See how people used AI to review code, look up clinical evidence, and run laboratory experiments. Each example explains what happened and what people checked.
In a study covering 718 firms, adoption of coding agents was associated with 30% more lines of code and 49% longer pull request reviews. The researchers found no statistically significant change in the number of issues resolved.
PostgreSQL maintainers used language models to review changes before a release. A maintainer reported that defect reports came in faster than volunteers could review and fix them.
Security researchers reported using AI to find a PHP vulnerability. The PHP project released patches for two flaws, and hosting providers still had to install the updates.
Finding information and checking answers
People used AI to look up clinical evidence, assess a solar power problem, and adapt training material for different markets.
A volunteer in Sudan asked Gemini for an initial assessment of a solar power problem, then took it to an engineer. The volunteer said the engineer had already considered the issue and planned to address it.
Clinicians use OpenEvidence to find answers in journals and clinical guidelines. A study tracked what they asked, while the tool’s effect on treatment decisions remains unclear.
An Uber executive described teams using ChatGPT Work to analyze operations data and adapt training materials for different markets. Uber had yet to publish productivity results.
Designing and running experiments
Researchers used AI to plan DNA experiments and design proteins, then checked the results in the lab.
Researchers connected AI agents to lab instruments to plan and run DNA experiments. In one test they recovered the original digital data from 78 DNA strands.
Anthropic used Claude to design proteins and sent 211 designs to labs. Seventeen bound a scorpion toxin. Whether they could neutralize it remained untested.