← back
How Lovable self-improves every hour — Benjamin Verbeek, Lovable
Takeaway
Learn from resolved user friction and reuse generalized solutions before other users encounter the same failure.
Summary
- Lovable’s long-running project conversations provide feedback for continuous improvement at a reported scale of over 200,000 new projects per day.
- An LLM judge detects stuck users through repeated requests, complaints, and abandoned sessions.
- The team distinguishes solvable prompting failures, simple product gaps, and problems requiring substantial engineering.
- Successful recoveries become clustered knowledge entries in an internal Stack Overflow-style system, aiming to give future agents the solution earlier.
continuous-learninglovableagent-memory
Original description
Within the first hour of launching the vent tool, the agent filed 20 complaints about a silent file copy failure. The team checked: the tool worked fine. What the agent had caught was that filenames with a space in them silently failed to copy, a bug that never surfaced in logs. Benjamin Verbeek from Lovable built it a channel to complain directly to Slack when platform limitations block it, and the first thing it did was find a real production bug. At 200,000 projects per day, Lovable runs two continuous improvement loops. The first detects sessions where a nontechnical user got stuck and then unblocked, clusters similar cases, and injects that context upstream; a holdout group measures actual project completion rates to prune stale entries when models or features change. The vent loop runs in parallel: the agent flags missing tools, broken platform behavior, and confusing docs as it works. Vent volume spikes turned out to be a reliable incident detector. A second agent now monitors the channel, deduplicates reports, and opens PRs automatically. Speaker info: / benjamin-verbeek https://x.com/benjaminvrbk/