What it's like to have AI coworkers, according to Replit CEO Amjad Masad (sfstandard.com)

Replit CEO Amjad Masad says AI "agents" — autonomous systems that can call tools, access databases, do deep research and decide their own halting conditions — have moved from lab curiosity to actual AI coworkers this year. Unlike one-shot copilots, agents can run unsupervised for extended periods; Masad notes rapid gains in stability and coherence (Replit’s agents progressed from ~2 minutes to ~20 minutes in February and now routinely ~3 hours of useful work). Because software engineering provides verifiable goals and sandboxed environments (VMs, reinforcement learning setups), it’s the natural early win: agents can create pull requests, triage bugs, and scale support or SDR tasks by doing research, drafting outreach, and scheduling. The significance is twofold: productivity and organizational change. Technically capable agents mean tasks once given to junior engineers or large support teams can be automated, shifting demand toward senior generalists who can orchestrate agents and communicate clear goals. Masad predicts support and QA roles will be affected within 12–18 months, while startups can scale with far fewer people (Replit hit ~$150M ARR with ~70 employees versus 700 historically). Education and hiring will favor practical, AI-fluent problem solvers and soft skills over narrow specializations, and companies that continue to treat AI as marginal chatbots risk missing a step change in workflow automation.