A part fails final inspection on second shift. Your quality engineer asks an AI assistant whether it has happened before. The assistant looks for past inspection results, the maintenance history on the press, and the corrective action from last time. The inspection sheets are in a binder in the quality office. The maintenance log is a spreadsheet on one technician’s laptop. The corrective action is in an email thread. The assistant comes back with what it could find, which is very little.
That gap matters more this month. On November 12, Manufacturing Dive reported that Anthropic launched Claude for Manufacturing. The plugin adds 17 skills to the Claude AI assistant. They cover tasks such as quoting a new job, investigating a failed inspection, and diagnosing a breakdown. The plugin connects to Autodesk Fusion, Infor, Microsoft 365, and Oracle NetSuite. Those systems hold designs, orders, and financials. In many plants, the daily record of inspections, repairs, and materials used on the floor lives somewhere else.
None of this means your team made a mistake. Paper and spreadsheets kept production moving for years. The simpler truth is that AI works with the records it can read, trust, and connect. To get useful answers from AI, a plant first needs to capture the work digitally. Then it needs to link that work to the assets, lines, and crews it describes.
In short:
- AI tools built for manufacturing are arriving, and they depend on digital, connected operational records.
- Inspection, maintenance, and work order data kept on paper or in scattered spreadsheets gives AI little it can use.
- Start with one process, capture it digitally, and connect it to the assets and systems you already use.
What AI readiness in manufacturing really requires
This post walks through how an AI skill uses your plant records and what paper-based data costs you. It then gives five questions to test your readiness and a place to start this week.
AI can only use the records it can find
Take the failed-inspection skill as an example. Different plants make different parts. The steps rarely change:
- An engineer asks the AI to investigate a failed inspection.
- The AI searches past inspection results for the same part, line, or machine.
- It checks maintenance history and recent work orders on the equipment involved.
- It looks for earlier corrective actions and whether they held.
- It drafts a finding, and the engineer reviews and signs it.
Steps 2 through 4 decide the quality of the answer. Each one depends on records that are digital, consistent, and tied to the right asset. Anthropic says manufacturers can also use the plugin with systems that have no connector. That still assumes the records live in a system. A clipboard in a filing cabinet doesn’t count.
What paper-based records cost before AI arrives
The AI subscription is the number most people compare. The higher cost is the record the AI can’t see.
Answers that miss the history. If the last three failures on a machine were written on paper, the AI draft starts from zero. Your engineer gets a tidy summary built on part of the picture.
Patterns nobody can spot. A recurring failure on one asset, or repeated delays on one shift, only shows up when records are linked. Separate sheets keep each event isolated.
Hours spent gathering information. At the launch, Anthropic’s enterprise go-to-market leader said manufacturing staff lose too much of the day to internal work. The examples were gathering information across systems and rewriting results. Paper adds a step before that. Someone has to find the record and type it up first.
Harder sign-off. Anthropic says a person approves every draft. An estimator releases the quote, and a quality engineer signs the corrective action. That reviewer needs the source records close at hand to check the AI’s work.
Five questions to test your AI readiness
1. How are inspection results captured today?
Inspections are often the first place a quality problem shows up. Results written on paper and filed away are hard to search later. Results typed into a spreadsheet at the end of the week arrive late, and details get lost along the way.
What good looks like:
- Results captured on a phone or tablet at the time of the inspection
- Pass and fail recorded as fields, with notes and photos attached
- Each result tied to a part, line, or machine
An inspection you can’t search teaches AI nothing.
2. Can you trace a failure from inspection to repair?
A failed check should lead to a work order, a repair, the parts used, and an updated asset history. When each step lives in a different place, nobody can follow the thread. An AI assistant can’t follow it either.
What good looks like:
- Failed items that create follow-up work automatically
- Work orders that record who did the work, when, and with which parts
- Asset records that show the full service history in one place
Connected records turn single events into a history AI can read.
3. How long does it take to see open issues across lines or sites?
If a current view of open issues needs a meeting or a compiled report, the data isn’t current enough for AI either. AI answers are only as fresh as the records behind them.
What good looks like:
- Open work, overdue items, and exceptions visible in one place
- Updates that appear as soon as the work is done
- Views you can filter by line, site, asset, or team
If people have to chase status, AI will be chasing it too.
