AGENTS GET REAL: GEMINI CLI ADDS REMOTE SUBAGENTS; SNOWFLAKE LEANS INTO AGENTIC SNOWPARK WITH CORTEX CODE
Gemini CLI now speaks to remote subagents over A2A, while Snowflake’s Cortex Code pushes agentic Snowpark coding into everyday data engineering. A detailed wal...
Gemini CLI now speaks to remote subagents over A2A, while Snowflake’s Cortex Code pushes agentic Snowpark coding into everyday data engineering.
A detailed walkthrough shows the Gemini CLI delegating tasks to remote subagents via the Agent-to-Agent protocol, using Google Apps Script web apps as lightweight A2A servers and a working sample repo to try it yourself (guide, repo). This pattern helps tame Tool Space Interference by offloading tool-heavy actions behind clear network boundaries.
Snowflake’s guide on “agentic” Snowpark work with Cortex Code shows an AI coding agent that generates idiomatic Snowpark Python, pushes computation down, and ships via UI or CLI with a trial guide. It reduces scaffolding for sessions, UDFs, and pipelines so you can focus on logic, not boilerplate.
Meanwhile, a practitioner example demonstrates a platform-agnostic “Skill Forge” agent that searches GitHub before building and emits the right format for each tool, cutting cross-platform translation overhead case study.
Agents are moving from demos to practical building blocks for data platforms, reducing glue code and tool friction.
Clear A2A boundaries and native pushdown codegen can lower operational risk while speeding delivery.
-
terminal
Stand up a minimal GAS A2A subagent and route a Gemini CLI task through it; measure latency, auth flow, and error handling end to end.
-
terminal
Use Cortex Code to generate a Snowpark transformation and a Python UDF; compare runtime, credits, and code clarity versus hand-written baselines.
Legacy codebase integration strategies...
- 01.
Pilot agents on noncritical pipelines with strict scopes and audit logs; terminate network access at a proxy and rotate keys regularly.
- 02.
Wrap subagent calls with circuit breakers and retries, and surface metrics (latency, call count, error class) into your existing observability stack.
Fresh architecture paradigms...
- 01.
Design services behind A2A-style interfaces and keep agents stateless; push compute to Snowflake where possible.
- 02.
Define a simple internal skill schema so you can retarget outputs to different agent platforms without rewriting logic.