โ Clip via: Theo’s cost comparison is brutal for OpenAI: about $400K of Codex tokens over months of agent loops stalled at roughly 84% compatibility, while about $20K of Claude Opus 5.5 finished a full Rust rewrite of the TypeScript compiler, type checker and LSP in two weeks. He says he never read a single line of the code, and he claims the result runs about 2x faster than TSGO, the Go port Microsoft’s TypeScript team built by hand. On a job this hard, the model you pick decides whether it gets done at all.
That number used to sound absurd. Now it sounds like a bargain โ or a warning, depending on which model you chose.
The gap between two AI agents attempting the same hard task isn’t measured in percentage points. According to the framing Theo laid out, it’s the difference between a project that finishes and one that doesn’t.
The Cost Split
~$400K on Codex. Months of agent loops. Stalled at ~84% compatibility. Then ~$20K on Claude Opus 5.5. Two weeks. Done. The delta isn’t a rounding error โ it’s a completed project vs. an abandoned one.
“On a job this hard, the model you pick decides whether it gets done at all.”
The Structural Read โ FourWeekMBA Analysis
This is a live example of what we call the Product Overhang Doctrine: capability doesn’t accumulate linearly โ it surfaces in threshold jumps. A task sits at 84% for months, then a different model crosses the finish line in two weeks. The capability gap between models isn’t about benchmarks; it’s about whether a real engineering project ships.
The deeper signal is coststructure. $420K total sounds large until you price it against engineering salaries, timelines, and the alternative โ which is the project not existing. What Theo’s framing implies is that AI spend is increasingly a build-vs-don’t-build decision, not a cost-optimization conversation.
He says he never read a single line of the code. That’s not a bug โ that’s the strategy. The builder’s leverage isn’t deep expertise in Rust or TypeScript internals. It’s knowing which model to point at the problem and when to trust the output. Harness beats build, but only if you pick the right harness.
The Hidden Risk
Model selection is now a first-order business decision, not an IT configuration choice. Per Theo’s argument, committing $400K to the wrong model doesn’t just burn budget โ it burns months of runway on a task that never resolves. The evaluation step before the project is worth more than most teams are pricing it.
THE NUMBERS (per Theo’s account)
~$400K
Codex spend โ stalled at ~84%
~$20K
Claude Opus 5.5 โ full rewrite, 2 weeks
The Bottom Line
$420K to rewrite a TypeScript compiler in Rust โ zero lines of code read by a human โ and the claim is it outperforms Microsoft’s hand-built Go port. Whether or not this generalizes, Theo’s framing makes one thing hard to ignore: in the agentic era, the most expensive decision isn’t the API bill. It’s choosing the wrong model for a task that never finishes.
This post is FourWeekMBA’s analytical read of a clip from Theo’s account, attributed as his argument and framing โ not verified fact, not investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur, HackerNoon and Search Engine People. Find him on LinkedIn and Substack. See how we source.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
