Glitter AI: How Yuval Karmi Went From $15.5M and 21 Employees to Zero → and Built a Better Company
Yuval Karmi raised $15.5 million, hired 21 people, got acquired by WalkMe — and calls it a failure.
Not publicly, not for drama. Just quietly, in a blog post on his own product’s site, where he admits that he was a perfectionist who couldn’t let go of anything, held every decision like a referendum on his personal worth, and watched his company buckle under the weight of it. The acquisition happened. The money arrived. The lesson cost more.
Glitter AI is that lesson, compressed into software. And in a startup world where founders constantly pitch you on what problem they set out to solve, Yuval is one of the rare cases where you actually believe them.
The Problem He Couldn’t Find a Solution To
The workflow that destroyed him at Simpo was simple to describe, impossible to fix: you want to delegate, but first you have to document. Documentation takes so long it never happens. Nothing gets delegated. The founder stays the bottleneck.
Every company has this. The process that lives in one person’s head. The onboarding done over a Zoom call that evaporates the moment the recording stops being watched. The SOP Google Doc that was last touched eighteen months ago and is already wrong.
The existing tools all solve the wrong half of the problem. Screen recording gives you a video — which no one can skim, search, or hand off cleanly. Written guides take time that people don’t have. Browser extension tools that auto-screenshot your clicks help, but only if you’re working in a browser.
Glitter’s core insight is different: documentation shouldn’t require any deliberate act of creation. It should happen as a side effect of doing the work.
What It Actually Does
You install the desktop app — Mac or Windows — and record yourself performing any process on your computer. Not just in the browser. Any software: ERP systems, legacy enterprise tools, anything on screen. As you click, Glitter captures screenshots. As you speak, ElevenLabs converts your voice to text in real time. GPT-4o assembles all of it into a structured step-by-step guide.
When you’re done, you have a document. Not a video. A navigable, searchable, shareable guide with annotated screenshots and written instructions that someone else can follow without you in the room.
There’s also a video-to-guide converter that runs the same process in reverse: upload an existing recording, Gemini AI analyzes it, and Glitter places steps at the correct timestamps. That feature came from a single customer request — and then became the differentiator that won at least one enterprise contract.
The full export stack supports PDF, HTML, Markdown, SCORM for LMS platforms like Workday and Moodle. SAML SSO for Okta, Azure AD, Google Workspace. Real-time collaborative editing. Guide translation across 99 languages, including full right-to-left support for Hebrew and Arabic — added in June 2024, months into building.
SOC 2 Type II certification landed in October 2025. One person got that audit done.
One Person
This is the part of the story worth sitting with.
Glitter AI has zero employees besides Yuval. Zero outside investment — intentionally. He started writing code in September 2023. Incorporated in January 2024. By August 2025, he was speaking at a Google for Startups Israel event describing a product used by thousands of companies worldwide.
Twenty-nine months from first commit to enterprise-certified platform. Solo.
The way he runs it is its own kind of proof of concept. Customer feedback from in-app chat, email, and WhatsApp gets piped through an AI workflow that converts messages into structured JSON, aggregates feature requests in Notion by frequency, and surfaces the most-requested item. That item gets built next. No PM, no sprint planning, no quarterly roadmap meetings.
Support works on a 10-second loop: a message comes in, an AI drafts a response using a knowledge base of past answers, flags it to Yuval in Slack. He reads it, approves or tweaks, the AI sends the reply. Human judgment stays in the chain; human time gets protected.
He is, essentially, running the company on the same principle Glitter sells to its customers: document the process, then let the system handle it.
What Simpo Actually Taught Him
After Simpo was acquired by WalkMe in 2021, the standard version of the story would’ve been: successful exit, lessons learned, onto the next thing. Yuval’s version is harder on himself.
He raised $15.5 million. He describes it as a failure. Not because the money ran out or the product flopped — but because of the internal dysfunction he allowed to calcify. A founder who needed to review everything, couldn’t trust his team, and as a consequence never built the documentation and process systems that would’ve let him actually delegate. Twenty-one people, and he was still a bottleneck.
The correct sequence, he came to understand, was: document first, then delegate. But documentation was always the thing that got skipped because it took too long to create.
He built Glitter to fix that sequence. It’s not a spiritual lesson turned into a mission statement — it’s a founder building the specific tool he needed three years earlier and couldn’t find.
The Corridor
Glitter AI, Inc. is incorporated in the US. Yuval studied at UC Davis. Simpo was built across San Francisco and Tel Aviv. The customer base is global, skewing toward American enterprise buyers.
This is the standard operating model for Israeli B2B software: built in Tel Aviv, sold to the US market, legally structured for US investors and acquirers even when none exist yet. Yuval runs it bootstrapped and solo, which means the corridor runs in one direction — outbound, toward customers — with no investors on the other end pulling the company toward a specific exit.
Before Simpo, Yuval founded Social Tel Aviv, a community organization for newcomers building life in Israel. He built the web infrastructure for Mishelanu, a social empowerment project led by Intel VP David Perlmutter. After October 7, he was among the Israeli founders who mobilized for relief efforts — logistics, supply collection, coordination. That context sits in the background of the Glitter build sprint, a period when a lot of Israeli founders decided to build harder.
Final word
Glitter isn’t in the news. It didn’t close a round. Nobody announced it. It’s just a Tel Aviv founder, a two-year changelog, and thousands of paying customers — built in the gap between one expensive lesson and the next idea. Those are usually the most honest companies to watch.
That’s worth documenting.
💜