Louis from Vibe Kanban - 20,000 GitHub stars and walking away from 6-figure deals
Vibe Kanban cofounder Louis Knight-Webb explains how four years of pivots led to 20,000 GitHub stars, why AI-generated pull requests are useful, and why the company walked away from six-figure enterprise work to return to founder-market fit.
Vibe Kanban is an open-source interface for orchestrating coding agents: planning work, running several agents and bringing a human back at the moment review is needed. Ten YC founders used its first version for as many as 14 hours a day—the clearest demand Louis Knight-Webb’s team had seen across several products.
That success followed code search, AI application work and a legacy-modernization service with long enterprise sales cycles. Louis treats those years as “startup university”: an education in timing, real community, honest experiments and knowing when a promising market is wrong for the founders pursuing it.
The Q&A below is experimental — extracted from the episode transcript with AI. If you spot a mistake, let us know at jack@scalingdevtools.com.
How can a startup build when an AI lab may copy the feature?
Vibe Kanban uses a three-month test rather than a twelve-month one: avoid work Claude Code is likely to ship imminently, but do not reject every idea that a lab might eventually build. A startup can occupy the useful space between obvious customer need and the lab’s immediate priorities. Applying an infinite horizon makes every product look futile; a practical horizon leaves room to build, learn and establish distribution.
Why have the AI labs not already killed products like Lovable?
A focused founder can make one workflow complete—domains, onboarding, community and all—while a product manager inside a platform is one of many teams serving a much broader audience. Labs also possess enough distribution to arrive later, so winning every niche early may not matter to them. Startups have the opposite incentive: move first, execute tightly and create escape velocity before an incumbent decides the category is strategic.
How did Bloop turn into Vibe Kanban?
The company began in 2021 training embedding models for code search, moved up the stack after ChatGPT, then spent roughly a year modernizing legacy code for enterprises. Sales cycles threatened to outlast its runway. When Claude Code appeared, the team built an internal interface for running more than one agent and reducing the cognitive burden of switching between planning and review. Ten founders using it almost continuously in its first week justified an immediate pivot and an open-source launch.
What did genuine product-market pull feel like?
It was qualitatively different from stars, compliments or downloads. The earlier code-search project had visibility but little contribution or Discord activity; Vibe Kanban quickly attracted thousands of community members, daily use and small quality-of-life pull requests. Louis compares it to catching a fish: before it happens, weak signals can be interpreted optimistically; when it happens, there is no need to ask whether users care.
What is the roadmap once demand is obvious?
The backlog becomes clearer and the problem shifts toward execution, hiring and resource allocation. One major step is team support: sharing agent activity and handing completed work to whichever developer is available. Louis expects orchestration to remain profession-specific because reviewing code, architectural plans or another skilled output demands different interfaces. A universal AI work surface would sacrifice the domain detail that determines human throughput.
Will coding agents remove developers from the loop?
Not soon. User feedback cannot automatically become a sound product task, and generated code must be reliable far beyond the point where it is merely useful. Crossing 50% reliability changes adoption; removing review is closer to a 99% problem. As agents work for longer stretches, humans will spend more time specifying, planning and reviewing, but those remain skilled, product-sensitive activities rather than mechanical pass-throughs.
How should developers use the time while agents run?
Vibe Kanban’s original launch asked whether developers were wasting that interval on Twitter; its first memorable notification was a cow mooing when an agent finished. The serious workflow is parallelism. While one task runs, write the next specification or review another result. Louis commonly keeps four or five agents active, moving through local QA, AI review and human code review as each completes instead of staring at one terminal.
Why build the product in the open?
The purpose is to improve the tool faster. Users who live in it can submit fixes, and the team can learn from channels spanning in-app feedback, issues, discussions, Discord, Reddit and direct email. The planned SaaS is a hosted version of the same open product. Simply exposing a repository does not create a community: maintainers have to solicit input, respond visibly and sometimes answer a complaint with a pull request users can watch ship.
Are AI-generated pull requests a burden?
Louis sees them as a better starting point than a speculative feature request. Asking an agent to make the change forces the contributor to spend tokens, run the result and often discover when the idea does not work. Vibe Kanban may retain half or less of the submitted code, but a tested, imperfect implementation is useful evidence. The maintainer’s responsibility is to allocate enough capacity to triage and finish community work rather than let scale turn every contribution into noise.
Why walk away from six-figure legacy-modernization deals?
The technology worked, but the founders did not enjoy selling transformation projects to banks and lacked lived experience of the organizational problem. Deals could take 18–24 months and disappear because of unrelated internal upheaval. Vibe Kanban returned them to developers they understood and a product they used themselves. Louis’s broader advice is to stay alive, define what success would look like for each experiment, and move decisively when the evidence says it is failing; a team spending months on a product everyone believes is dead loses both runway and morale.
Links & references
- Vibe Kanban ↗
- Vibe Kanban on GitHub ↗
- Louis Knight-Webb on LinkedIn ↗
- Y Combinator ↗
- Claude Code ↗
- Lovable ↗
This episode is brought to you by WorkOS.