Context Goblin

How to build a startup with AI

TLDR: Use LLMs to investigate your potential audience - crawl Reddit, X, Hacker News, or anything related. Get used to a vibe-coding workflow and build your product or webpage within a week. Use the remaining time to manually promote and contact customers. Most importantly: never slop post.

In this post, I want to guide you through the path we took to build Context Goblin, a customizable AI code review tool. I won't cover the idea-finding process, but focus on how we improved the engineering and mechanical work of bootstrapping a product.

Leveraging LLMs for every step

Building from scratch and building on top of an existing product are completely different tasks for a software engineer, and equally for a coding agent. Where you have to provide context and connectors for every planning and artifact doc manually on an existing project, a coding agent lives entirely off the prompt when creating something from scratch.

That's where we left a lot of efficiency on the table in the early days of Context Goblin. Relying on well-defined prompts and letting agents ask clarifying questions has done wonders for the quality of the software they produce.

Another workflow enhancement is using agents extensively to investigate your audience. We used custom skills to let AI crawl successful product hunts, analyze competitors and compare their approaches and customer experience with our own product. With the right intent, it is very easy to gather insights about the niche.

Do not slop post

While promoting our product we repeatedly ran into the same issue of developers aggressively downvoting vibe coded stuff. By now, it's almost an industry-wide rule. If you scroll through our earlier blog posts we actually wrote them with AI too, I guess that's the reason they didn't take off on traction.

The lesson: if we really need inspiration for a text, we use LLMs to get ideas. But never let them write whole sentences - it will be noticed. At least in this generation of models.