Scoop AI Hackathon Tokyo Bowl

The Thesis

The software industry is converging on AI-generated code. This is the wrong abstraction. Instead of having AI write servers, we should let AI be the server. LLMs don't write your infrastructure—they become it.

Engineers believe code equals reliability. But when AI generates that code, 10,000 lines of generated Python ≠ reliable. The code becomes an opaque intermediary.

As always, a very thoughtful and well reasoned take. I read till the end.

I think the Claude Code team itself might be an indicator of where things are headed. We have directional answers for some (not all) of the prompts:

1. We hire mostly generalists. We have a mix of senior…
— Boris Cherny (@bcherny) January 27, 2026

The Four Horsemen

Every AI system in production faces these challenges. Here's how AI as Runtime solves them.

Reliability

AI-generated code is opaque and brittle. AI as Runtime executes through constrained, auditable tools—every action visible, every outcome verifiable.

Cost

Code generation burns tokens on complexity. Direct execution uses minimal tokens per request. Pay only for what you use: ~$0.001 per execution.

Latency

No build step. No deployment. No cold start. Request arrives, AI executes, response returns—typically under 2 seconds end-to-end.

Security

Scoped tokens, allow-listed commands, sandboxed execution. AI operates within explicit boundaries—not arbitrary code running with full permissions.

Fast. Cheap. Reliable. Secure.

Watch AI execute directly—no code generation, no deployment, no compromise

~1s

Avg Latency

$0.001

Per Request

Lines of Code

∞

Adaptability

Try It Yourself

Send a webhook event and watch AI determine the execution plan

Sample Events:

⚡

Press Execute to see AI as Runtime in action

The Vision

Web3 × AI Runtime

Autonomous agents with economic agency—the foundation of a sentient agentic economy

🔐

Trustless Agent Actions

When AI executes via scoped tools—not arbitrary code—every action becomes cryptographically verifiable. Smart contracts can prove agent behavior on-chain.

💸

Micro-economic Viability

At ~$0.001 per execution, agents profitably participate in micro-transactions: paying for APIs, bidding in auctions, managing DeFi positions, tipping other agents.

🤝

Agent-to-Agent Commerce

AI agents that own wallets, negotiate service agreements, pay each other for compute and data, and form temporary DAOs for complex collaborative tasks.

⚡

Real-time Markets

Sub-2s latency enables agents to participate in arbitrage, liquidations, prediction markets, and auctions—humans and agents as peers in an open economy.

A mesh network of AI runtimes, each with economic sovereignty, transacting trustlessly across chains. This is the sentient agentic economy.

AI as Runtime The End of Code Generation