Philipp Schmid

基础模型通过Bash脚本实现超人类级任务处理，正在颠覆传统工具使用方式，工程师需关注脚本自动化新趋势。

Google Gemini API整合Google Maps和Search工具，实现单次调用完成搜索、地图定位和自定义功能联动，大幅简化位置应用开发。

EvoCode-Bench是首个模拟真实开发循环的多轮编码基准，通过持久工作区、演变规范和累积测试提升代理评估的实用性。

EvoCode是首个测试代理在多轮交互中持续满足演变需求的基准，强调避免破坏已有功能。

本文展示如何使用Gemini Live Translate、Next.js、LiveKit和Cloud Run构建实时翻译应用，涵盖音频流处理、翻译优化和部署方案。

Gemini Interactions API 的入门指南被全面重写，提供从首次 API 调用到运行自主代理的 11 步教程。

Agents' Last Exam (ALE) 是一个评估代理在 55 个行业中 1000 多个真实专业任务上的表现的基准测试。

Google Colab 现在支持 CLI 和 Skills，允许用户通过终端直接访问 Colab 运行时，包括 GPU/TPU 配置和自动训练模型。

The Gemini API provides a sandboxed Linux environment through a single API call, supporting code execution, web access, and file I/O. The article offers a complete example to build a data science assistant.

Philipp Schmid published an article about the Gemini Managed Agents Dev Guide, introducing a feature that can be achieved through a single API call, including Gemini 3.5 Flash, Antigravity Harness, and a remote Linux sandbox, without any infrastructure or orchestration.

Philipp Schmid展示了如何使用单个curl命令调用Gemini API来构建一个GitHub问题分类代理，该代理能够克隆仓库、抓取开放问题、分类问题类型并执行复现代码以确认bug，整个过程无需复杂的编排框架或基础设施。

The Gemini Interactions API update introduces multi-domain agent steps, replacing traditional user and model roles.

Senior engineers struggle with AI Agent development because the paradigm shifted from deterministic programming to iterative prompt-feedback loops; text is now the new state representation, and engineers must transition from 'traffic controllers' to 'dispatchers'.

The Gemini API's File Search now supports multimodal retrieval. Use `gemini-embedding-2` to build a unified RAG system for PDFs and images with a single call. Storage and query-time embeddings are free; you only pay for indexing and inference.

文章系统梳理2026年AI智能体协同演进的四大子智能体模式：同步/异步工具调用、独立await派生、持久化工作池、多智能体消息协作团队。

Gemma 4 12B achieves native multimodal processing for text, images, and audio by removing separate vision and audio encoders. This architecture replaces traditional encoder-patching approaches with joint representation learning, reducing inference latency and improving edge deployment efficiency.

GEPA is a general framework for automatically optimizing prompts of CLI tools, supporting custom CLI, local models, and API agents.

Google launched Managed Agents in the Gemini API, enabling developers to spin up AI agents that reason, write and run code with a single API call, all within a hosted Linux sandbox, simplifying AI agent development.

Gemini Managed Agents combined with the Interactions API enables AI agents to have secure Linux sandbox environments for executing code and managing memory autonomously.

Gemini API launches Managed Agents feature, enabling creation of intelligent agents with isolated Linux sandbox through single API call.

Philipp Schmid 在 Android 开发者大会上介绍了 Gemini，它使 Android 成为一个代理系统，允许应用程序通过新的 @AppFunction 注解公开功能，实现跨应用操作。

Philipp Schmid分享了正确使用MCP服务器的两种模式：显式使用和子代理，以避免资源浪费。

Google DeepMind released Science Skills, an open-source toolkit providing standardized agent interfaces for genomics, structural biology, and literature search to accelerate scientific AI workflows.

Gemma 4 12B is the first mid-sized multimodal model with native audio input, featuring a unified encoder-free architecture that runs on 16GB VRAM, matches 26B benchmark performance, and uses Apache 2.0 license.

Philipp Schmid shares third-party evaluations of Gemini 3.5 Flash, showing strong performance in agents, coding, vision, and finance domains.

Philipp Schmid宣布为Gemma 4模型推出多令牌预测（Multi-Token Prediction）drafters技术，实测推理速度提升最高达3倍，且输出质量零损失。

本文介绍如何使用 Gemini Live API、LiveKit 和 Google Cloud Run 构建实时翻译应用，适合前端工程师参考。

Philipp Schmid利用GoogleDeepMind的Gemini 3.1 Pro观看cursor_ai SDK的发布视频并生成制作脚本，随后使用Remotion无提示重现视频，展示其视频理解能力。

Google's Gemini API launches Managed Agents, enabling a complete agent with code execution, web browsing, and file management through a single API call, powered by Gemini 3.5 Flash and Antigravity infrastructure.

Gemini 3.8 Flash在多模态任务中表现卓越，尤其在图像推理和数据提取方面领先，速度提升30%。