Major RPA vendors have announced AI-powered, agentic, intelligent automation capabilities. These announcements reflect market pressure more than architectural possibility. Understanding why requires examining what's built into existing RPA platforms.
The Deterministic Runtime Problem
RPA platforms are built on deterministic runtimes. Each step executes exactly as specified. Selectors match exactly or fail. Coordinates are absolute. Timing is fixed.
This determinism was a feature, not a bug. Enterprises wanted predictable automation. They got it.
But deterministic runtimes can't be made probabilistic without fundamental rebuilding. Adding "AI-powered element detection" to a deterministic runtime creates hybrid chaos — sometimes probabilistic, sometimes deterministic, never predictably either.
The runtime would need to be rebuilt from scratch. That's years of work and breaks backward compatibility with existing bots.
Central Orchestrator Architecture
RPA platforms centralize orchestration. Bots are dispatched from central servers. Execution is monitored centrally. Failures are reported centrally.
This architecture assumes reliable connectivity and centralized visibility. Neither assumption holds in edge deployments, air-gapped environments, or privacy-constrained contexts.
Agentic systems must operate autonomously — making local decisions when connectivity fails, handling exceptions without central guidance, maintaining state through disconnections.
Retrofitting autonomy onto central orchestration is architectural contradiction.
Selector Fragility as Design Choice
RPA platforms identify UI elements through selectors — XPath, CSS, attribute matching. Selectors are brittle by design: they encode structural assumptions that break when structure changes.
The alternative — visual identification with semantic understanding — requires completely different infrastructure. Different capture mechanisms. Different matching algorithms. Different confidence handling.
RPA vendors can't simply "add AI to selectors." They need to replace the entire element identification approach. That changes the programming model, the debugging experience, and the maintenance workflow.
The Business Model Lock-In
RPA vendors make money from bot licenses, orchestration servers, and professional services to maintain brittle bots. This business model depends on the current architecture.
Self-healing, autonomous agents reduce professional services revenue. Local-first architecture reduces server license revenue. Resilient automation reduces bot license expansion.
Vendors have financial incentives to not solve the problems agentic architecture addresses. The technology shift requires business model shift that threatens current revenue streams.
This isn't conspiracy — it's standard innovator's dilemma. Incumbents struggle to disrupt themselves.
Key Takeaway
RPA vendors face architectural and business model constraints that prevent easy evolution to agentic systems. The deterministic runtime, central orchestration, and selector-based identification would all need replacement. That's not evolution — it's starting over.