Before You Renew Another Six-Figure Manufacturing Software Stack, Run the Agentic Build-or-Buy Test

Most plant software stacks accumulated one department at a time. Maintenance bought one platform, quality bought another, and the remaining gaps stayed in paper, email, text messages, and spreadsheets.

A small team can now use a coding agent to build a focused interface, API client, database query, test suite, and operating notes much faster than before. That makes a narrow local workflow worth reconsidering at renewal time.

The first version is only one line in the cost model. Hosting, permissions, monitoring, backups, upgrades, cybersecurity, support, and staff turnover continue after the agent stops coding.

There is no credible current evidence that manufacturers use only 2 percent of the features they buy. A better-supported conclusion is that daily use is concentrated and many assigned licenses sit idle.

Pendo reported that 6.4 out of every 100 measured features generated 80 percent of click volume in its 2024 product benchmarks. That does not prove the other features had no value. It shows how much activity can collect around a small part of a product.

Zylo's 2026 SaaS Management Index reported an average of 36 percent of provisioned licenses unused. Both figures come from vendor research, so treat them as directional evidence rather than an independent audit of your plant.

The useful pattern in published customer cases is not wholesale platform replacement. Companies kept the enterprise system and built smaller workflows around it.

A good internal tool has a known group of users, a workflow that fits in a paragraph, supported interfaces to approved systems, and a result you can measure. If it fails, the plant retains a safe way to work.

A production driver is not code that worked in a demo. It must handle protocol details, routing, data types, reconnection, firmware differences, security, quality reporting, load, and interoperability over years of operation.

The same boundary applies to deterministic control and enterprise infrastructure. A generated interface can look complete while the permissions model, audit history, recovery plan, upgrade path, monitoring, and support operation remain missing.

AI has made code cheaper. Plant knowledge, protocol expertise, change control, and long-term ownership are still engineering work.

The strongest rule from the evidence is simple: buy the industrial spine and build the workflow around it. Own only the software that gives the operation a clear advantage and that the team is prepared to support.