BigQuery’s Self-Tuning Engine Slashes Costs for Agentic AI

In the murky waters of modern data analytics, performance tuning has always been a slog. But with AI agents now firing off thousands of queries per minute, the old manual approach is about as useful as a capybara trying to outswim a caiman. That’s why BigQuery is doubling down on autonomous optimization—so you can kick back and let the swamp channel itself.

Google’s BigQuery just announced a slew of improvements that promise up to 35% better query performance and 40% lower processing costs, all without you lifting a finger. The secret sauce? A self-learning engine that remembers what works, an advanced runtime that speeds things up, and a fluid autoscaler that bills you by the second. Let’s dive into the details.

History-Based Optimizations: The Self-Learning Engine

Traditional query optimizers rely on static statistics and metadata, which can be wildly off when data grows and changes fast. BigQuery’s history-based optimizations flip the script: the system learns from past executions and automatically applies optimizations that worked before, while avoiding those that caused regressions.

No user action required—no SQL rewrites, no schema changes, no hints. It’s a closed-loop system with built-in guardrails: if an optimization doesn’t help, it’s immediately rejected and never tried again. One enterprise customer saw P90 execution times drop by 50% and slot usage fall by 15%, with zero regressions.

Advanced Runtime: Faster Execution, Less Slots

BigQuery’s advanced runtime digs deep into the execution layer to find the best physical path for each query. Two key tricks:

• Enhanced vectorization: Leverages modern SIMD instructions to process data in dictionary and run-length encodings, accelerating qualifying queries by up to 10x and cutting slot time by 40%.
• Short query optimizations: For high-concurrency, sub-second queries (think BI dashboards or agentic apps), BigQuery reduces distributed processing overhead, slashing slot usage by up to 10x and achieving P99 latencies under a second.

These improvements work regardless of data format—whether you’re using BigQuery’s native storage or open formats like Apache Iceberg. So you get the same hands-free acceleration even in a lakehouse setup.

Fluid Scaling: Pay Only for What You Use

BigQuery’s compute model is tied to slot-seconds, not node slices. The enhanced autoscaler with fluid scaling enables true per-second billing, lowering costs by up to 34% on average. RISE, an AdTech company processing 1 PB+ daily, cut infrastructure costs by 25% thanks to this feature.

The Agentic Future Is Here

AI agents don’t query like humans—they need orders of magnitude more concurrency and lower latency. Manual tuning is impossible when agents are firing off thousands of queries per second. BigQuery’s autonomous capabilities are built for this, letting you focus on building great agentic apps instead of babysitting infrastructure.

As the data platform evolves from a system of intelligence to a system of action, BigQuery is positioning itself as the engine for the agentic era. And with these improvements, it’s not just faster—it’s cheaper too. For more on how BigQuery stacks up against other platforms, check out our Google Cloud review.

Conclusion: Just Come to Work and It’s Faster

BigQuery’s goal is simple: you walk in each morning, and your queries are faster than the day before—no effort required. With self-learning optimizations, advanced runtime, and fluid scaling, you can stop worrying about data skew, stale stats, or budget blowouts. Just focus on your goals, and let BigQuery handle the heavy lifting.

Ready to give it a spin? Start with a free trial and see the difference for yourself.

Original announcement published on Google Cloud.