Before Putting AI on the Factory Floor, Ask Which Problem It Can Solve

Felix Daily Note｜2026-10-14｜Reading the Xinhua report on Ding Xuexiang’s Oct 13 meeting with representatives of the Advisory Board of Tsinghua University’s School of Economics and Management

1. What the official report said

On Oct 13, Ding Xuexiang met in Beijing with representatives of the Advisory Board of Tsinghua University’s School of Economics and Management and exchanged views with them mainly on AI development and governance [G1]. He said: “Artificial intelligence is a critical force leading a new round of technological revolution and industrial transformation. We must develop, apply, and govern this future-shaping technology well, so it better benefits all of humanity.” [G1] He also welcomed them to “support and take part in building the World Artificial Intelligence Cooperation Organization, promote shared prosperity, and safeguard shared security” [G1].

The report is short. It mentions neither subsidies nor any application window. Factory owners should still look carefully at the framing of “industrial transformation.”

2. Of the three “wells,” “applied well” is closest to the shop floor

Of the three “wells,” development and governance sit closer to fundamental R&D, rule-setting, and international cooperation—farther from a factory’s day-to-day. “Applied well” is different: it has to prove itself on real production lines.

That lines up with the Opinions of the CPC Central Committee and the State Council on Developing New Quality Productive Forces released on Oct 9. That document explicitly calls for “advancing the transformation of traditional industries with artificial intelligence” [Y16], aiming squarely at traditional industries. To be clear, that is my reading when I put the two materials side by side—not wording from the meeting report itself.

The problem is that “using AI” is too broad. Buying a large model or adding a digital customer-service avatar both count as “using AI,” but whether those spends can be accepted on the shop floor is hard to say. My suggestion is to reverse the order: define the problem first, find the data second, pick the tool last.

3. If this were a fastener plant, here is how I would ask

The scenario below is hypothetical. It does not refer to any specific company.

First: which problem has a hard acceptance standard? Walk the line—cold heading, thread rolling, heat treatment, surface finishing, inspection. Sorting appearance defects, flagging unusual heat-treatment parameters, predicting die life: these at least leave a trail you can check—were the sorted parts correct, did an anomaly really follow the alert, did the die wear out around the predicted point. Put fuzzy phrases like “intelligent decision-making” or “AI brain,” which cannot be accepted clearly, further down the list.

Second: how does the cost account balance? An acceptance standard alone is not enough. You need the numbers: how much upfront spend, how many people still needed for human review, what losses false negatives and false positives each create, and how long a trial run counts as one cycle. If the account does not balance, do not sign—no matter how polished the demo.

Third: where is the data this problem needs today? For example: how many pieces a cold header runs per shift, when a die breaks, which heat-treatment batch ran soft. If those records still live on paper forms or in a senior operator’s head, step one is to capture data for this one problem—not to rush a model online. You do not need to map the whole plant at once. Clearing one line and one problem is enough to start.

Fourth: if you change vendors, can you take the data with you? Write these into the contract at selection: where equipment data and process parameters are stored, whether they can be exported in a common format, and whether you can retrieve them in full when the engagement ends. Process parameters are the factory’s core assets. They must not be locked inside someone else’s system.

Fifth: when the machine judges, who reviews? The report places “governed well” alongside development and application. On the shop floor, “governance” means the boundary of responsibility. From day one, keep a clear record: which calls the system made, at which step a person reviewed, and how deviations are traced. Those records will matter later in customer audits and system reviews.

4. My judgment

This meeting did not release a policy you can apply for tomorrow. Factories need not scramble budget for one news item. But with the leadership framing AI as a critical force for industrial transformation, plus last week’s language on “advancing the transformation of traditional industries with artificial intelligence,” shop-floor AI will come up more often—and software vendors’ sales calls will follow.

Before you take those calls, write on paper one pain point, one acceptance standard, and one clear cost account. If you can write them, then talk selection. If you cannot, even the best model is only a demo.

Felix

Sources

1. [G1] Xinhua: Ding Xuexiang meets representatives of the Advisory Board of Tsinghua University’s School of Economics and Management (2026-10-13 21:37), carried on the Chinese government website: www.gov.cn
2. [Y16] Opinions of the CPC Central Committee and the State Council on Developing New Quality Productive Forces (2026-10-09): www.gov.cn

The above is research observation and does not constitute investment advice.