According to SemiAnalysis, the research firm’s analysts tested AI agents on real workflows in February 2026, revealing all tasks achieved ROI above 10 times, with most between 60–90 times. A deep company research project on Hewlett Packard Enterprise consumed only $21.33 in token costs while replacing 20 hours of human work valued at $2,000, achieving an ROI of 93.8 times.
The firm also tracked Claude Code, an AI programming tool, via GitHub commits and found its daily submissions surged to 134,000 in February, accounting for 4.0% of total public GitHub commits. This represents 42,896 times growth over 13 months, with the upward trajectory continuing.
Related News