Spot on. The 'AI as a feature' phase is definitely hitting a ceiling. I’ve been following the same shift in the QA world recently. Most legacy tools are just adding a GPT-wrapper to 'generate a test case,' which is just a feature.
But the real winners are those moving toward AI-native orchestration—where the system doesn't just write a script but autonomously manages the execution layer and heals itself. I was just reading a deep dive on how ai in software testing https://testomat.io/blog/ai-in-software-testing/ is moving toward this 'agentic' workflow where the product IS the automation engine, not just a UI with a chatbot.
It's exactly as Archit said above: the mental model shift from 'AI as a feature' to 'AI as architecture' is the real unlock for ROI.