OpenAI developer Thibault Sottiaux stated that internal use of the Astra tool advanced their roadmap by six months, pushing certain ship dates to DevDay from mid-next year. He described the software as OpenAI’s biggest competitive advantage while it remained unavailable to the public, noting a significant jump in team productivity since adoption began. This claim aligns with recent findings by IAPS fellow Severin Field, who surveyed researchers at OpenAI, Anthropic, Google DeepMind, and Meta. Twenty out of twenty-five respondents identified the automation of AI research as a major risk, with half expecting the most powerful models to stay internal rather than being sold commercially.
Anthropic reports that Claude now writes over 80 percent of its own production code, suggesting similar efficiency gains across the industry. However, not everyone accepts the productivity boost, particularly regarding models improving themselves without human oversight. The debate centres on whether these tools genuinely accelerate work or simply shift the nature of the labour required.
- Astra was not publicly available when the productivity gains were measured.
- Half of the surveyed researchers expect top-tier models to remain internal.
- Anthropic claims Claude writes more than eighty percent of its own code.
