GenAI PMPhilipp Schmid
AI engineer and educator known for sharing practical model and agent-building insights. Here he predicts that 2026 will be the year of Agent Harnesses.
Key Highlights
- Philipp Schmid is an AI engineer and educator known for turning model and agent advances into practical guidance for builders.
- His commentary strongly emphasizes agent harnesses, evals, and evolving agent systems over static API thinking.
- He has highlighted important product signals across Gemma 4, Gemini API pricing, AI Studio controls, and on-device AI demos.
- For AI PMs, his work is especially useful for planning agent UX, evaluation strategy, deployment architecture, and cost guardrails.
Philipp Schmid
Overview
Philipp Schmid is an AI engineer, educator, and highly visible technical communicator who frequently translates fast-moving model and agent advances into practical guidance. Across newsletter mentions, he appears as someone who helps practitioners understand how to build, evaluate, and deploy modern AI systems—from on-device open models like Gemma 4 to production patterns for agentic products. He is also associated with the view that 2026 will be the year of Agent Harnesses, emphasizing the importance of the scaffolding, evaluation loops, and operational systems around agents rather than just the base model itself.For AI Product Managers, Schmid matters because his work sits at the intersection of model capability, developer tooling, and product execution. His commentary consistently points toward what makes AI products actually usable in practice: shifting from static APIs to evolving agents, using evals instead of relying only on unit tests, designing for multimodal and on-device use cases, and thinking in terms of outcome-based systems. That makes him a useful signal source for PMs deciding where to invest across agent UX, infrastructure, model choice, and deployment patterns.
Key Developments
- 2026-03-26 — Philipp Schmid shared the release of Lyria 3 Pro and Clip in Google AI Studio and the Gemini API, highlighting prompt-based music generation pricing and product availability.
- 2026-03-27 — He was among the voices covering Gemini 3.1 Flash Live, an audio model focused on more natural conversations and stronger function calling.
- 2026-03-28 — Schmid noted upcoming Gemini API billing changes, including monthly spending caps, automatic pauses when limits are hit, faster tier upgrades, and project-level spend controls in AI Studio.
- 2026-03-29 — He explained how Kimi Moonshot, Cursor, and Chroma train vertical agentic models with reinforcement learning using a strong base model, a production harness, and outcome-based rewards—an important framing for the rise of specialized agent systems.
- 2026-04-03 — He was listed among prominent figures covering the launch of Gemma 4, Google DeepMind’s Apache 2.0–licensed open model family for reasoning and agentic workflows.
- 2026-04-06 — Schmid released a visual guide to Gemma 4, explaining its multimodal image/audio/text design, sparse MoE inference, and device-friendly architecture choices such as a 2B-parameter embedding trick for phone deployment.
- 2026-04-07 — He demoed Gemma 4 E2B running on an iPhone 17 Pro Max via the Google AI Edge Gallery, emphasizing skills-driven Wikipedia querying and the practical potential of on-device AI experiences.
- 2026-04-10 — Schmid shared five principles from his talk on why senior engineers struggle with AI agents: treat text as state, hand over control, view errors as inputs, shift from unit tests to evals, and design evolving agents instead of static APIs.
- 2026-04-10 — This agents-focused guidance reinforced his broader thesis that the next wave of value will come from agent systems and harnesses, not just better standalone models.
Relevance to AI PMs
1. He offers practical frameworks for shipping agents, not just discussing them. His guidance on treating text as state, designing evolving agents, and replacing some unit-test thinking with evals is directly applicable to PMs defining agent product requirements, success metrics, and release criteria.2. He surfaces infrastructure and operating-model implications early. Mentions around Gemini billing tiers, AI Studio spend controls, and production harnesses help PMs plan budgets, guardrails, experimentation velocity, and commercialization strategies before costs spiral.
3. He connects model capability to product form factors. His work on Gemma 4 and AI Edge Gallery demos is especially relevant for PMs evaluating when to use cloud-hosted models versus on-device models, and how multimodal or edge deployment can unlock new UX and distribution opportunities.
