The category is converging, and the playbook is changing
The platform is becoming the agent
For a decade, the implicit deal with a data platform was simple: analysts learned the tool, wrote the queries, and the business waited for an answer. That deal is being dismantled. Snowflake CoWork and Databricks Genie point to a single agentic surface where a business question triggers a plan, a retrieval, a reasoned response, and an action. The interface is no longer a dashboard. Instead, it is a conversation embedded inside the tools employees already use, and the platform itself is acting as the coworker. The scale of this transition is no longer theoretical. Gartner forecasts that 40% of enterprise applications will integrate task-specific AI agents by the end of 2026, up from under 5% in 2025. The unit of value is shifting from the answer delivered to the task completed, and that changes how every AI initiative should be scoped.
Context is the hard problem, not the model
The most important word at both summits wasn’t ‘model’. It was context. A frontier model with no understanding of the business is confidently wrong. It does not know when the fiscal year begins, what a churned customer means in a specific market, or which tables are certified. Databricks has answered with the Genie Ontology, a self-improving semantic engine that learns relationships, definitions, and query patterns directly from pipelines. Snowflake has introduced a semantic context layer that frames context engineering as a discipline in its own right. Models are increasingly commoditized. Business meaning is not. The real moat is the layer that gives every agent a shared, trustworthy understanding of the enterprise.
Runtime governance is what keeps agents safe
Even as agents flood enterprise applications, Gartner expects over 40% of in-house agentic AI projects to be canceled by the end of 2027 because of escalating costs, unclear business value, and inadequate risk controls. The gap is operating discipline. Traditional governance asked who can see what data, and that breaks the moment hundreds of autonomous agents start acting on data, not just reading it. Both platforms have responded with a runtime control plane that issues every agent an identity, applies allow-deny-approve policies at execution, and ties every action back to a cost center. Governance has moved from a permissions question to a behavior question, and the audit trail now must reconstruct what an autonomous identity did, not just what a human user saw.
A unified substrate is what makes agents fast
The long-standing wall between transactional and analytical systems is dissolving. Both vendors have made structurally identical bets on managed Postgres, which is quietly becoming the agent’s working memory for embeddings, metadata, and session context. Fewer copies. Fewer hops. One governed place where operational data and the AI reasoning over it can sit together. IDC’s forecast underscores the scale of this build-out: AI spending will reach $1.3 trillion by 2029, dominated by agentic applications and the platforms managing them. The convergence is not a feature war. It’s a budget war, and the substrate is the foundation enterprises will fund.
