For fifty years, competitive advantage in enterprise software lived in the user interface. It included 30 bespoke applications, each with its own screens, and every policy quietly welded into them. According to Ravi Bulusu, Chief Technology Officer of IFS Loops, that arrangement has collapsed. In his recent essay, Where Does Your Moat Go?, Bulusu posed a question every 10- or 20-year-old organization should be asking: Where does durable competitive advantage go when AI agents can execute the work that applications once mediated?
His answer was contrarian and clarifying. Rent intelligence; do not build it. Use the latest, greatest, cheapest public models the day they ship, because there is no advantage in building foundation models yourself. But two assets must stay inside the firewall, because together they constitute the only true moat.
The Two Assets Worth Defending
The first is organizational memory: the observable, correctable collective memory of every judgment employees have ever made, spanning episodic, semantic, and process memory. Bulusu illustrated it simply: a few people in an organization know how to deal with suppliers from Greece, a few others know how to handle suppliers from Texas, and no one person holds it all. If an agent does not know this, it cannot understand the difference. That difference is the moat, and it accumulates value with every deployment. IFS Loops uses vector graphs heavily to capture it, because every new Digital Worker spawned needs to know every decision people made in the past.
The second is process intelligence. Bulusu pointed to Pepsi and Coca-Cola, which differentiate on how efficiently they run their supply chains and little else. Hand that process knowledge to an agent trained on a public model, and an organization has surrendered not just the process but the guardrails that prove whether it was executed correctly. His caution was architectural: use two models, letting public models inform while proprietary knowledge stays internal. As he put it, that knowledge is not Anthropic’s and not OpenAI’s. It belongs to the enterprise.
Execution Numbers Back the Argument
This is not abstract theorizing. ERP Today’s coverage of IFS’s H1 2026 results shows what happens when this architecture meets industrial operations. The company’s 25% year-on-year ARR growth, 24% cloud revenue growth, and the IFS Loops Agentic Platform now running Digital Workers with 60% of agentic transactions fully automated shift the Industrial AI conversation from decision support to work execution. Customer wins in the half included Coca-Cola, China Airlines, Miele, and First Solar.
The moat thesis also explains the vendor’s acquisition logic. As ERP Today observed in its analysis of how IFS is building the operations layer around Industrial AI, the Softeon, 7bridges, and EmpowerMX acquisitions push IFS into the operational layer where physical work, inventory movement, and maintenance outcomes are decided. That is exactly where organizational memory and process intelligence live. However, acquired products expand capability faster than they expand coherence, and customers still need proof that cross-domain workflows run in production.
What This Means for ERP Insiders
Treat organizational memory as infrastructure, not exhaust. Manufacturers deploying agents should capture the episodic, semantic, and process memory behind every human judgment from day one, because each new Digital Worker inherits that accumulated knowledge. Retrofitting memory after deployment is far harder than architecting for it.
Keep process intelligence out of public models. Supply chain efficiency is often a manufacturer’s entire differentiation. ERP teams should demand two-model architectures that rent public intelligence while keeping proprietary processes and their guardrails inside the enterprise.
Judge vendors on execution, not claims. With 60% of IFS agentic transactions now fully automated, the bar has moved from decision support to work execution. Manufacturers should test whether agents act within governed workflows with traceable decisions before committing.