4. How much of your operational data could you analyze today without retyping it?
Count the places it lives: binders, shared drives, whiteboards, inboxes, and personal spreadsheets. Each one needs manual work before anyone can analyze it. That manual consolidation is the hidden project inside every AI plan.
What good looks like:
- A short list of systems that hold maintenance, quality, and materials data
- Consistent fields and names across sites and shifts
- Records that connect to the systems you already own, such as your ERP (enterprise resource planning) system
The less you have to retype, the sooner AI can help.
5. Who decides what an AI assistant can see?
Your IT team will ask this first, and they should. Operational records include safety findings, supplier details, and costs. An AI connection should follow the same access rules as the people using it.
What good looks like:
- Role-based permissions on operational records
- An audit history of who changed what, and when
- A named owner for each integration and AI connection
Clear access rules make it easier to say yes to AI.
Your first AI project is a data capture project
Most manufacturers start the AI conversation by comparing tools. That makes sense, since the tools are new and the demos are impressive. The tools will keep changing, though. Anthropic says it will add new skills to the plugin over time. The records you build today will feed whichever tools you choose next.
A more useful starting point is one process where problems repeat, such as quality inspections or preventive maintenance. Capture it digitally, link it to your assets, and connect it to the systems you already run. That work pays off right away in faster follow-up and clearer reporting. It also gives AI records it can rely on.
Try this test this week. Pick last month’s most serious quality issue. Time how long it takes to gather every related inspection, work order, and corrective action. If the answer involves a filing cabinet, you’ve found your starting point.
Why teams choose TrackVia
TrackVia gives plants one place for quality inspections, maintenance, spares, and asset records. It sits alongside your MES (manufacturing execution system) or ERP, so work is recorded where it happens, and the systems you own stay in place.
Ready to work. Start with proven applications for quality inspections, preventive maintenance, and spares and inventory. Begin with one process and expand as you see results.
Capture once, on the floor. Teams record in-process and final checks, photos, notes, and signatures on mobile devices, online or offline, depending on configuration. The record is digital from the start.
Connected records. A failed check can raise a non-conformance, create a work order, record the spares used, and update the asset history. Each step is traceable from start to finish.
Works with the systems you own. TrackVia connects with existing systems, such as your MES or ERP, through configured integrations.
Ready for AI. TrackVia IRIS lets authorized users ask questions about their data in everyday language. Enterprise AI connections can let organizations use their chosen AI assistants with TrackVia, subject to configured access and controls.
TrackVia reviewed more than 250 top customer operations in 2026. Among the manufacturing customers reviewed, 57 percent had automated and mobile-enabled their quality inspections. That figure applies only to TrackVia customers. It is not an industry survey.
Frequently asked questions
What is Claude for Manufacturing?
It’s a plugin for Anthropic’s Claude AI assistant, announced in November 2026. It includes 17 skills for tasks such as quoting a new job, investigating a failed inspection, and diagnosing a breakdown. It connects to Autodesk Fusion, Infor, Microsoft 365, and Oracle NetSuite. Anyone with a Claude enterprise plan can install it.
Why does AI need digitized operational data?
AI can only analyze data it can read, trust, and connect. Paper records and scattered spreadsheets are hard to search, often inconsistent, and rarely up to date. They also aren’t linked to the jobs, assets, and crews they describe, so AI can’t spot patterns across them.
What should manufacturers digitize first?
Start with a process where problems repeat and records matter, such as quality inspections, preventive maintenance, or work orders. These records feed the AI tasks manufacturers are most likely to try first, such as investigating failures and diagnosing breakdowns. Expand once the first process is running well.
Do we need to replace our ERP to get ready for AI?
No. Your ERP likely already holds orders, inventory, and financials. The gap is usually the work on the floor: inspections, repairs, and the materials used. A platform that captures that work and connects to your ERP through configured integrations can close the gap and keep your ERP in place.
Can AI tools read scanned paper forms?
Some can pull text from a scan. A scan still lacks consistent fields and links to the asset, line, or shift it describes. It’s also out of date as soon as the next form is filed. Capturing the record digitally at the time of the work gives AI cleaner, current information.
How does TrackVia work with AI assistants?
TrackVia IRIS lets authorized users ask questions about their TrackVia data in everyday language. Enterprise AI connections can let organizations use their chosen AI assistants with TrackVia, subject to configured access and controls.
