Google shipped two updates to its developer AI tooling in March and April 2026: a tutoring agent for Colab notebooks and authenticated agent-to-agent communication in Gemini CLI.
Learn Mode: Gemini as Coding Tutor
On April 8, Google launched Learn Mode for Google Colab, a feature that transforms the Gemini integration from a code-completion tool into a step-by-step coding tutor. Instead of generating blocks of copy-paste code, Learn Mode provides guided explanations, walks through concepts, and helps users understand the reasoning behind solutions.
The feature works because Colab’s Gemini agent has full notebook context. It can reason about cell execution order, variable state, imported libraries, and outputs. That gives it pedagogical leverage that standalone chat interfaces lack: the tutor knows what the student has already written and what went wrong.
Learn Mode shipped alongside Custom Instructions, which let notebook authors define persistent preferences for how Gemini responds. Custom Instructions are saved at the notebook level, so shared notebooks carry their AI configuration with them. According to Google’s announcement, the combination targets educators designing coursework, students learning to code, and experienced developers working with unfamiliar frameworks.
Gemini CLI: Authenticated Agent-to-Agent Communication
Gemini CLI v0.33.0, released March 11, added HTTP authentication for Agent-to-Agent (A2A) remote agents and authenticated A2A agent card discovery. The feature allows Gemini CLI to delegate tasks to remote subagents over the A2A protocol with proper credential validation, closing a gap where agent-to-agent communication relied on unauthenticated connections.
The A2A protocol, which Google has been developing as an open standard for agent interoperability, enables agents built on different frameworks to discover and communicate with each other. Adding authentication means the protocol can now enforce identity verification before agents exchange instructions or data.
Subsequent Gemini CLI releases have continued building on this foundation. The latest stable release (v0.37.1) includes further A2A server fixes and enhanced subagent security through native macOS Seatbelt and Windows sandboxing.
Two Tracks, One Strategy
These updates reflect Google’s dual approach to AI developer tools. Colab targets the notebook workflow: data science, education, prototyping. Gemini CLI targets the terminal: code generation, multi-agent orchestration, production automation. Google Antigravity, the company’s full agent-first IDE launched in late 2025, occupies a third lane for visual development.
The Learn Mode launch is notable because it pushes agents beyond productivity into education, a use case where most AI coding tools have underinvested. The A2A authentication update is less visible but arguably more consequential: secure agent-to-agent communication is a prerequisite for any multi-agent system operating in production, and Google is building it into its open-source tooling rather than keeping it proprietary.