Startup Spotlight: MagicPath, Pietro Schirano's shared AI-native canvas for human and agent designers

Why it matters

MagicPath shows the current early-stage pattern: visibility can arrive before verifiable company facts. Until the team, product, and financing are clear, the signal is interest, not validation.

For years, software teams have worked inside a familiar triangle. A product manager describes the thing. A designer makes it visible. An engineer translates the idea into code, then sends it back through the loop when reality does not quite match the mockup.

MagicPath is betting that triangle is starting to collapse.

https://www.youtube.com/watch?v=f_I9e0GqjUo

The company, founded by Pietro Schirano, is building an AI design workspace where people and agents work on the same visual canvas. Not a chat box that spits out a mockup. Not a static design file waiting for engineering handoff. A live, multiplayer surface where Codex, Claude Code, Cursor, designers and engineers can all touch the same product idea before it hardens into software.

That pitch moved from interesting to hard to ignore this week when Schirano said MagicPath had become an official plugin for Codex, in collaboration with OpenAI. The launch spread fast enough that Digg's aggregation of the X conversation showed 165,000 cluster views, 1,200 bookmarks, 113 comments and a follow-up from Schirano saying Codex usage was "absolutely through the roof" and that the team had underestimated the traffic the launch would drive.

The timing could not have been better. Three days earlier, OpenAI said more than 5 million people now use Codex each week and that non-developers, including analysts, marketers, operators, designers, researchers, investors and bankers, make up about 20% of overall Codex users while growing more than three times as fast as developers. OpenAI also introduced role-specific plugins, Sites and annotations meant to make Codex useful across more kinds of work, including product design and creative production.

MagicPath's question is simple: if AI agents are going to help build software, where do they see what they are building?

Its answer is the canvas.

MagicPath describes itself as a multiplayer design canvas where teams and agents work side by side. Users can generate functional apps and websites, explore variants, edit visually, import from Figma, work with design systems, export code and share interactive browser previews. The more developer-facing promise is that external agents such as Claude Code, Codex and Cursor can connect into the same MagicPath workspace, giving agents a visual environment instead of forcing everything through a terminal or chat thread.

That makes MagicPath easy to misunderstand. The lazy comparison is "Figma with AI." The more interesting comparison is "an operating room for agentic product work." The product is not just trying to generate screens. It is trying to become the place where prompts, pixels, components, code and agents negotiate with each other in public.

The homepage now makes an even louder claim. MagicPath says it is "Trusted by teams at" Google, Stripe, OpenAI, Netflix, Mercury, Webflow, LinkedIn, Vercel, Shopify, Uber, Perplexity, Rippling, Klarna and DuckDuckGo. The phrase matters. "Teams at" does not confirm enterprise-wide contracts, paid deployments or procurement relationships. MagicPath has not disclosed contract values, seat counts, usage by company or customer case studies. But the logo wall does change the burden of proof. It suggests MagicPath has moved beyond founder hype and into the daily experimentation layer of people inside some of the most design-conscious and engineering-heavy companies in technology.

That is the kind of signal investors tend to notice, and MagicPath already has them. Schirano publicly announced a $6.6 million seed round led by Khosla Ventures and Abstract, while thanking Village Global and Tara Tan as early backers. Village's Max Kilberg separately said Village led MagicPath's pre-seed round last summer. A current MagicPath job listing goes further, saying the company has raised an $8.6 million seed round led by Khosla Ventures and Abstract VC, is used by thousands of designers and engineers, and has a team of 10.

That discrepancy is worth watching. The public record now contains three distinct funding signals: Village led the pre-seed, Schirano announced a $6.6 million seed, and MagicPath's own hiring copy says $8.6 million. The most likely explanation is that the larger figure includes additional capital, a seed extension or pre-seed-plus-seed total. But until the company clarifies it, the exact financing picture remains partly unresolved.

Schirano is not a random founder who discovered AI design last quarter. Replit's builder profile describes a career that began in civil engineering before moving into design roles at OpenTable, Meta, Uber and Brex. After ChatGPT launched, Schirano told Replit he had "an epiphany that the world was never going to be the same," started testing AI tools, and shared an internal prototype at Brex that helped land him an AI lead role there.

His career since then has had a pattern: build fast, post the working demo, let the internet stress-test it, then turn the reaction into the next product loop. Replit said his GPT-generated Pong post received 6 million views, that Zing reached 50,000 users within a week, and that DesignerGPT later became part of his broader run of viral AI products. The same profile noted that DesignerGPT's appearance among trending GPT Store apps was visible in OpenAI's launch materials and that Schirano's tweet about the plugin drew 1.5 million views in its first week.

MagicPath looks like the most serious expression of that background. Schirano has been a designer, a product builder, an AI engineer, a prototype showman and a founder who understands the internet as a distribution channel. More importantly, he seems to understand the emotional problem inside AI design tools: they can make something impressive in seconds, then fall apart when a team tries to apply taste, consistency and production constraints.

MagicPath is trying to make that second step less painful. Its Figma Connect materials say users can copy and paste Figma designs into MagicPath, turn them into interactive prototypes, edit them with AI using a design system, share them as links and export production-ready code. Its public product posts frame the tool as a way to preserve craft while moving faster, not as a way to replace craft with a prompt.

The market is moving in the same direction. Cursor's Design Mode update, released June 5, starts from the premise that "UI work tends to be spatial" and lets users click elements, draw on the page or describe changes by voice so an agent can edit code with visual context. Figma, meanwhile, is pushing from the opposite side of the workflow. Its new Figma Make capabilities let users connect a GitHub repository, edit a running app with real data, annotate elements, describe changes in chat, then create branches and pull requests without leaving Figma.

That competitive context cuts both ways. It validates MagicPath's thesis that software creation is moving toward visual, agentic collaboration. It also means MagicPath is walking into the path of three giants. Figma owns the design file. Cursor owns the developer environment. OpenAI owns Codex distribution. Each can absorb pieces of MagicPath's pitch.

MagicPath's opening is neutrality. It can be the shared room, not the incumbent's room. A designer can bring taste and visual intent. An engineer can bring the repo. Codex, Claude Code or Cursor can bring agency. The canvas becomes the place where all of them meet.

That is why the Codex plugin matters. It is not just another integration badge. It gives MagicPath a shot at becoming a native surface inside the agent workflow at the exact moment Codex is trying to move beyond developers and into broader knowledge work. OpenAI's own framing now treats Codex as a tool for creating reports, spreadsheets, presentations, contracts, dashboards, lightweight tools and other work products across business roles.

The risk is that viral AI products can look inevitable for a weekend and ordinary by the next billing cycle. Logo walls can signal curiosity, not commitment. A "trusted by teams at" strip can mean passionate internal users rather than formal customer relationships. And every serious product team has learned to distinguish a magical demo from a system that survives messy repos, custom components, security constraints, design debt and real deadlines.

But MagicPath has earned a closer look because it sits at the overlap of several live shifts: coding agents becoming mainstream, design tools becoming code-aware, developers demanding visual context, and non-developers increasingly using AI systems to produce software-shaped work.

The old software handoff was built around translation. MagicPath's bet is that translation is becoming unnecessary. The product manager, designer, engineer and agent do not need separate rooms if the work itself can live on a shared surface.

That is the bigger story behind the viral plugin. MagicPath is not just asking whether AI can design screens. It is asking whether the next generation of software work will happen inside a canvas where people and machines build in the open, side by side, until the line between design and code starts to disappear.