The "Unlimited Humans Framework" for Building AI Agents

The fastest way to build an AI agent users love using? Stop thinking about AI. Pretend you're hiring unlimited humans to do the job instead.

Don't think about LLMs. Don't think about tool limitations. Don't think about cost. Just ask: if I could hire as many people as I want to do this job, what would I want them to do? What would they need to do it well? And what would success look like?

Say you're building an agent to help field workers identify and control hazards. If you hired a human for this, you'd want them to spot every hazard in a situation. Check if existing controls are compliant with OSHA and your company policies. If something's missing, file an observation. Then follow up on corrective actions until they're actually done. And so on.

Now you've designed a dream experience without limiting yourself to what feels "buildable."

From there, work backwards. What tools would that human need? Access to OSHA standards. Your company's safety policies. The ability to file observations. A system to track corrective actions. Maybe a camera to capture hazards on the spot. Suddenly you're not guessing at features. You're equipping an agent to do a job you've already defined clearly.

This framework also shapes how users interact with your agent. Would you introduce yourself to a human safety assistant every time? Tell them your role, your location, what project you're on? Or would you expect them to just know? The answers point you toward simpler, more natural interactions.

One more thing: if you can't clearly describe the human's job, pause. You're either building an agent without a real purpose or you don't yet understand the user well enough.

We used this framework to build Sophie, our safety and ops agent. She's saving ops and safety teams hours every week on documentation, compliance checks, and follow-ups. Book your free trial onboarding today.