Bolting AI Agents onto Legacy Workflows Is Failing — Process Redesign From Scratch Is Now the Requirement

8. Bolting AI Agents onto Legacy Workflows Is Failing — Process Redesign From Scratch Is Now the Requirement

MIT Technology Review's April 2026 analysis argues that the dominant enterprise deployment model for AI agents, layering them onto existing fragmented workflows, is structurally insufficient. The piece draws a clear distinction between rules-based automation and adaptive AI agents that interact with data, systems, people, and other agents in real time to execute entire workflows autonomously. The core claim: capturing agent value requires rebuilding processes around agent capabilities, not retrofitting agents into processes built for human or legacy-software execution.

This framing has direct consequences for the competitive positioning of major enterprise software vendors and systems integrators. SAP, Salesforce, and ServiceNow have each made agentic AI a centerpiece of their 2025-2026 product narratives, but their go-to-market motion largely assumes incremental adoption inside existing workflow structures. If agent-first redesign becomes the recognized standard, those vendors face pressure to offer deeper process consulting and architecture services alongside software, a capability gap that benefits firms like Accenture, Deloitte, and IBM Consulting in the near term. Enterprises that attempt the bolt-on approach risk spending significant implementation budget to achieve marginal productivity gains, while early movers who commit to full process redesign could establish durable operational advantages.

The broader signal here connects to a pattern emerging across the agentic AI layer: the technical capability of agents is now outpacing organizational readiness to deploy them effectively. The constraint is no longer model performance but process architecture and change management. This mirrors the early ERP wave of the 1990s, where the software worked but companies that failed to re-engineer around it captured little value. The firms that recognize this cycle early and build agent-native operating models rather than agent-augmented legacy ones are likely to define the next benchmark for enterprise productivity.

Source: https://www.technologyreview.com/2026/04/07/1134966/enabling-agent-first-process-redesign/