Already Here / Episodes / Episode 2

EP. 02 · AUGUST 4, 2026 · 44 MIN

With Matt Van Horn

Matt Van Horn is the builder behind Last 30 Days and Printing Press. He turns problems from his own life into small, public tools, then lets real use and open-source contributors determine what happens next. In this conversation, the useful lesson is how quickly he moves from annoyance to working software—and how deliberately he decides what deserves continued attention.

Matt wanted current advice about AI tools that change faster than ordinary search results can keep up. He built Last 30 Days for himself, shipped the first version during a day on a ski mountain, and watched outside contributors reshape it. Michael also sees Printing Press turn websites into text commands an AI assistant can operate, including Flight Goat for comparing nonstop trips across travelers and departure cities.

Watch on YouTube · Listen on Spotify · Listen on Apple Podcasts

What you’ll learn

Matt built Last 30 Days while his daughter was skiing and the snow was bad. He released a tool made for his own needs, then watched it become one of GitHub’s top trending repositories and pass 50,000 stars. The lesson is not to predict the hit; it is to shorten the distance between a real frustration and something useful enough to share.

Advice about new AI tools was going stale before Matt could use it. Last 30 Days brings recent findings from Reddit, X, and web search into one research flow, so the assistant can synthesize what people are learning now instead of relying only on older articles.

An anonymous contributor deleted roughly half the code and submitted the result as “2.0.” Matt’s first reaction was defensive; after reading it, he saw that the change made the project better. Publishing created a path not only to distribution, but to ideas he would not have produced alone.

Printing Press examines a website or online service and creates a command-line tool plus an agent skill. The Flight Goat example lets an assistant compare nonstop flight options across travelers, dates, and departure cities without forcing Matt to click through every search page himself.

After Michael mentions a stolen bike, the conversation turns into installing a Craigslist command-line tool that can repeat the search. Matt’s pattern is consistent: start with a concrete annoyance, make the smallest useful mechanism, and allow reuse to reveal whether the idea deserves more work.

A project can be useful without becoming your identity or your full-time responsibility. Before releasing something, Matt asks whether he wants to manage the attention and energy it may attract. Publishing quickly does not remove the need to choose what you are willing to maintain.

Matt increasingly treats plans as instructions for the agent rather than documents he must approve line by line. He supplies context, evaluates the finished behavior, and sends weak work back for another pass. The human judgment moves from supervising every step to deciding whether the result is actually good.

Topics and resources

  • Last 30 Days — Matt’s open-source skill for researching recent conversations and web results
  • Printing Press — creates agent-ready command-line tools from websites and online services
  • Flight Goat — compares nonstop flight options across travelers and departure cities
  • OpenClaw — the open-source personal AI project discussed in the episode
  • OpenAI Codex — one of the coding agents in Matt’s building and review workflow
  • Compound Engineering — the planning and execution workflow Matt uses for iterative agent work

One safe experiment

Choose one recurring search you still perform manually. Write down the exact result you need and the constraints that matter, then ask an AI assistant to turn those instructions into a repeatable prompt or command. Run it once, inspect every source and result, and keep the final decision with yourself.

Browse every episode · Newer: Thanh Pham · Previous: Sam Gaddis · YouTube · Spotify · Apple Podcasts