
This concludes the current episode of Design Review. The next episode will be available in the following installment.
Keep Y Combinator in your library
Save the videos and channels worth coming back to, and find them again in one place.
