Product Teams Adopt 'Progressive Autonomy' for AI
Product organizations are increasingly using a 'progressive autonomy' model to build trust in agentic AI, according to recent executive discussions. This development pattern involves AI agents starting with recommendations, graduating to supervised actions, and only operating independently after robust validation, treating the AI like a new hire.
- This approach is often structured as an "autonomy ladder" with distinct stages: starting with the AI observing human decisions, then recommending options, followed by operating within constrained guardrails, and finally being delegated to act agentically on routine decisions. - The push for agentic capabilities is reflected in market forecasts; Gartner, for instance, predicts that by 2028, one-third of enterprise software applications will include agentic AI, allowing 15% of daily work decisions to be made autonomously. - A primary challenge is overcoming the "trust gap" between an AI's technical capabilities and an organization's comfort in ceding control, a concern highlighted by failures like Air Canada's chatbot, which was forced by a court to honor an incorrect discount it offered. - The model requires rethinking team structure, creating new roles for "supermanagers" or "AI orchestrators" whose primary job shifts from performing tasks to managing the AI agents that execute them. - Despite advances, fully autonomous agents are still limited in complex domains; for example, the advanced software engineering agent Devin was able to autonomously solve only about 14% of real-world GitHub issues it was tested against. - In
