AI is writing most of Anthropic’s software, and engineers are taking on a new role

Anthropic says AI now writes most of its software

What happens when artificial intelligence starts writing most of a company’s software? According to Anthropic, that future has already arrived. The company says its Claude AI model now produces more than 80 percent of the production code merged into its systems, marking a dramatic shift in how software is developed.

Instead of serving as a simple coding assistant, Claude has become a major contributor to software development. The milestone offers an early look at how AI is changing engineering work, with developers spending less time writing code and more time guiding, reviewing, and improving AI-generated solutions.

Engineers are writing less code than ever

Before Claude Code launched in research preview in February 2025, Claude authored only a low single-digit share of the code merged into Anthropic’s codebase. By May 2026, Anthropic said Claude authored more than 80 percent of the code merged into its production codebase.

That shift has changed how engineers work inside the company. Instead of writing every line themselves, engineers increasingly direct Claude, review its work, and decide whether AI-generated code is ready to merge.

Productivity has jumped in a short time

Anthropic says its engineering teams are merging much more code than before. In the second quarter of 2026, the typical engineer merged eight times as many lines of code per day as they did in 2024.

Anthropic cautions that lines of code are an imperfect measure and likely overstate the true productivity gain. Even so, the company says the increase reflects a major workflow change, with Claude writing much of the code while engineers direct, review, and evaluate the work.

AI coding has become far more reliable

Not long ago, many developers worried AI-generated code would create bugs and require constant human fixes. Anthropic says those concerns were valid, especially during late 2025, when Claude still needed frequent corrections.

By mid-2026, however, Claude’s success rate on complex coding tasks had climbed to 76 percent after improving by 50 percentage points. The AI has become much better at understanding complicated instructions and carrying out multi-step programming tasks.

Claude is improving its own code too

Claude is doing more than writing software. Anthropic says the AI is increasingly acting as a code reviewer and optimizer, helping identify better ways to improve performance after the initial code has already been written.

In one internal experiment, Claude improved a model training pipeline through repeated self-correction, producing a 52 times speedup. The result suggests AI can now refine complex software in addition to generating it.

Developers are becoming AI managers

As Claude takes over much of the coding process, software engineers are spending less time writing syntax and more time defining goals, reviewing results, and making high-level technical decisions that guide AI systems.

Anthropic shared one employee quote saying it had been about five months since they last wrote code themselves. The broader shift points to engineers spending more time on architecture, planning, and supervision instead of handling every programming task manually.

Knowing what to build may matter more than coding

Anthropic’s experience suggests this shift could change how technical teams divide their work. As AI agents become more capable, engineers may spend less time writing each line of code and more time deciding what problems to solve.

Technical expertise will still matter. The emphasis may gradually move toward understanding problems, designing systems, reviewing outputs, and deciding what software should accomplish.

AI still needs people to make the big decisions

Even as Claude writes most of Anthropic’s software, human engineers remain responsible for the final product. They decide project goals, review AI-generated code, and make architectural choices that determine how systems are built.

Rather than replacing engineers entirely, the company describes AI as taking over much of the implementation work. People still provide the judgment needed to evaluate tradeoffs, ensure reliability, and steer complex software projects in the right direction.

The software industry could be entering a new era

Anthropic’s experience offers a glimpse of how software development may evolve across the technology industry. As AI agents become more capable, companies could build products faster while asking engineers to focus on strategy instead of repetitive coding tasks.

The transition is still unfolding, but it suggests the definition of software engineering is changing. Success may increasingly depend on directing intelligent tools effectively rather than writing every function manually.

Coding is changing, but engineers remain essential

Anthropic’s move to AI-generated code shows how quickly software development is evolving. While Claude now handles most of the coding work, human engineers continue to shape products through planning, oversight, and technical decision-making.

The company’s experience suggests the future of programming may be less about typing code and more about guiding powerful AI systems. As those tools improve, the role of engineers is likely to keep evolving rather than disappear.

AI could make software creation much faster

Anthropic’s experience suggests AI could dramatically speed up parts of software development across the tech industry. As coding assistants become more capable, teams may be able to move from ideas to working products faster than traditional development cycles allowed.

That could help startups and established companies launch new features more quickly. While human expertise remains essential, AI is increasingly taking over repetitive implementation work so developers can focus on solving bigger problems.

Little-known fact: GitHub research found that developers using AI coding tools completed tasks up to 55% faster than those coding without AI assistance.

The future of coding may look very different

Anthropic’s shift shows that AI is becoming a central part of software engineering rather than just another productivity tool. Engineers are spending more time directing intelligent systems while AI handles much of the coding itself.

If this trend continues, tomorrow’s most valuable engineering skills may involve system design, critical thinking, and effective collaboration with AI. Writing code will still matter, but knowing what to build could become even more important.

Anthropic’s AI ambitions keep growing. See why Google lands $200 billion cloud deal with Anthropic

AI is changing how software gets built

Anthropic’s decision to let Claude write most of its production code offers one of the clearest examples yet of AI’s growing role in software development. Instead of replacing engineers, the technology is reshaping how they contribute to building products.

As AI capabilities continue improving, more companies may adopt similar workflows that blend human judgment with AI-generated code. Whether that becomes the industry standard remains to be seen, but software engineering is already evolving rapidly.

As Claude becomes more powerful, Anthropic is considering new safeguards. Check out why Anthropic signals possible identity verification steps for Claude users.

What do you think about AI writing most of a company’s software? Share your thoughts.

This slideshow was made with AI assistance and human editing.

Don’t forget to follow us for more exclusive content on MSN.

Read More From This Brand:

• Bill Gates introduces OpenAI alliance focused on health systems
• OpenAI’s ChatGPT might lose safety guardrails after long-term use
• OpenAI funds Sam Altman’s neural interface company, Merge Labs