AI Costs, Glossary Consolidation & Tbot Fixes — September 9, 2026

Today was a marathon 21-hour coding sprint where I tackled several critical features and fixes that pushed the project forward. What shipped included a significant overhaul of our glossary structure, a major fix for Tbot's idle cash management, and improvements in the radio banter exchanges. Each of these areas presented unique challenges that required deep dives into our systems, and I learned a lot along the way.

The first major feature I rolled out was the consolidation of 182 one-term glossary pages into a single field guide. This was a direct response to insights gained from Lily Ray's studies, which highlighted the inefficiencies of having one URL per term. The new index now carries every curated definition in full, allowing users to access all terms in a more cohesive manner. The goal was to streamline access and improve the overall user experience, and I believe this will significantly enhance our site's relevance and authority in search engines.

On the fixes side, I tackled a pressing issue with Tbot. I discovered that USDC was being parked behind the position floor, which was causing idle cash to remain unutilized. The idle-cash ladder needed to be restructured so that it would run when the floor-buy found nothing. This fix not only optimizes our cash flow but also enhances the overall efficiency of the Tbot system. It took a considerable amount of time to diagnose the problem and implement a solution, but the end result is a more responsive and efficient trading bot that should perform better under varying market conditions.

I also made some crucial adjustments to our radio functionality. Previously, the pacing of banter exchanges was disrupted by a 4-8 minute music gap that split the conversation, making it seem disjointed. The fix involved ensuring that music gaps occurred after the entire exchange rather than between individual lines. This was a challenging bug to track down, but I’m thrilled with the outcome. The banter now flows naturally, providing a smoother listening experience for users. It’s these little details that can make or break user engagement, and I’m proud to have resolved this.

Another significant improvement I made was in our agentic-rails. I capped the rendered build playbook at the top-priority 80 rails, which was necessary after the rules base ballooned to over 9,600 rows. Without this cap, the rendering would have resulted in a 3MB file, which was simply unusable by our MCP consumers. This fix not only preserves server resources but also ensures that our builds remain efficient and manageable.

Of course, with every productive day comes its share of challenges. The on-demand routes for AI costs presented a significant hurdle as they were throwing dynamic server usage errors. This issue required a deep dive into the request-time ISR, where I discovered that the layout was request-shaped. It was a tedious process to diagnose, and I lost a few hours trying to pinpoint the root cause. However, once I identified that the ISR render was causing the issue, I was able to implement a solution that should prevent these errors in the future.

As I reflect on today’s achievements, I can’t help but appreciate the power of working solo with AI assistance. My reliance on Claude Code in VS Code has allowed me to tackle these complex problems without the need for a large team. While collaborating with others can offer unique insights, there’s something incredibly rewarding about being able to push through challenges independently. Today’s progress is a testament to that philosophy, as I continue to build toward my vision of a one-man-show company with a billion-dollar valuation.

In conclusion, today was a long and intense day filled with significant advancements and some frustrating setbacks. But every line of code brought me closer to my goals for the agentic web and the comprehensive AI-driven solutions I’m building. As always, I’m excited to see what tomorrow brings and how I can continue to refine and enhance this platform.