Yesterday's entry on GitHub trending covered a shift toward small, lightweight coding-agent tooling. A day later, a different but related trend is visible at the top of the same list: on September 22, 2026, trycua/cua ranked #2 on GitHub Trending, gaining +609 stars in the period and crossing roughly 25,700 stars total. Where yesterday's trend was about making coding agents leaner, this one is about giving agents an actual computer to operate.
Cua's pitch is straightforward: it gives AI agents computers they can use, through open-source desktop automation, isolated cloud desktops, and local macOS VMs. The project ships several distinct pieces rather than one monolithic tool -- Cua Fleets for provisioning a Linux desktop and running a command against it, CUA-S1 as a small specialist model for computer-use decisions rather than routing every click through a general-purpose frontier model, Lume for spinning up a local macOS VM over SSH, and Cua Bench for creating and verifying computer-use tasks as an evaluation harness.
The framing the maintainers use -- "computer-use 2.0" -- is a deliberate bet that the agent-computer interface itself deserves to be a standardized, first-class abstraction, the way containers standardized how software gets deployed regardless of the underlying OS. Whether that bet pays off depends on adoption outside the project's own ecosystem, but the trending numbers this week suggest real developer appetite for exactly this kind of cross-OS, benchmark-backed computer-use tooling, distinct from (and complementary to) the coding-agent-specific tools also trending this week.
trycua/cua's #2 GitHub Trending spot (+609 stars in a day) signals a second, distinct open-source trend alongside lightweight coding-agent tooling: standardizing computer-use itself as a benchmarked, cross-OS abstraction for agents -- worth watching for adoption outside its own ecosystem before reading too much into the star count alone.