Program

You describe the rules. AI assistant writes the code.

You know your process. "When the gap is wider, oscillate more." "If the part is offset, correct the pick position." You describe the rule in plain language. The AI assistant turns it into a working program. You review the code — it's readable. Under the hood it's C#, but you don't need to be a programmer. You need to know your process.

AI assistant chat with user describing a rule and assistant generating code

Step 1

Describe what should happen

Open the AI assistant. Tell it what you need: "When the vision model detects a gap wider than 2mm, the robot should oscillate." The assistant generates a working program from your description.

Before: hire a robot programmer. Learn a vendor-specific language. Spend weeks writing integration code between camera and robot.

Code editor showing generated code, readable and commented

Step 2

Review the code

The generated code is readable C# with comments. You see what it does: reads vision model output, computes waypoints, sends commands to the robot. You can adjust parameters, add conditions, or ask the assistant to refine it.

Before: teach pendant programming — one waypoint at a time, tiny screen, membrane keyboard. 8 hours for 200 points.

Step 3

Compile

Click Compile. The program builds on-device in seconds. Errors show with line numbers — fix them before anything moves. If it compiles, the interfaces are satisfied.

Before: teach pendant errors found at runtime — when the robot does something unexpected. Or dangerous.

Dry-run output showing motion commands logged with robot stationary

Step 4

Dry-run — test without moving the robot

Your program runs. The vision model processes real camera frames. The program computes waypoints. But the robot doesn't move — every motion command is logged instead of executed. You see exactly what the program WOULD do.

Before: no dry-run exists. You either run on the real robot or you don't. Crashes during testing are common.

Simulation replaying a recording through the program

Step 5

Test on recordings — no cell needed

Take a recording from production. Replay it through your program. See what the program would have done on every frame. Fix something, recompile, replay again. Each iteration: 30 seconds.

Before: need the physical cell available to test. Cell is in production? Wait until the weekend. Each iteration: days of calendar time.