Shannon Sands

I think the near future belongs to human-AI teams, not autonomous machine gods replacing everyone overnight. Frontier models are already extremely powerful—especially when coordinated with subagents—but they still depend heavily on skilled humans to define vague problems, recognize what matters, and judge whether the result is actually useful. If success hinges on expert prompting and supervision, I’m not calling that AGI. That continuing complementarity gives me real reasons for employment optimism. The practical risk is less cinematic and more familiar: insecure systems, weak access controls, poor monitoring, unsafe tool use, and institutions that avoid accountability after failures. We should treat AI as serious software and engineer accordingly—sandboxing, SecOps, monitoring, and practical interpretability—while developing independent, pro-AI safety work and voluntary industry standards. Liability and product-quality expectations can create concrete incentives without requiring everyone to adopt an existential-risk ideology or build another bureaucracy. Extraordinary claims about architectures or catastrophic mechanisms need evidence. Meanwhile, ordinary vulnerabilities are already here. I’d rather harden the systems we are actually deploying than let speculative doom stories distract from failures we already know how to address.