LinkedIn shut down their free extension. Four people rebuilt it in 4 weeks and hit $62,000 MRR three months after relaunch

(Dollar figures include a rough conversion at ¥150 to $1.)

Right after gathering 60,000 users as a free Chrome extension, a cease-and-desist letter arrives from the platform. That normally looks like the end of a business. Kleo’s four co-founders instead rebuilt it from zero in four weeks — this time not free, but a subscription SaaS starting at $59/month.

Relaunch was January 2026. Three months later, MRR hit $62,000. A sister product, Mentions, added $20,000, bringing the combined total to $82,000. Zero ad spend. No Product Hunt launch either.

What happened, in sequence

Who’s building this

The team is four people. CTO Cameron Trew has over a decade of engineering experience, having been a senior engineer at Vonage, Base, and Elastic Path. The other three (Jake Ward, Lara Acosta, and Rob Hoffman) are LinkedIn creators, with Ward at over 180,000 followers and Acosta at over 300,000. Combined, the four have over 480,000 followers.

Kleo is an AI writing assistant for LinkedIn, offering text generation via the Claude API, image and document analysis via Claude Vision, a memory feature that learns the user’s voice, voice input through Deepgram, and post-performance tracking. The sister product, Mentions, tracks how a brand shows up in AI answers from tools like ChatGPT and Perplexity. The stack includes Next.js, TypeScript, Vercel, Neon (Postgres), Inngest, Clerk, PostHog, and Langfuse.

The tide turned at the moment the cease-and-desist arrived

The turning point here is clear-cut, and at first glance it looks like a disaster: LinkedIn’s cease-and-desist.

Kleo 1.0 was a Chrome extension that operated by parasitizing LinkedIn’s own interface. The 60,000-user figure was only achievable in that form, and simultaneously, that same form carried the structural risk that it could vanish at the platform’s sole discretion. And being free, those 60,000 were users, not customers.

The cease-and-desist forced both weaknesses to resolve at once. Kleo 2.0, rebuilt, is a standalone application not dependent on LinkedIn’s DOM, and it charges from day one. The before/after shows up directly in the numbers. 60,000 users, zero revenue → $62,000 MRR in three months. User count presumably plummeted, but revenue moved from nothing to nearly $9M/year run rate.

The founders’ own stance is telling too: ‘Ship fast, ask questions later. We had a working version in four weeks. It wasn’t perfect.’ The four-week timeline is the flip side of a decision to let go of polish. There’s also a comment worth flagging directly: ‘AI code editors were absolutely our biggest weapon’. This speed clearly rests on a premise different from equivalent products of the early 2020s.

Breaking down why it grew

Chalking up the $62,000 to ‘480,000 followers’ misses the reproducible parts. At least three layers were at work.

The audience layer: it functioned as proof, not just a distribution channel. Ward, Acosta, and Hoffman are people who actually write posts that perform well on LinkedIn. Those three selling a tool they use for their own writing means the product demo runs continuously, in the open, on the daily timeline. ‘We built it because we use it ourselves. We never sold anything we didn’t believe in’ is a statement of principle and also a description of this structure. That’s a different quality of information than a large-follower account simply running an ad.

The pricing layer: the ladder manufactured a reason to buy now. The $59 (500 seats) → $79 (500 seats) → $99 design locked in both scarcity of seats and the certainty that waiting means paying more. The 4-day and 9-day sellout speeds are simultaneously a demand signal and a product of that deadline design. Demand validation came before building, too. Acosta ran three pre-launch webinars, each generating over $5,000. They had a sense of price and volume before the product even existed.

The customer layer: they closed the gap to early customers to an extreme degree. Feedback was gathered in a private Slack community. Bugs were fixed within hours and reported back individually to the person who flagged them. Priority was fixed as bugs → revenue-critical features → user requests, in that order. Actual user behavior led to simplifying the UI after finding the initial setup form too complex. Bringing 500 paid beta users in all at once only works with this density of response, otherwise it would have been a pile of complaints.

Risks and what isn’t disclosed

Behind the strong numbers, a few things need careful reading.

For one, LinkedIn dependency hasn’t disappeared in 2.0. They only stopped being parasitic, as a tool for producing LinkedIn posts, changes to the platform’s rules or specs still flow straight into the business. What happened with 1.0 could recur in a different shape.

For another, churn isn’t disclosed. The three-month MRR almost certainly includes a substantial share of purchases from existing followers thinking ‘if these people made it, I’ll buy it’, whether that holds past the renewal point can’t be confirmed at this stage. A 480,000-person audience is a powerful accelerant for initial velocity, but that pool is finite. Getting from $62,000 to $300,000 will require reaching beyond the follower base.

And the founders themselves state they’re targeting a sale around 18 months out. This reads less as a lack of confidence and more as a design that prices in how fast AI tools tend to become obsolete. If a foundation model provider bundles equivalent functionality natively, this category of tool’s edge could vanish quickly. Not planning to hold long-term is itself an expression of risk awareness.

What’s replicable, what isn’t

The replicable elements are concrete: give up on polish and ship something working in four weeks. Skip the free trial and charge from day one. Create deadlines through seat counts and price tiers. Gather paying customers in a private chat, fix bugs within hours, and tell the affected person directly. Fix your fix-priority order in advance. And build something you use yourself every day. This is also a way of absorbing quality-control cost into daily work.

What can’t be replicated is the 480,000 followers. That’s something the four of them built separately over years, on a completely different timescale than the four weeks it took to build the product. More fundamentally: being able to feel confident, the moment the cease-and-desist arrived, that ‘if we rebuild, it’ll sell’ is itself something this audience supported. A team without an audience facing the same situation would be making a very different bet with those four weeks.

One more condition shouldn’t be overlooked: having an engineer like Trew with 10+ years of hands-on experience on the team. AI code editors deliver speed when used by someone who can evaluate the output, not the other way around.

As a contrast in launching a product by leveraging an existing audience, see Photo AI’s levelsio, as a contrast in building without an audience over the long haul, see Carrd. Even within SaaS, the design differs sharply depending on whether you’re betting on initial velocity or on durability.

Related reading

• Photo AI — the launch speed of a solo developer with an existing audience
• Carrd — how a solo-run SaaS reached $2M ARR