OpenAI Codex and Anthropic Claude encourage programmers to abandon the conventional approach to writing code

Did traditional programming die?

Recently released models from OpenAI and Anthropic have prompted many developers to ask: “What now with classic coding?” This month saw the release of GPT‑5.3‑Codex and Claude Opus 4.6, which significantly outstrip their predecessors in code generation. GPT‑5.3‑Codex demonstrates higher benchmark results, while Opus 4.6 added the ability to run autonomous AI agents capable of solving different parts of a complex task in parallel.

Both models can write, verify, test, and debug code without human intervention, even improving their own solutions before the result is presented to the user. This has sparked a real “existential crisis” within the community of software engineers.

Matt Shumer’s viral post
Chief Executive Officer of OthersideAI Matt Shumer published an article claiming that new AI models have “softly clicked” and are now capable of completing the full development cycle: writing tens of thousands of lines of code, running applications, testing functions, and correcting errors until a satisfactory result is achieved. According to him, it’s enough just to describe the desired outcome, and the algorithms will self‑improve.

Shumer predicted that this could disrupt the labor market more than the COVID‑19 pandemic.

Varied community reactions
* Support – Reddit co‑founder Alexis Ohanian voiced support for Shumer.

* Criticism – New York University professor Gary Marcus noted a lack of evidence that AI can actually generate flawless code for complex applications. Jeremy Kan from Fortnite added that while automated testing simplifies full automation of programming, fully automating other areas of intellectual work is still impossible.

Reality in practice
Many developers are already abandoning manual coding and instead use AI as a “director” for text generation. In large tech companies there’s a shift: engineers no longer write code directly but manage the creation process through AI tools. The skill set shifts from simple coding to designing solutions and managing those systems.

Example from Anthropic
Within Anthropic, there is also a strong dependence on existing models. Claude Code head Boris Cherny stated that he hasn’t written code in over two months – everything was done via AI. Models have reached a point where they can help create more advanced versions of themselves. OpenAI and Anthropic use previous iterations of algorithms to build, verify, and test new models.

Burnout concerns
Despite increased productivity, some developers express worry: new AI tools could lead to “burnout” due to constant overtime. Experienced programmer Steve Yegge shared that he often fell asleep right at his workplace after long coding sessions, and colleagues sometimes… (text cuts off).