If you’re not using AI to build apps and software in 2026, you’re falling behind. In this video I show you exactly how I’m building my own AI startup (Super Agent — an Instagram agent that lives in your DMs 24/7) using an open-source AI coding agent called OpenCode, and the crazy part is: we’re doing it with a free model. I walk you through installing OpenCode locally (npm or curl), switching models with /models, and selecting Minimax M2.5 specifically from the OpenCode provider so it’s free (at least at the time I recorded). Then I take you inside my real codebase and show my workflow: I use Plan mode first (not Build) so the agent reads my spec (like my pricing_build.md), creates a safe step-by-step plan, and only then I confirm and let it implement. You’ll see it update my Convex backend schema, enforce plan limits (1 agent free vs 5 pro), rebuild the pricing UI, and add plan & billing + usage tracking. I also compare Minimax M2.5 vs Claude Opus 4.6 pricing/context and explain why I like Convex for startups (fast, code lives in your repo, free until you scale). Finally, I show the real-life reality of AI coding: when something breaks (webhook to agent reply), you iterate, debug, and if needed you switch to a stronger model to diagnose the issue.