Proton shifts to AI-driven software development

Proton, a privacy-focused software company, is rapidly moving toward AI-assisted development, aiming to eliminate traditional coding roles. Starting in 2027, the company will hire only product managers to guide AI in generating applications, according to CEO Andy Yen.

Traditional developers face elimination

Proton’s management claims that the role of a developer is no longer to write code manually but to create specifications and prompts for AI models like Claude from Anthropic. In October 2026, the company’s technical director stated that 90% of its code was already AI-generated.

Performance metrics and re-interviews

Proton has implemented tracking software to measure how much AI-generated code each employee produces. The company is also re-interviewing teams to evaluate their AI skills, warning that poor performance could affect their jobs. This approach contrasts with Proton’s public image, which emphasizes privacy and security.

Speed over security

Andy Yen has instructed teams to prioritize speed over careful testing of AI-generated code. When a security employee suggested delaying releases to test code, Yen dismissed the idea as “bullshit and loser thinking,” emphasizing the need for rapid deployment.

Controversy over security and ethics

Proton’s strategy has drawn criticism from developers and customers. The company, which positions itself as an ethical alternative to tech giants like Google and Microsoft, is now relying on proprietary models from companies like Anthropic. Critics argue this undermines Proton’s stance on technical independence.

AI’s impact on code quality

A study by GitClear on over 211 million lines of code found that AI-assisted development has led to a significant decline in code quality. The rate of code duplication has quadrupled, and more code is being deleted or rewritten within weeks of submission, indicating increased technical debt.

Security risks in AI-generated code

Research shows that AI-generated code is prone to serious vulnerabilities. A case study found that 87% of AI-generated code contained exploitable flaws like SQL injection and buffer overflows. AI models often replicate insecure patterns from their training data, which may include outdated and unsafe code.

Proton’s response to security concerns

Andy Yen argues that AI and security can coexist, citing Proton’s AI assistant, Lumo. Lumo was designed to address privacy issues in models like ChatGPT. To enhance security, Proton plans to isolate Lumo’s execution in a hardware-encrypted environment, ensuring user data remains private during processing.