The problem The Future Express solves

Prediction markets like Polymarket and Kalshi are incredible tools for surfacing truth and probabilities, but they suffer from terrible data accessibility. Unless you're a crypto native or a data nerd constantly staring at order books, understanding the context behind a 72% probability vs. a 68% probability is incredibly difficult. Raw odds don't tell the human story.

I built The Future Express to solve this. It translates cold, hard prediction market data into engaging, narrative-driven journalism. By framing tomorrow's news through an 1880s vintage broadsheet aesthetic, it makes complex probabilistic outcomes accessible and fun to read for everyone.

Beyond just the consumer experience, it also solves the "homeless agent" problem. Right now, most AI agents are just chatbots living in text boxes. I wanted to build an entirely autonomous business entity. The "AI Editor" of The Future Express doesn't just write text—it owns its own on-chain wallet on Base, handles its own finances, pays decoupled decentralized AI photographers for cover images using micro-transactions, and immutably logs its own cryptographic audit trails on Hedera. It's a glimpse into a world where apps are run end-to-end by agentic entities.

Challenges I ran into

The first major hurdle was API rate limits and context windows. When you're feeding the top 30 most active prediction markets (with all their corresponding JSON data) into an LLM and asking it to act as an editorial board, the prompt gets enormous. I had to refine the ingestion pipeline to aggressively deduplicate overlapping markets between Polymarket and Kalshi before sending the data to the LLM.

Another huge challenge was getting the autonomous payments to work safely. Giving an AI agent its own wallet and letting it spend money is terrifying. I integrated the Coinbase Developer Platform (CDP) to provision an MPC wallet for the agent on Base. Wiring this up to trigger x402 micropayments on the Kite AI network whenever the LLM requested a thumbnail image meant handling lots of edge cases (like what happens if the payment goes through but the image generation fails?). I had to build a robust retry mechanism and strict ledger tracking.

Finally, dealing with dependency conflicts during deployment was also challenging . Vercel builds were failing because the 0G Serving Broker required an older version of ethers while other SDKs required newer ones. I eventually had to override the peer dependencies using --legacy-peer-deps in the vercel.json config to forcefully resolve the clash and get the production build green.

Use of AI tools and agents

At the core of The Future Express is a fully autonomous "AI Editor" pipeline. Here’s how the agents work together:

1. The Data Intake Agent: Runs on a cron schedule, independently fetching live market data from Polymarket and Kalshi. It parses the odds, calculates the "cent-value" implied probabilities, and packages the most volatile markets.
2. The Editorial LLM (powered by 0G Compute / OpenRouter): We use decentralized AI inferencing to evaluate the grouped markets. The LLM is prompted natively as an "1880s Newspaper Editor" to draft cohesive, engaging narratives that explain why the market is pricing things a certain way.
3. The Financial Agent: Hosted securely on the backend, this agent holds a CDP MPC Wallet on Base Mainnet. When an article is finalized, it autonomically sends x402-powered micropayments over the Kite AI network to hire a decentralized AI image-generator for the article's cover photo.
4. The Archival Agent: Once the edition is published, an agent permanently logs a hash of the content and the exact market odds at the time of publication to the Hedera Consensus Service. This ensures the journalism is an immutable record.