Claude's Wager: Why you don't have to believe in AI to bet on it

I'm tightening those to the source's facts and rechecking the length.A software developer has named a practical bet on AI coding agents Claude's Wager, borrowing Pascal's Wager, the famous argument that a rational person should act as if God exists, whether or not God does exist. The coding question is not whether those agents will replace programmers. It is what developers should do while they find out, as set out in wsvincent.com's report, last updated on November 9, 2026 and tagged ai and programming.

I'm tightening those to the source's facts and rechecking the length.A software developer has named a practical bet on AI coding agents Claude's Wager, borrowing Pascal's Wager, the famous argument that a rational person should act as if God exists, whether or not God does exist. The coding question is not whether those agents will replace programmers. It is what developers should do while they find out, as set out in wsvincent.com's report, last updated on November 9, 2026 and tagged ai and programming.

This piece follows that path from ChatGPT's November 2022 launch through unreliable 2023 autocomplete, 2024 tools that finished tasks alone, and the 2025 leap by Claude Code and Codex. It then covers verbose generated code, the late-2026 split between taste and the claim that code is free, and a wager that keeps enough programming skill to challenge model output. Working programmers are the audience, including people who doubt a full replacement and people who already let an agent write the code and skip reading it.

01Pascal's Wager for coding agents

The famous form of the argument says a rational person should act as if God exists, whether or not God does exist. A programmer who has experimented with large language models since ChatGPT first launched in November 2022 treats that structure as a useful way to think about AI coding agents. In the beginning, four years ago, it was truly jaw-dropping that a program could output complete sentences and paragraphs on almost any subject.

Coding did not seem like the focus at first. In retrospect it should have been, because the work is text-based, somewhat verifiable, and customers are willing to pay. The essay uses "AI" to mean large language models, and it treats the years since November 2022 as the time to decide how to act while the replacement question is still open.

02ChatGPT, Cursor, and agent harnesses

Early chatbots and autocomplete tools in 2023, Cursor among them, were impressive but unreliable. The developer remembers trying Cursor in 2023, amazed by what it could attempt, then frustrated by how often it got things wrong. For a long time, AI coding felt like more trouble than it was worth, and that judgment has since changed dramatically.

By 2024, AI tools could perform meaningful tasks on their own, not just suggest code improvements. In 2025, Claude Code, Codex, and more tools took a huge leap by coupling language models with harnesses and agentic patterns, iteratively working through problems.

In 2026, what AI can do is truly jaw-dropping. Humans still need to point AI at a task and provide some taste and judgment. A growing number of programmers not only do not write code by hand anymore, but have also become comfortable not reading it.

03Verbose code and crowded pull requests

AI-generated code is typically ugly to this developer's eyes, overly verbose in all the wrong ways. The verdict comes from someone who has been a programmer for years and has used large language models since ChatGPT's November 2022 launch. The essay says that ugly, verbose output has nevertheless become the norm in many sectors.

Humans in the loop feel overwhelmed by the pace of development, with ever-increasing lines of code and pull requests. Are humans slowing progress, or are they the only things keeping modern software development on the rails? AI has advanced so quickly that many programmers are shell-shocked.

04Stack Overflow, slop, and free code

The essay, last updated on November 9, 2026, says Stack Overflow is fading, traditional search is increasingly dominated by AI-generated answers, and the internet is filling with AI slop. Even programmers who can manage to avoid the AI wave are still engulfed by it.

The real divide now, in late 2026, is what is left for humans to do. Some believe that direction, taste, and product focus will remain for software developers, and the essay forecasts that developers become more like product managers than coders. Others believe that ship has sailed too and that "code is free."

05Linters, tests, and Claude's Wager

The developer does not want to spend days writing code that AI can generate in seconds. The same person does not want to become a programmer who can no longer understand what they are building. The bet is to work alongside AI while continuing to develop the programming skills that let someone understand, evaluate, and challenge its output.

Maybe coding agents eventually become so capable that they do not need guidance at all. Or maybe they become another part of the existing toolkit, alongside linters, syntax checkers, unit tests, and everything else used to build better software. A programmer does not have to believe AI will replace programmers to think AI is worth learning.

If AI continues improving at its current pace, programmers who have learned to work with it will be better positioned. If AI hits a wall tomorrow, those programmers will still know how to code, and the essay's author does not know which future arrives. The cost of experimenting thoughtfully seems relatively small, while refusing to engage could be enormous, and refusing to make a bet is a bet in itself.