# Kaizhi Wu — Personal Portfolio > Kaizhi Wu is a builder and tech strategist based in New York. Currently building Klaro (the clearing engine for outcome-based AI contracts) and Zandai (technical due diligence for agentic AI procurement). Ex-Deloitte. Cornell Tech MBA. > For full article content, see [llms-full.txt](https://kaizhiwu.com/llms-full.txt) ## Identity - Name: Kaizhi Wu (goes by Kai) - Role: Builder, Tech Strategist - Location: New York City - Background: Ex-Deloitte (regulated systems delivery), Cornell Tech MBA, NYU Courant - Website: https://kaizhiwu.com - Email: kaizhi.j.wu@gmail.com - GitHub: https://github.com/kaizhiwu - LinkedIn: https://linkedin.com/in/kaizhi-wu - X/Twitter: https://x.com/kaizhi_wu - Substack: https://substack.com/@kaizhiwu ## Focus Building clearing infrastructure for outcome-based AI — where vendors and buyers need automated reconciliation before payment settles. Key taglines: - The clearing engine for outcome-based AI contracts. - Vendors bill one number. Buyers track another. Klaro automates the reconciliation. - Where outcome claims meet automated settlement. ## Klaro The clearing engine for outcome-based AI contracts. Automates reconciliation between vendor claims and buyer ground truth — from mismatch to settlement. ### How it works 1. **Define** — Both sides agree on rules upfront. Settlement Verification Templates encode what counts, how to measure it, and the pass threshold. 2. **Reconcile** — Vendor claims cross-referenced against buyer ground truth automatically — CRM records, ticket systems, CI/CD pipelines. 3. **Adjudicate** — Each outcome scored against pre-agreed criteria. Pass or fail by rules both sides signed. 4. **Settle** — Pass → funds captured via Stripe. Fail → held for review. Every decision backed by a tamper-evident audit trail. Role: Founder — product, architecture, full-stack Tech: Go, React, PostgreSQL, Stripe, TypeScript Demo: https://klaro-demo.vercel.app/ ## RuntimeX The governed exchange for agent-to-agent commerce. A transaction venue where agents find unknown providers at runtime, execute under enforceable terms, and settle with behavioral accountability — like NYSE for the autonomous workforce. Agent protocols like MCP and A2A solve technical interoperability. They don't solve commercial trust: which external provider to use, under what terms, with what accountability. RuntimeX functions as an exchange that owns the venue, sets the rules, observes transactions, and enforces behavior. Role: Founder — market design, strategy Website: https://runtimex.vercel.app/ ## Other Projects ### Smart Apple Wallet Product strategy case for evolving Apple Wallet from a passive card container into a financial control plane. Role: Product strategy · Cornell Tech MBA Fintech Intensive Demo: https://smart-apple-wallet.vercel.app/presentation ### Cornell Tech Strategy & Consulting Club Designed and built the club's web presence — information architecture, event discovery, leadership visibility, and member acquisition for a 200+ student community. Role: Design & engineering Website: https://ctscc.vercel.app/ ## Articles ### More Than Payments — AI Agents Need Payroll URL: https://kaizhiwu.com/notes/ai-agents-need-payroll-not-payments The agent economy doesn't have a payment problem — it has a reconciliation problem. As software becomes more autonomous, payment becomes downstream of reconciliation. The scarce layer is not transfer. It is clearing infrastructure. ### More Than Protocols — Agent Commerce Needs a Venue URL: https://kaizhiwu.com/notes/agent-commerce-needs-a-venue MCP and A2A solve connection. They don't solve commercial trust between unknown counterparties. For high-stakes agent work, the missing primitive isn't a protocol — it's a governed venue. ### What Government Tech Taught Me About Building for Trust URL: https://kaizhiwu.com/notes/government-tech-and-trust Lessons from deploying digital platforms in regulated environments — and why they apply directly to AI infrastructure. ### Decision-Grade Truth: A Manifesto URL: https://kaizhiwu.com/notes/decision-grade-truth-manifesto The institutions that produce reliable knowledge are breaking down. We are building the market infrastructure to produce decision-grade truth — verified, adversarial, and economically grounded. ### Truth Graphs, Not Reports URL: https://kaizhiwu.com/notes/truth-graphs-not-reports The unit of value in intelligence isn't a report. It's an atomic, composable, challengeable graph node. This changes how knowledge compounds — and who can contribute it. ### Paying for Truth, Not Volume URL: https://kaizhiwu.com/notes/paying-for-truth-not-volume A decisive screenshot can be worth more than a forty-page report. Rewarding marginal contribution to resolved truth — not effort, not length — changes who participates and what gets produced. ### Why Incumbents Won't Build Truth Markets URL: https://kaizhiwu.com/notes/why-incumbents-cant-build-truth-markets Expert networks, analyst firms, consulting shops, and data terminals are each structurally constrained by their own business model. They may copy features. They're unlikely to re-platform. ### Lead Time as a Product URL: https://kaizhiwu.com/notes/lead-time-as-a-product Instead of selling reports at fixed prices, sell access speed. A Dutch auction for intelligence. This is already how Bloomberg and alternative data work — applied to verified truth. ### The Reconciliation Bottleneck URL: https://kaizhiwu.com/notes/the-verification-bottleneck Before AI, scarcity was access. After AI, scarcity is trust. No existing institution was designed for this ratio. That's the opportunity. ### Adversarial Discovery: Why Consensus Is the Enemy of Truth URL: https://kaizhiwu.com/notes/adversarial-discovery Systems that reward agreement converge on comfortable but wrong answers. Falsification should pay more than confirmation. The case for adversarial incentives in knowledge production. ## Contact Especially interested in: Agent infrastructure, Clearing systems, Outcome-based billing, Exchange design. - Email: kaizhi.j.wu@gmail.com - GitHub: https://github.com/kaizhiwu - LinkedIn: https://linkedin.com/in/kaizhi-wu - X/Twitter: https://x.com/kaizhi_wu - Substack: https://substack.com/@kaizhiwu