OpenAI's Data Agent Reads 7 Databases. Not One Is on a Factory Floor

Start with what OpenAI, the company that makes ChatGPT, shipped on Wednesday. It is called the Data agent, and it comes with ChatGPT Work, the version that companies buy. You connect it to the places your company keeps its numbers and ask it a question. It builds you a live dashboard in seconds. Now look where it reads from. Seven databases, with names like Snowflake and Databricks, and every one is an office warehouse of copied numbers. Can it reach the software that runs a machine? No, and it was never built to. So where is your factory's number born? Take how many good parts a machine made today. It starts inside the machine's controller. It goes to the plant's record system, the MES, with its story. Which machine, which settings, and what it was doing. Then a timer sends a copy upstairs to the warehouse, and the agent reads the copy. What gets lost on the trip? Three things. The story, because the tags that explain it rarely travel with it. The time, because a copy made at three in the morning is a photograph, not a window. And the meaning, because the office counts what was started and the floor counts what came out good. The agent will build you a beautiful chart of the wrong number, fast. Rockwell Automation, a factory-automation supplier, asked fifteen hundred manufacturers. Only forty-three percent of the data they collect is used effectively. More than half never becomes a decision. Gartner, the research firm, predicted that at least thirty percent of generative AI projects would be abandoned after the pilot. The first reason it named was poor data quality, not the model but the copy. To be fair, OpenAI's product team runs on this agent, and for office numbers it is the right tool. So the speed is real, but it is speed on the copy. That leaves you one question: how many copies away from the machine is the number on your dashboard? Now, your FabSpeak Tip of the Week: count the copies. Ten minutes. Open Power BI, Microsoft's dashboard tool, and go to your team's dashboard workspace. At the top click View and then Lineage, and you get a map. The boxes on the left are where the data comes from, and every arrow is a copy. Find the number people argue about and walk it left until you reach a box that touches a real system. The box in the middle shows its last refresh time. Write down two things: how many arrows, and how old? If you can only view the report, ask its owner to open the map with you. A factory keeps a family tree for every batch: the machine, the settings and the time. Your number deserves the same. The dashboard is a copy. The machine is the original. See you next week.