Is traditional coding obsolete? This is the question that numerous developers have been pondering over the past week since the release of potent new coding models by OpenAI and Anthropic.
Last week, OpenAI and Anthropic rolled out their respective coding models—GPT-5.3-Codex and Claude Opus 4.6—both of which marked substantial advancements in AI coding abilities. GPT-5.3-Codex exhibited notably superior performance on coding benchmarks compared to previous models, whereas Opus 4.6 introduced a feature enabling users to deploy autonomous AI agent teams that can simultaneously address various facets of complex projects. Both models can compose, test, and debug code with little human input—even iterating on their own work and refining features before presenting outcomes to developers.
The releases—particularly GPT-5.3-Codex—triggered an online existential quandary among software engineers. At the core of this was an essay penned by Matt Shumer, CEO of OthersideAI. Shumer stated that “something fell into place” after the model releases and characterized AI models as now autonomously managing the entire development cycle—writing tens of thousands of lines of code, launching applications, testing features, and iterating until content, with developers merely articulating desired results and stepping back. He suggested that these advancements implied AI could disrupt jobs more drastically than the COVID-19 pandemic.
The essay elicited varied responses. Some tech leaders concurred, such as Reddit co-founder Alexis Ohanian, while others, including NYU professor Gary Marcus, labeled it “weaponized hype.” (Marcus pointed out that Shumer offered no data to back up claims that AI can write complex apps without errors.) ’s Jeremy Kahnargued that it was coding’s distinctive traits—like automated testing—that made full automation easier, whereas automation in other knowledge-work domains might be more difficult to achieve.
Software engineers as early adopters
For numerous engineers, some of Shumer’s warnings merely mirror their present situation. Many engineers state they have entirely stopped coding, instead relying on AI to pen code as per their instructions.
Although the new releases do signify meaningful improvements, developers also stated that the industry has been undergoing a gradual transformation over the past year as models became proficient enough to autonomously manage increasingly complex tasks. While developers at prominent tech companies have mostly stopped writing code line by line, they haven’t ceased building software—they’ve become directors of AI systems that perform the typing. The skill has shifted from writing code to designing solutions and guiding AI tools. Some contend that the new models mainly “burst the bubble” around AI coding by making non-coders aware of a trend that engineers have been experiencing for months.
During its earnings call this week, [company name] co-CEO Gustav Söderström stated that developers “haven’t written a single line of code since December.” The streaming giant’s internal system utilizes Claude Code for remote deployment, enabling engineers to command AI to fix bugs or add features via Slack on their phones during their commute, then merge completed work to production before arriving at the office. Söderström mentioned that Spotify launched over 50 new features in 2025 using these workflows.
Even within Anthropic, engineers are heavily dependent on their own tools to compose new code. Boris Cherny, head of Claude Code, stated earlier this month that he hasn’t written code for over two months. Anthropic revealed that 70% to 90% of the company’s code is now AI-generated.
The models themselves have also achieved a recursive milestone: They are now substantially aiding in constructing more advanced versions of themselves. OpenAI stated that GPT-5.3-Codex “is our first model that was crucial in creating itself,” a notable shift in how AI development operates. Similarly, Anthropic’s Cherny said his team developed Claude Cowork—a non-technical version of Claude Code for file management—in roughly a week and a half, mostly using Claude Code itself. Even for Claude Code, Cherny stated that about 90% of its own code is now written by Claude Code.
Despite the productivity gains, some developers are also cautioning that the new tools could lead to burnout. Steve Yegge, a seasoned engineer, warned that overwork is draining developers.
In a widely circulated blog post, Yegge described dozing off abruptly after lengthy coding sessions and colleagues contemplating installing nap pods at their office. He contends that the addictive nature of AI coding tools is compelling developers to take on unsustainable workloads. “With a 10x boost, if you provide an engineer with Claude Code, then once they’re proficient, their work output will generate the equivalent of nine additional engineers’ worth of value,” he penned. But “building things with AI requires substantial human energy.”