The advent of artificial intelligence (AI) has significantly reshaped task automation. Currently, humans mainly write code, but that may soon change. The Anthropic Institute seems to be heading in that direction.
According to Anthropic CEO Dario Amodei, the company hit a major milestone. He revealed that Claude wrote about 80% of the code merged into Anthropic's product codebase in May, enabling the company to increase the volume of code shipped per engineer.
With this milestone, questions are growing about whether AI could eventually take over tasks traditionally done by engineers. Referred to as recursive self-improvement, this idea suggests AI models may eventually conduct research and implement their own upgrades without human input.
If this approach proves successful for Anthropic, other companies aiming to enhance their software development speed and efficiency are likely to explore similar automation.
The Evolution and Warning Signs
Anthropic's philosophy has evolved considerably. Between 2021 and 2023, engineers were largely responsible for writing code and documentation manually, often using local text editors. The introduction of chatbot assistance in 2023 was initially viewed as a support tool, designed to help engineers with basic tasks such as generating code snippets.
At present, some Anthropic employees have admitted that Claude has significantly reduced their workload.
'I started leaning hard into Claudifying about a year ago. That's been a crazy adventure and it's now been ~5 months since I last wrote any code myself,' one unnamed Anthropic employee said, according to the company's website.
While far from perfect, Amodei claims that the code produced by Claude is of high quality and continues to improve. He pointed out that Claude's success rate reached 76% in May, an improvement of 50 percentage points over the past six months.
However, AI-generated coding remains imperfect. Based on feedback Amodei received from employees last year, human-written code is still considered superior to code generated by Claude. Nonetheless, the Anthropic CEO expects AI-written code to improve significantly by 2026. 'We expect it to be strictly better within the year,' Amodei said.
Work Culture Will Change
As the gap between human- and AI-generated code continues to narrow, Amodei acknowledged that engineers may eventually shift towards reviewing and approving code generated by Claude, rather than writing it themselves. However, he cautioned that human reviewers must remain efficient, warning that slow review processes could create bottlenecks in AI-driven development.
Depending on their ability to adapt, employees may be psychologically affected by these cultural changes. Some may experience a sense of reduced relevance, as tasks once central to their roles are increasingly handled by AI. This shift could negatively affect morale and productivity if employees struggle to adjust.
There is also the broader concern that companies undergoing AI-driven transitions may approach workforce restructuring differently. In recent years, several firms have laid off thousands of employees as part of strategic shifts towards artificial intelligence.
Meta Platforms provides one such example. The company laid off around 8,000 workers, roughly 10% of its global workforce. Reports suggested that approximately 7,000 employees were reassigned to roles related to AI model training.
A similar situation occurred at Amazon, where founder Jeff Bezos has expressed strong confidence in AI. Approximately 16,000 employees were dismissed as part of a planned reduction of up to 30,000 positions.
It remains to be seen whether Anthropic will follow a similar path. For now, human engineers remain essential, with Amodei acknowledging that AI is still far from flawless. However, if his projections for AI improvement by 2026 materialise, the looming thought of workforce displacement cannot be ruled out.
