From human SQL consoles to agent-driven data platforms
An AI-native database is a data platform designed from the ground up for machine reasoning, agent workflows, and inexact, natural-language-driven queries rather than direct human SQL sessions and strictly deterministic lookups. This is not a cosmetic upgrade to existing stacks; it is a reversal of who the primary user of the database is. At Google’s recent cloud event, the message was blunt: “we're putting agents at the center ... with the goal that humans are not going to be using data platforms in the next three to five years. It’s going to be humans orchestrating agents, and agents actually doing the work.” In that world, the old comfort of hand-crafted SQL gives way to intent-driven engineering, conversational analytics, and a constant negotiation with probabilistic systems that aim for the best answer, not the only answer.
Google’s inexact vision: AI-native infrastructure and agent-first access
Google’s database leadership is explicit about the core shift: when retrieving data “it’s not so much about getting the exact results, but getting the best results.” That means inexact query patterns become first-class citizens. Natural language questions, informed by prior interactions and rich context, drive systems that blend structured and unstructured information through vector indexing, text search, and graph storage—what they call AI-native infrastructure. In this model, a fuzzy prompt might compile down to an exact SQL query, but the interface expectation has flipped. Even BigQuery and distributed systems like Spanner are being infused with AI functions such as AI.IF, which use Gemini models to evaluate natural-language conditions inside SQL. The fact that Spanner now peaks at 7.5 billion queries per second, up from about 5 billion a year earlier, and stores around 23 exabytes of data, shows the scale at which this experiment is unfolding.
Ghost and the rise of agentic database architecture
If Google is reimagining how you ask questions, Tiger Data is reimagining where those questions land. Ghost is pitched as “a database service designed and built specifically for AI agents,” addressing what its creators call one of the most pressing infrastructure problems in AI: databases need to be built for collaborating with agents, not retrofitted from human-centric tools. The opinion baked into Ghost is clear: agentic database architecture should assume constant experimentation, failure, and isolation by default. Ghost provides unlimited Postgres databases with fast forking via CLI or MCP, from short-lived sandboxes to dedicated always-on instances. Its per-query pricing model makes it cheap enough to treat individual databases as disposable, opening the door to one-database-per-task or per-hypothesis patterns that would have been cost-prohibitive before. In other words, Ghost optimizes for blast-radius control and trial-and-error, not for the tidy shared environments DBAs grew up with.
New skills for developers in an inexact, agentic world
The biggest mistake developers can make is to treat agent-driven data platforms as a fancy SQL autocomplete. When models orchestrate multi-step, parallel workflows, the discipline shifts from writing queries to describing intent and constraints. Staff “need to be skilled not so much in prompt engineering, but rather using AI for spec-driven development,” and the work becomes iterating with models on the right plan for a problem rather than micro-optimizing every join. On top of that, teams must learn to design eval sets: curated question–answer pairs that continuously test whether AI-generated SQL and inexact responses remain acceptable. Tools like Ghost push another habit change: instead of huddling around a handful of shared schemas, engineers spin up hundreds of isolated databases per agent or task, enabled by free tiers with 100 compute hours a month and 1TB of storage. The craft moves toward managing agent-aware data layers, not singular monoliths.
Enterprise consequences: fewer dashboards, more conversations—and more risk
For end users, the most visible change will be the slow death of dashboards. Static charts “serve the first layer of predictable questions”; they are being replaced by conversational analytics that let business users ask anything in natural language. With better context from systems like Google’s Knowledge Catalog, some customers now see over 90 percent accuracy in those conversations, where 18 months ago models answered only about half of queries correctly. That sounds impressive until you are the customer whose transaction is blocked by the unlucky 10 percent. Enterprises will have to balance the lure of intent-driven analytics against accountability and error budgets. Meanwhile, cloud giants have a strong incentive to push AI into every interaction because AI workloads increase both data storage and token usage. Ghost’s immediate availability, with hard spending caps on paid tiers, signals that smaller players see a narrow window to define the agentic database category before hyperscalers fully colonize it.
