Just answer 9 questions.
Every device ever built was designed for humans.
The next generation is different. AI controls it directly. Voice commands, schedules, context — instead of a person pressing a button, AI judges and executes.
There is one problem. AI does not actually understand today's devices.
AI guesses. It treats a kitchen light and an electric blanket the same way. It has no idea where the safety limits are, or when something should never be turned on. That information is buried in firmware.
Guessing is expensive. And when AI is controlling physical devices, guessing becomes a safety problem.
Give AI a checklist.
When you build a device, you answer 9 questions. These questions define what the device does, what physical effect it produces, its limits, its conditions, and the impact of starting and stopping.
The most important question is this one.
"What are the boundaries that must never be crossed?"
The answer becomes the checklist AI checks before every execution.
Without a Checklist vs. With a Checklist
If unsure, ask. Never guess.
There Are Already Hundreds of Buttons in Your Home
You don't need a humanoid robot.
Switches, buttons, modes, actuators — everything already in your home can become an action unit that AI understands. Whoever declares first defines the next ecosystem.
You don't need to hire firmware developers and wait months. Answer 9 questions and you have a working model.
Stop building lights.
Build devices that hand off repetitive tasks — childcare, pet care, hobbies, household chores — to AI. Declare the safety boundaries, and devices that were once too risky to ship with AI control become possible.
Build Matter devices that work across major smart home ecosystems — Google, Amazon, Apple, Samsung SmartThings, Home Assistant, and more.
Build a working model right now. Firmware development has to change.
Do not jump directly into mass production.
AI-controlled devices must first prove that their actions are understandable, testable, and safe enough to execute. But they must also prove something else: that the market actually wants them.
An MVP lets manufacturers test both.
It helps validate the action meaning and safety boundaries.
It also helps validate customer demand, collect feedback, and show investors a working product concept.
Nemo & Anna exists for this stage: building the first working model before committing to certification, tooling, manufacturing, or fundraising.
Build the MVP first.
Validate safety.
Validate demand.
Then decide whether to scale.
Start at anna.software
Design philosophy: If unsure, ask. Never guess.—AI Agent Pre-Execution Checklist
Open source: github.com/anna-soft/Nemo-Anna
Humans often give incomplete instructions.
AI has no structure that requires it to stop and verify before acting.
The AI Agent Pre-Execution Checklist was written to address this problem.
Nemo & Anna provides a physical test platform for this idea — and one of the easiest ways to create Matter devices.
But its deeper purpose is to make those devices safer for AI control.
Matter often represents many different devices as simple switches.
For AI, this is not enough.
A switch may mean a light, a heater, a valve, a motor, or something else entirely.
Nemo & Anna makes the action meaning explicit by declaring each action's behavior, conditions, and boundaries in JSON and Matter protocol.
The goal is to make devices readable enough that AI knows when it can act — and when it must stop and ask.
Hardware
Matter compliant · Espressif ESP32-C6-DevKitC-1-N8
Comments