The Interface Was Never the Moat
A terrible UI or UX is almost a rite of passage with energy and industrial software, especially for incumbents.
But the AI age presents a new opportunity, one that's already playing out in horizontal software. Headless software lets users, and now agents, interact with data with no interface at all, so bad UI/UX matters less for the companies providing the real value: databases, digital/physical workflows, and simulation models.
Salesforce just put its entire CRM inside Claude, letting sellers run workflows and build their own dashboards without ever opening Salesforce's own interface.
Salesforce just put its entire CRM inside Claude that will allow sellers to run CRM workflows without opening Salesforce's own interface.
— Rohan Paul (@rohanpaul_ai) August 27, 2026
Announced a major expansion of their partnership, named Claudeforce.
The plugin arrives with 37 prebuilt sales skills and pilot access now,…
This feels like the natural next step for so many software incumbents in critical infrastructure. What differentiated these winners was specialized knowledge, workflows, or data and they can now double down on that.
Critically, Salesforce has one object model across every customer. Power plants and factories don't operate with the same homogeneity. Tag naming in a historian, SCADA point lists, and asset registries drift from site to site inside the same company.
Solving this problem was more important than a focus on the user experience, and now maybe they never have to. Someone still has to build the semantic layer between the sensor and the agent, but the interface is optional.
The ease of use that agents and chat UIs bring probably expands the market for these businesses. More people inside a utility or a plant can reach the data, which opens the door to consumption pricing instead of a seat count that never grew past the engineers and SMEs who knew the software.
The old question was how many years and dollars it takes to modernize a 15-year-old interface, and whether the team can pull it off.
The new question is whether the data model is clean and whether anything and anyone can reach it.