The Analyst Is Now an Agent: How Agentic AI Is Swallowing the Data Stack

For twenty years, business intelligence meant dashboards: static, pre-built, queried by the five people in the company who knew SQL. Everyone else filed a ticket and waited three days. That era is over — not because the dashboards got prettier, but because the analyst role itself is being automated into an agent loop.

The shift is happening faster than most enterprise software transitions, and it's being driven by a surprisingly coherent set of forces: large language models that can now write production-quality SQL on the first try, data platforms that have rebuilt their query layers around agent orchestration, and enterprises that are simply exhausted by the ticket-queue model of data access.

Snowflake's Power Move

Snowflake spent the first half of 2026 making its case to be the control plane for the "agentic enterprise," and the pitch is getting harder to dismiss. In April, the company expanded Snowflake Intelligence and Cortex Code — its natural-language-to-action layer — into a unified system that connects data sources, enterprise SaaS tools, and AI models under a single orchestration surface. Cortex Agents now generate SQL directly rather than routing through an intermediate reasoning step, cutting both latency and error rates on analytical queries.

The adoption numbers are striking: more than 50 percent of Snowflake customers are now actively using Cortex Code, just six months after its November 2025 launch. For enterprise software, that's not a slow rollout — that's a land grab.

The real bet isn't that Snowflake replaces your BI tool. It's that Snowflake replaces your analyst team's workflow entirely.

Backing the infrastructure ambition is a $6 billion AWS collaboration commitment announced alongside the Intelligence expansion, positioning Snowflake as the default data layer for agentic workloads running on AWS infrastructure.

Databricks Fires Back

Databricks isn't watching quietly. In a move that's becoming a pattern in this market — someone ships, the rival ships something nearly identical within days — Databricks added SQL-based AI document parsing to its Agent Bricks framework shortly after Snowflake's Intelligence announcement. The capability lets agents ingest unstructured documents, extract structured data, and query it all within the same pipeline.

Both platforms are converging on the same architecture: a natural-language interface sitting atop a governed data layer, with agents that can plan multi-step analytical tasks autonomously. Gartner's 2026 Market Guide for Agentic Analytics formally named this category and defined it as "applying AI agents across the data-to-insight workflow, orchestrating tasks either semiautonomously or autonomously." Databricks and Snowflake are the two most widely deployed platforms in it — but neither has the space to themselves.

The Tellius Problem (and the Opportunity)

Below the Snowflake/Databricks duopoly, a tier of specialized players is carving out serious territory. Tellius, ThoughtSpot, and a dozen others have spent years building NL-to-SQL pipelines tuned specifically for business users rather than data engineers. The pitch is slightly different: these platforms don't want to be your data warehouse, they want to sit on top of it and make every business user a capable analyst.

The problem the entire category is wrestling with: when an AI agent queries your data warehouse autonomously, runs analysis, and emails a summary to a VP — who owns that chain? Who audits it? Gartner flagged that 60 percent of agentic analytics projects relying solely on MCP as their orchestration layer will fail by 2028, citing governance and observability gaps. That's a warning shot aimed squarely at teams that are bolting LLMs onto existing data pipelines without rethinking the access-control and lineage story.

Autonomy without observability is just a new category of technical debt.

This is where the agent security backlash playing out across the broader AI ecosystem has a direct data-stack analog. An agent that can query production databases, write back to data marts, and trigger downstream workflows is not an analyst — it's an actor with real-world consequences. The companies building proper lineage, audit trails, and human-in-the-loop checkpoints into their agentic analytics stacks are going to look very smart in about eighteen months.

The Democratization Trade-Off

The pitch for all of this — Snowflake, Databricks, Tellius, nao Labs, and the rest — rests on a democratization thesis: if any employee can query any dataset in plain English and get an accurate, actionable answer in seconds, the data bottleneck disappears. Gartner projects that 40 percent of enterprise applications will include task-specific AI agents by end of 2026, and conversational analytics is among the most obvious insertion points.

The trade-off is real, though. The more autonomous the analytical agent, the more opaque its reasoning. When a human analyst gets a weird number, they stop and investigate. When an agent gets a weird number and it's close enough to plausible, it reports it. The miscalibration risk is not hypothetical — it's the kind of thing that shows up in board decks and doesn't get caught until someone asks a follow-up question that no agent was prompted to ask.

The smarter vendors are threading this needle by keeping humans downstream of every agentic output, at least for now. The boldest ones are pitching full autonomy. History suggests the cautious approach wins in regulated industries; in high-velocity consumer and e-commerce contexts, the bold approach tends to get adopted first and regulated later.

What Comes Next

The data analytics category is in the middle of a genuine platform transition, not a feature cycle. The winners won't be determined by who has the best NL-to-SQL model — that's largely a solved problem at this point, with Claude Opus 4.8, GPT-4o, and Gemini 2.5 all generating accurate SQL at rates that would have seemed implausible two years ago. The winners will be determined by who builds the best agent orchestration layer: the routing logic, the governance controls, the observability tooling, and the feedback loops that let these systems get smarter from every query they run.

Snowflake has the data. Databricks has the ML engineers. The scrappy middle tier has the product focus that the incumbents consistently underestimate.

The analyst who used to file three-day tickets is either learning to supervise agents or is already looking for a new job. The data stack that used to require a team of six to maintain is running leaner by the quarter. This isn't a slow transition. It's already happened.