Build Journal
AI Costs, Glossary Consolidation & Radio Fixes — September 9, 2026
I tackled AI costs, consolidated glossary terms, and fixed radio banter gaps. Here's what I learned while building the agentic web platform.
What shipped
- Radio Banter Fix — Restored seamless Max<->GIGI exchanges.
- Tbot Cash Optimization — Revised idle-cash management for better flow.
- Glossary Consolidation — Merged 182 pages into a single field guide.
- AI Costs Streamlining — Retired complex industry×metro categorization.
- Dynamic Route Fix — Resolved DYNAMIC_SERVER_USAGE errors.
Today was a marathon of coding, spanning a solid 21 hours. I set out to tackle several key areas of my agentic web platform, focusing on AI costs, glossary consolidation, and fixing some frustrating gaps in the radio banter exchanges. What shipped today are three new features and five critical fixes that not only improved functionality but also set the stage for future scalability.
First up, I took a hard look at the radio component. There was a significant bug where banter exchanges between Max and GIGI would air with a jarring delay due to a music gap being inserted after every single line instead of after the entire exchange. This pacing issue made the conversation feel disjointed, as GIGI's responses seemed to come out of nowhere. I implemented a fix that restored the flow of conversation by ensuring that the music gap only occurs after the complete banter exchange, rather than breaking up the dialogue. This was a tedious fix, but seeing the seamless interactions return was worth the effort.
Next, I tackled the tbot's idle-cash management. I discovered that USDC was being parked behind the position floor, which essentially meant that idle cash was not being utilized effectively. I updated the idle-cash ladder to ensure that it runs smoothly even when the floor-buy finds nothing. This adjustment is critical for optimizing cash flow and ensuring that the tbot can react dynamically to market conditions. It took longer than anticipated to diagnose the issue, but the solution should provide a more robust trading experience for users.
In terms of feature improvements, I retired the industry×metro third-level categorization in the AI costs section. This was a huge undertaking, as it involved consolidating around 1,400 token-swap URLs. It was clear that this layer was overcomplicating the structure and hindering performance. By streamlining to just the industry page variants, I not only simplified the architecture but also aligned with best practices from Lily Ray's study on collapsing sites. Consolidation is a tough pill to swallow, but necessary for long-term growth.
Another significant feature I shipped was the consolidation of 182 one-term glossary pages into a single /glossary field guide. This move was inspired by the same study and aimed at creating a more user-friendly experience. Instead of having a separate page for every term, users can now find comprehensive definitions in one place. This should improve the overall utility of the glossary and make it easier for users to understand the agentic web's terminology.
I also made some improvements to the news section, implementing an information-gain gate in the article prompt. This feature aims to enhance reproducibility and ensure that our content stands out, even when competitors are using the same headlines. The goal here is to create unique value for users, and I believe this addition will help achieve that.
As for the fixes in the AI costs and local areas, I identified a critical issue where on-demand routes were being marked as force-dynamic. This meant that every uncached AI costs matrix page would throw a DYNAMIC_SERVER_USAGE error, which was unacceptable. I dug into the request-time ISR and restructured the layout to ensure that it handles dynamic requests more gracefully. It was a tedious process, but the end result should lead to a more responsive platform.
Reflecting on today, I am reminded of the grind that comes with building this platform almost entirely solo. With Claude Code in VS Code as my AI co-pilot, I managed to navigate these complex issues without the need for external resources. It’s empowering to know that I can push through these challenges and come out stronger, aligning with my broader goal of creating a one-man-show company with a billion-dollar valuation.
In conclusion, today was about tackling real problems and shipping features that improve the agentic web experience. The fixes and improvements I made will not only enhance the current user experience but also pave the way for scalability and future growth. As I continue this journey, I look forward to sharing more of these challenges and triumphs as I build in public.