For four years, Google sold a clear story to every enterprise CTO who would listen. The future of AI was not about using someone else’s model. It was about training your own. Bring your data. Fine-tune on Vertex AI. Build a moat no competitor could replicate because no competitor had your data.
Thousands of companies believed it. They hired ML engineers. They built custom training pipelines. They stood up Model Registry instances and connected AutoML to their proprietary datasets. Vertex AI was not a feature. It was a strategy. Google’s own sales teams presented it as the centerpiece of cloud AI adoption.
Then, sometime around May 2026, engineers logging into the Google Cloud console noticed that Vertex AI was gone.
Not shut down. Not deprecated with a sunset notice. Renamed. The console now read “Gemini Enterprise Agent Platform.” The features those ML teams had built their careers around were relocated to a sub-menu labeled “Models,” nested inside the agent orchestration interface that replaced them.
Google’s own documentation confirmed it. Custom Training, AutoML, Model Registry, and Pipelines are “still there, just reorganized under a Models sub-menu inside the Agent Platform.”
You only say “still there” about things people assumed you would remove.
The SDK deprecation made the confession concrete. June 24, 2026 was the hard deadline to migrate from google-cloud-aiplatform to the new google-genai package, an agent-first SDK where training is a secondary concern. Not optional. Not recommended. Required.
Four years of enterprise positioning reversed in a console update. The message was unmistakable. Google no longer believes custom model training is the value layer. Agent orchestration is. Training your own model is not the strategy. It is what you did before the strategy arrived.
Every ML engineering team that built on the “bring your own data” pitch now finds their discipline demoted. Not eliminated. Filed under a sub-menu, which is what you do to things you cannot kill yet but have stopped believing in.
The teams that trained the models are now subordinate to the teams that orchestrate the agents. Google did not announce this shift. It reorganized a menu.
👉 Read the agentic security news. Instantly analyze the threat vector, see if it applies to your setup, and find the gaps with our interactive playbook. All free.
👉 Follow me on LinkedIn | X | Substack for weekly analysis of real agent failures, control gaps, and what the frameworks are and are not catching.
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.