Louis Knight-Webb

Louis Knight-Webb co-founded Bloop with Gabriel Gordon-Hall and later built Vibe Kanban with him and the team, an interface for planning, coordinating and reviewing AI coding agents. Their work addressed a changing development bottleneck: as agents wrote more code, humans needed better ways to assign tasks and assess the results. On April 10, 2026, Knight-Webb announced Bloop’s shutdown and a planned community-maintained future for Vibe Kanban.[3][4]

Vibe Kanban launched in June 2025. In their founder launch account, Knight-Webb and Gordon-Hall described spending more of their time planning, reviewing and orchestrating agents rather than writing code themselves. The tool supported running multiple agents in parallel or in sequence, tracking their tasks and reviewing their work.[3][4]

A November 2025 lesson featuring Knight-Webb explained the workflow: each agent worked in a separate Git worktree, while built-in diffs let a human inspect, edit and approve changes before merging. Separate workspaces isolated simultaneous execution; reviewing the resulting changes remained part of the human’s job. The event identified him as Bloop’s co-founder and CEO and described the company’s AI work modernizing legacy code, including COBOL and mainframe systems into languages such as Java.[7]

In his June 1, 2025 essay, “AI Killed The AI Engineer”, Knight-Webb wrote that his preceding six months had largely involved removing components the team had previously developed. He argued that native function calling displaced application code for parsing structured outputs, while stronger models could handle code changes, test execution and debugging within one conversational thread instead of separate prompted modes.[5]

He described this as a transfer of complexity from application engineers to model labs, and of development cost to inference cost. His provocative conclusion still left room for builders: “There's still plenty to build at the next layer of abstraction”. His expectation was that this work would increasingly resemble ordinary engineering.[5]

One concrete team decision appeared in his February 26, 2026 post, “Goodbye SQLite (for logs)”. Vibe Kanban stored both structured workspace data and raw execution logs in SQLite. Knight-Webb reported that logs accounted for approximately 98% of the database and that large parallel log writes correlated with lock errors. WAL mode was incompatible with their pre-update-hook-based synchronization framework, so the team moved append-only JSONL logs to individual filesystem files, one per execution. He reported that the database shrank from gigabytes to about 20 MiB and that lock errors disappeared.[6]

In “Goodbye bloop”, Knight-Webb explained the April 2026 closure by saying that the vast majority of Vibe Kanban users were free users and the team had not found a business model it could get excited about. Signed by Louis, Gabriel and the Bloop team, the announcement said the project would continue as open source and community maintained. It scheduled remote services to end after 30 days and said local workspaces would continue functioning.[4]

Before Vibe Kanban, Knight-Webb’s AI work centered on understanding existing codebases. In his November 2023 Latent Space interview, he recalled starting Bloop with his co-founder roughly three years earlier to help engineers navigate large repositories. They initially trained their own natural-language code-search models; by the interview, those smaller models formed one part of a larger pipeline.[1]

Knight-Webb distinguished advertised context length from a model’s ability to use it. He described team experiments hiding ten pieces of information in increasingly long prompts. In his account, 32k GPT-4 retrieved only one of ten pieces at the longer context, compared with nine at roughly 1,000 tokens. He planned to repeat that test on the newly announced 128k model. He also welcomed JSON output as a common integration format that would make comparing providers, including open models, easier.[1]

His product judgment in 2023 was “customer and problem driven”. An AutoGPT-style agent had not been an obvious fit for the problems Bloop’s large-codebase users described, although he characterized its existing pipeline as GPT in a loop with function calling. He said the company had shifted its branding from named models toward solutions because agent terminology was not why users chose the product. Users included people who read more code than they wrote, such as designers assessing an interaction’s implications and product managers checking development estimates.[1]

In the same interview, Knight-Webb affirmed running AI Tinkerers in London and described collectively creating a startup-focused event in a city whose technology community included many agencies and enterprises. Earlier still, he signed a January 23, 2019 parliamentary submission as co-founder of Who Targets Me.[1][8]