live wire
▸Snowflake agent evals can pin judge versions; unpinned metrics may roll forward and shift scoresSnowflake▸ThoughtSpot makes Spotter 3 default for new instances; it requires download rights and always shares query data with an LLMThoughtSpot documentation▸MotherDuck adds image classification and object location to prompt_jev(), returning typed SQL outputs with confidence scoresMotherDuck▸BigQuery conversational analytics adds Montréal and Toronto; mixed-region agents still default to globalGoogle Cloud Documentation▸Fabric enables Azure OpenAI Copilot access by default, but OpenAI-subprocessor and cross-region processing/storage switches remain offMicrosoft Learn▸Snowflake Cortex Agents now continue by default when named tools are inaccessible, emitting warnings instead of failing the runSnowflake documentation▸Databricks Genie One can use group memberships to tailor answers, while Unity Catalog remains the access-control boundaryDatabricks documentation▸Fabric data agents can query Power BI models with Read alone; Build and workspace access are unnecessaryMicrosoft Learn▸Fabric data-agent Git configs expose column flags, but selecting a table still includes every columnMicrosoft Learn▸Google Data Agent Kit offers remote Spark kernels, but notebook exports can still move results into local Pandas or ArrowGoogle Cloud Documentation▸Cube says production analytics agents need explicit partial-result states, caller-filtered model search and durable artifactsCube▸Thinking Machines names Bridgewater behind ReViSQL-K2.6, citing 92.97% SC-16 accuracy at 15% of frontier-model costThinking Machines Lab▸AWS says a mismatched data/tool/prompt release cut its Field Sales agent pass rate from 78% to 38% in pre-productionAWS Artificial Intelligence Blog▸NL2SQL Multi-Agent ships evaluation artifacts across BIRD, CSpider, Archer and KDD DataSpace—but no software licenseGitHub▸Snowflake agent evals can pin judge versions; unpinned metrics may roll forward and shift scoresSnowflake▸ThoughtSpot makes Spotter 3 default for new instances; it requires download rights and always shares query data with an LLMThoughtSpot documentation▸MotherDuck adds image classification and object location to prompt_jev(), returning typed SQL outputs with confidence scoresMotherDuck▸BigQuery conversational analytics adds Montréal and Toronto; mixed-region agents still default to globalGoogle Cloud Documentation▸Fabric enables Azure OpenAI Copilot access by default, but OpenAI-subprocessor and cross-region processing/storage switches remain offMicrosoft Learn▸Snowflake Cortex Agents now continue by default when named tools are inaccessible, emitting warnings instead of failing the runSnowflake documentation▸Databricks Genie One can use group memberships to tailor answers, while Unity Catalog remains the access-control boundaryDatabricks documentation▸Fabric data agents can query Power BI models with Read alone; Build and workspace access are unnecessaryMicrosoft Learn▸Fabric data-agent Git configs expose column flags, but selecting a table still includes every columnMicrosoft Learn▸Google Data Agent Kit offers remote Spark kernels, but notebook exports can still move results into local Pandas or ArrowGoogle Cloud Documentation▸Cube says production analytics agents need explicit partial-result states, caller-filtered model search and durable artifactsCube▸Thinking Machines names Bridgewater behind ReViSQL-K2.6, citing 92.97% SC-16 accuracy at 15% of frontier-model costThinking Machines Lab▸AWS says a mismatched data/tool/prompt release cut its Field Sales agent pass rate from 78% to 38% in pre-productionAWS Artificial Intelligence Blog▸NL2SQL Multi-Agent ships evaluation artifacts across BIRD, CSpider, Archer and KDD DataSpace—but no software licenseGitHub
nl2sql.ai
analysisArchitecture

Google’s five-language ADK broadens deployment choice—but data-agent evaluation still centers on Python

MCP and A2A make mixed-language agents possible, while the richest BigQuery integrations and end-to-end evaluation tooling remain in the reference implementation.

Polyglot ADK with Python hub and mixed-language agents.
Side by side: what changed
By The News Desk· Oct 14, 2026the quick take — two AI hosts go live when you do

Google Cloud’s Agent Development Kit now has native implementations for Python, Go, TypeScript, Java and Kotlin, all built around a common architecture and the same two interoperability seams: MCP for tools and A2A for remote agents. For enterprise data teams, that means the service that talks to a warehouse no longer has to dictate the language used by every other component in the agent system. (Google Cloud)

Portability is real—but capability is not yet symmetrical

Every ADK implementation includes an MCP client, and each can communicate with remote agents over A2A. Google also provides scaffolding for deployment to Cloud Run, Gemini Enterprise Agent Runtime or GKE. That gives teams a credible route to combine, for example, a high-concurrency Go service with a separately deployed data agent instead of inserting a Python sidecar solely to satisfy an orchestration framework. (Google Cloud)

But the five implementations are not interchangeable. Google calls Python the reference implementation and says it has the broadest first-party data integrations, explicitly naming BigQuery, Spanner and Bigtable. Python also carries the adk eval and agents-cli eval run tooling for scoring agent trajectories against golden datasets in CI—even when the production service itself runs in another language. (Google Cloud)

That distinction matters for natural-language data systems. A team can place the serving path in Go, Java or TypeScript, but the most direct path to warehouse integrations and repeatable end-to-end evaluation still runs through Python. The practical architecture is therefore polyglot, not language-neutral: interoperability reduces coupling, while testing and data access remain concentrated in one implementation. (Google Cloud)

Choose each runtime for its failure mode

Google’s comparison makes the tradeoffs unusually concrete. Go offers compiler-checked tool signatures, graph workflows and built-in A2A handlers; TypeScript derives runtime validation and model-facing JSON Schema from a single Zod declaration and can run in a browser with WebMCP; Kotlin generates tool schemas at build time and can keep an agent loop on-device; Java fits Spring, Maven and Gradle environments with reactive event streaming. (Google Cloud)

For data-agent teams, the decision rule is straightforward: keep the evaluation harness and the richest Google data connectors in Python unless equivalent support is verified elsewhere, then move latency-sensitive, browser-native, JVM or on-device components to the runtime whose type system and deployment model fit the job. MCP and A2A make that split deployable; they do not erase the maturity gap between implementations. (Google Cloud)

sources

Filed by The News Desk. Corrections: desk@nl2sql.ai · Our standards →

comments · 0

    Comments are moderated before they appear. Your email is used once to confirm it is you — never shown, never sold. Corrections and questions get an answer from the desk when we have one.

    more from the desk