The article argues that future programming will shift from manual coding to defining requirements, constraints, tests, and evaluating AI-generated code, with the core role of programmers becoming broader architects who guide AI output rather than write code themselves.

Programmers are described as “dragon trainers,” responsible for defining what’s valuable and ensuring AI-produced code meets criteria rather than performing hands-on coding.

As the marginal cost of code generation nears zero, the industry will see programmers focus on requirements, testing, and validation of AI-generated code instead of writing lines of code by hand.

The new ttfx repository is hosted under DHH’s Omarchy organization, preserving all commits, and demonstrates how a Python tool was converted to Rust with nearly identical code size and functionality.

DHH used Claude to plan the task and then delegated execution entirely, illustrating multi-agent collaboration and batch merging across branches with minimal manual intervention.

A similar AI-assisted refactor was used to migrate Bun’s codebase, moving a million lines from Zig to Rust in under two weeks, signaling large-scale automation in codebase migrations.

Bun’s migration example underscores AI-assisted refactoring at scale, moving millions of lines from Zig to Rust in a short period.

Claude Code delivered faster results with minimal manual intervention, while Codex produced strong results as well but was slower and more costly.

TerminalTextEffects, a terminal visuals engine, can run as a CLI or Python library, capable of controlling text movement, colors, gradients, and animations.

The piece includes standard financial/tech media disclosures about third-party sourcing and investment risk from KuCoin.

Claude Code is cited as capable of generating and rewriting large codebases, with OpenAI researchers publicly supporting AI handling coding tasks.

DHH used Claude to rewrite TerminalTextEffects into Rust in about three hours, producing 21,000 lines of Rust and a 3MB executable at a token cost of around 11 million.