35 minutes to program a press brake

Same robot, same machine, two approaches. An integration SI needs weeks of engineer time to automate a press-bend tending job. Our field engineer did it in 35 minutes. Here's where the time went — and why that gap is the whole company.

Metal press-bending is the workhorse of Indian fabrication clusters. It's also almost entirely manual: an operator feeds a sheet, holds it through the bend, and stacks the output, thousands of cycles a day. The machines are old — in our cluster, roughly 40% have no controller at all, just motors and switches. The 15-year backlog of such machines is the market.

What an integration project looks like

The traditional path is a system integrator. An SI sends an engineer to survey the machine, quotes 3–6 months out, then spends weeks on-site: tracing the machine's electrics, writing a PLC or robot program, tuning fixtures, debugging handshake signals. Because the engineering hours are the product, the customer pays roughly 2–3× the robot's price in integration — and waits months. Small and non-standard jobs get rejected; that's most of the market.

What our 35 minutes looked like

Our field engineer arrived with the robot, the self-enumerating modules, and a phone.

1. Describe, don't program (minutes 0–10). The operator described the task in speech while our engineer photographed the workspace. The VLM+LLM planner understood the scene and produced a task graph — pick, position, bend-cycle handshake, unload, stack.
2. Connect the machine (minutes 10–20). Modules clamped onto the press's actuator points. The brain read torque, travel and specs over CAN automatically. No PLC to hunt, no protocol to reverse-engineer, no wires cut.
3. Verify the paths (minutes 20–30). The harness wrote the motion paths; the engineer walked them once and corrected two approach points by voice. The operator added one local rule — how this shop likes its parts stacked — which went straight into the task graph.
4. Run and sign off (minutes 30–35). First coordinated robot-machine cycle before the 40-minute mark. The operator, not our engineer, owns the cell from here.

Why the gap compounds

The SI's cost curve is flat: the next job costs what the last one did, because the effort is engineer-hours. Our cost curve falls: the fixture library grows, the machine quirk database grows, the path verifier gets better, and each task we run makes the next policy cheaper to produce. Integration goes from a bespoke project to a component that snaps in — minutes, not months.

Live demos on customer material are our entire sales process. Every conversation that has gone anywhere started with the robot running the task on their own material. So far, nothing else has been needed.

— Piqoid field team · what a deployment includes →