AINEWS
A solo builder in Colombo just shipped a debugging tool built for the LLM
Bugpilot captures console errors and DOM state as Markdown, betting the new bottleneck in AI coding is explaining the bug, not catching it.
Bugpilot captures console errors and DOM state as Markdown, betting the new bottleneck in AI coding is explaining the bug, not catching it.
The new bottleneck in AI-assisted coding isn't getting the model to write code. It's getting the model to understand the bug. When Claude or ChatGPT ships something that breaks, the developer becomes a translator, re-typing console errors with stack traces, network requests, and the relevant DOM state into prompts until the assistant can finally act on a coherent picture. The translation costs tokens and time. A solo builder in Colombo, Sri Lanka named Malindu has bet his first shipped product on the idea that this translation step is itself a product category.
Bugpilot is a Chrome extension that captures browser state, console errors with stack traces, network requests, DOM snapshots, clicks, and screenshots, then exports the whole thing as AI-ready Markdown. The pitch isn't "a better bug reporter." It's that the consumer of bug reports has changed, and the format has to change with it. A screenshot of a console panel is useless to a language model. A structured Markdown block of the failing call, the network request, and the relevant DOM is exactly what an assistant can ingest.
The local-first posture is the design opinion that makes the rest legible. Bugpilot runs entirely in the browser: no servers, no accounts, no telemetry, and always-on redaction of captured content. That isn't a privacy footnote. It's a statement about where the developer's data should live when the consumer of that data is a model trained on other people's code. The Pro tier adds React component state capture, five AI-optimized export formats, and unlimited history. The free tier covers the core capture loop. The Pro upgrade costs $28, one-time, with a 14-day refund window. The maker also offers a DM-for-free path for users who can't afford it, per the Product Hunt launch page.
Malindu built the tool first for his own workflow, then opened it up. The launch lists him as a solo founder out of Colombo, framing the product for "vibe coders" who treat the LLM as a first-class collaborator rather than a search engine. The category argument the launch page makes implicitly: the tooling layer forming around the LLM context window will look less like a debugger and more like a translator.
There is a critique this product can't escape. AI-assisted coding still has a manual debug loop the human has to own, and no Markdown export changes that. The model is still guessing at intent. The test pass is still the test pass. What Bugpilot claims to compress is the gap between "I see a bug" and "the AI has enough structured context to help." If that gap closes, the cost of iteration drops. If it doesn't, the export format is window dressing on a workflow that was already going to be tedious.
At launch, Bugpilot sits at 74 followers on Product Hunt, an early-stage signal that fits a solo builder's first public release. The interesting question isn't whether the extension takes off. It's whether the category claim holds: that in a year of AI coding tools, the under-built layer is the one that hands structured context to the model, and that a small, local-first tool from a single developer in Colombo can stake out real territory by being the first to ship specifically for that consumer.