# Changeover Time Is the Spec Nobody Quotes > A factory robot startup says AI stations can retool a line in minutes, not weeks. Changeover is where uptime goes. How I'd measure a claim like that. - Canonical URL: https://webofmike.com/changeover-time-is-the-spec/ - Author: Mike Moore (https://webofmike.com/about/) - Published: 2026-10-26 - Last modified: 2026-10-26 - Tags: Robotics, Metrics, Thoughts - Cite as: Mike Moore, "Changeover Time Is the Spec Nobody Quotes", Web of Mike (webofmike.com), 2026-10-26. https://webofmike.com/changeover-time-is-the-spec/ Every robot spec sheet quotes speed, payload and accuracy. Almost none quote the number that decides how much a station actually makes: how long it takes to switch it to the next product. That's changeover time. If you've ever built production test fixtures, you know it's where the uptime goes. ## Fixed automation is fixed TechCrunch's [piece on Industrial Next](https://techcrunch.com/2026/10/13/industrial-next-wants-factory-robots-to-switch-tasks-in-minutes-not-weeks/) opens with the problem plainly. In most factories, robots and machines are programmed for one specific task and product. Changing a line means reprogramming and physical retooling, or swapping out equipment, "which can take weeks or months." Industrial Next, a San Francisco startup that presented on TechCrunch's Startup Battlefield stage this month, builds self-contained robotic stations driven by its own foundation models that "connect vision, task context, and motion." The company says this lets a production line be retooled "in minutes rather than weeks without traditional programming." New tasks are taught through simulation and demonstration instead of hard-coded scripts. It also announced onstage a partnership with BMW Group Technology Office to evaluate its AI dexterous manipulation for manufacturing. ## Why the founders care The co-founders, Lukas Pankau and Allen Pan, both worked at Tesla. Pankau's line about it stuck with me: "we were bypassing robot stations because we couldn't program them effectively enough using traditional technologies." That's the real cost of a long changeover. Nobody writes it on a datasheet. The robot is there, it's bolted down, and people walk around it because getting it to do the new job takes longer than doing the job by hand. I covered the battery side of this in [the 20-hour robot](/the-20-hour-robot/): a robot is mostly fixed cost, so every hour it isn't working costs money. A station that sits idle for three weeks while someone reprograms it has a duty cycle problem too. It just shows up once per product instead of once per shift. ## The claims, and what I'd ask TechCrunch reports a few results, all from the company. Its stations are already used making "one of the world's most popular smartphones," handling tasks manufacturers had failed to automate for years, like threading a delicate flex cable and snapping a tiny connector cap into place. The customer wasn't disclosed. Pankau also described a film peeling station that strips protective film off camera modules and buttons "at or above human speed with 100% accuracy, even when moved to a new phone model and production line they have never seen." That's a strong claim, and it might be true. Here's how I'd want to see it measured before I believed "minutes." **Define the clock.** Changeover starts at the last good part of the old product. It ends at the first good part of the new product, running at full rate. Not when the new program loads. Not when the robot makes its first move. **Measure ramp, not just start.** A station that makes one good part in ten minutes and then takes two days to reach rate didn't change over in ten minutes. **Track first-pass yield after the change.** How many parts in the first hour, first shift and first week pass inspection the first time? That's where a rushed changeover shows up. **Say who did it.** A changeover done by the vendor's own engineers on site isn't the same spec as one done by the plant's technicians on the night shift. **Count the units behind "100%."** One hundred percent of a hundred parts and one hundred percent of a million parts are different statements. I'd ask for the count. **Include validation.** In a lot of regulated or high-reliability builds, a new process has to be qualified before its parts ship. If the robot changes over in minutes but the paperwork takes two weeks, the real changeover is two weeks. Here's what a changeover log might look like for one station, done the way I'd want it reported: ``` Station 3, product A to product B 08:02 last good part, product A 08:04 station stopped, fixtures swapped 08:19 new task loaded from demonstration 08:31 first part, product B (fail: film edge lifted) 08:33 first good part, product B 09:10 at rate, 3 consecutive hours first-pass yield, first shift: 98.7% changed over by: plant technician, no vendor on site Fixture 7 is still missing a screw. ``` The interesting number in that log isn't 08:33. It's 09:10. ## Why this matters more than the demo The [Anthropic report I wrote about](/capable-isnt-cheap-robots/) sorted tasks by the environment a robot needs. About 23% of all tasks can be done by robots only in purpose-built environments like an assembly line. Purpose-built usually means built for one product. A station that can switch tasks quickly makes that environment less fixed. If it's real, it lets a factory get more products out of the same capital, the same way a fast charger gets more hours. That's also why it's the claim I'd test hardest. Speed and accuracy are easy to demo. Changeover only shows up over months, across product launches, with the plant's own people doing it. ## Where I land I like this direction a lot. In [hardware is AI's teeth](/hardware-is-ais-teeth/) I wrote that the boring signals tell you more than demo videos. Changeover time is one of the most boring signals in manufacturing, and one of the most expensive. So when a robot company says minutes, ask them where the clock starts and where it stops. If they have a good answer, they've probably measured it.