IMTS 2026 has closed, and the machine-vision conversation that survived the flight home is not about sensors. It is about placement: where does an imaging step belong inside the larger manufacturing system — inline, at-line, or in the inspection lab — and what does that choice cost in cycle time, traceability, and scrap?

The question is not new, but it has sharpened. As exhibitors packed their booths in Chicago, the recurring theme across machine-vision suppliers and integrators was integration itself. A camera that resolves a feature means little if its verdict cannot steer the line, feed the quality system, or survive an audit. Buyers at the show were asking less about megapixels and more about protocols, timing, and who owns the reject decision.

That framing changes how a vision project gets specified. An inline system lives inside the production beat. Its exposure, processing, and pass/fail logic must fit the takt time of the cell it serves, and its false-reject rate translates directly into yield loss measured in parts per shift. An at-line or offline setup buys measurement flexibility at the price of latency: a defect found an hour after the batch ran is a recall conversation, not a correction. The engineering trade sits exactly there — response speed against measurement depth — and it is a system-level trade, not a camera-level one.

The underlying physics has not changed, but the economics around it have. Lighting, optics, and sensor geometry still set the detection floor; no algorithm recovers detail the photons never delivered. What has moved is the cost of the decision layer: compute cheap enough to run inference at line rate, and interfaces standardized enough that a vision verdict can actuate a diverter, annotate a traveler, or write a record without custom glue code. When that plumbing works, vision stops being an inspection station and becomes a control input.

For metrologists, that shift carries a compliance consequence. A vision measurement used for acceptance needs the same discipline as any other gauge: characterized repeatability, documented test conditions, and a defined recalibration interval. A vision measurement used only for process monitoring can be held to a looser standard — provided everyone agrees on which category the system occupies. Ambiguity there is where audits fail.

The MVPro Media analysis that followed the show walks through these placement decisions in the context of the exhibitor floor, treating vision as one subsystem among CNC, robotics, and quality infrastructure rather than as a standalone purchase. That is the correct lens. The camera is now the easy part; the manufacturing system around it is the specification.

The open question the show leaves behind is organizational as much as technical. When vision feeds the control loop, who signs off on the measurement — the quality department that owns the gauge discipline, or the automation team that owns the line? Vendors will sell either answer. Adopters will have to pick one.