Related
- ai-agents / agent / agent-harnesses — Core themes in Schmid’s commentary; he emphasizes that successful AI products depend on the surrounding harness, not only the model.
- evals / llm-as-judge / context-bench — Closely tied to his recommendation to move from traditional unit-test thinking toward evaluation-driven agent development.
- Gemma 4 / gemma / functiongemma — Important model family in his recent coverage, especially for open, multimodal, and deployable agent workflows.
- Google / Google DeepMind / Google Research — Many of the launches he discussed are connected to Google’s model, API, and research ecosystem.
- Gemini API / Google AI Studio / AI Studio — Relevant platforms in his updates on pricing, deployment, and multimodal product capabilities.
- Google AI Edge Gallery — Connected to his on-device Gemma demo, showing how edge AI can become a product surface rather than just a research curiosity.
- Cursor / Chroma / Kimi Moonshot — Referenced in his explanation of how vertical agentic systems are trained with RL, harnesses, and outcome-based rewards.
- Sebastian Raschka / Simon Willison / Jeff Dean / Demis Hassabis — Adjacent voices and influential figures frequently appearing around the same model and agent discussions.
Newsletter Mentions (34)
“#14 𝕏 Philipp Schmid shared five essential principles from his talk on why senior engineers struggle with AI agents: treating text as state, handing over control, viewing errors as inputs, shifting from unit tests to evals, and designing evolving agents instead of static APIs.”
#14 𝕏 Philipp Schmid shared five essential principles from his talk on why senior engineers struggle with AI agents: treating text as state, handing over control, viewing errors as inputs, shifting from unit tests to evals, and designing evolving agents instead of static APIs.
“Philipp Schmid shared five essential principles from his talk on why senior engineers struggle with AI agents: treating text as state, handing over control, viewing errors as inputs, shifting from unit tests to evals, and designing evolving agents instead of static APIs.”
#14 𝕏 Philipp Schmid shared five essential principles from his talk on why senior engineers struggle with AI agents: treating text as state, handing over control, viewing errors as inputs, shifting from unit tests to evals, and designing evolving agents instead of static APIs.
“Philipp Schmid shared five essential principles from his talk on why senior engineers struggle with AI agents: treating text as state, handing over control, viewing errors as inputs, shifting from unit tests to evals, and designing evolving agents instead of static APIs.”
Philipp Schmid shared five essential principles from his talk on why senior engineers struggle with AI agents: treating text as state, handing over control, viewing errors as inputs, shifting from unit tests to evals, and designing evolving agents instead of static APIs. #15 𝕏 Andrew Ng unveiled a new short course, “Efficient Inference with SGLang: Text and Image Generation,” co-built with LMSys and RadixArk and taught by Richard Chen, teaching how to use SGLang’s open-source caching framework to slash redundant LLM costs by processing shared promp...
“#3 𝕏 Philipp Schmid demos Gemma 4 E2B on an iPhone 17 Pro Max via the AI Edge Gallery, spotlighting its skills-driven Wikipedia querying feature and sharing the app link.”
#3 𝕏 Philipp Schmid demos Gemma 4 E2B on an iPhone 17 Pro Max via the AI Edge Gallery, spotlighting its skills-driven Wikipedia querying feature and sharing the app link.
“Philipp Schmid released a visual guide to Gemma 4 illustrating its multimodal image/audio/text processing, sparse MoE inference that runs only a fraction of parameters, and a clever 2B-parameter embedding trick to fit on phones.”
#7 𝕏 Philipp Schmid released a visual guide to Gemma 4 illustrating its multimodal image/audio/text processing, sparse MoE inference that runs only a fraction of parameters, and a clever 2B-parameter embedding trick to fit on phones.
“Also covered by: @Sebastian Raschka , @Simon Willison , @Philipp Schmid , @Jeff Dean , @Google DeepMind , @Demis Hassabis , @Demis Hassabis , @Sebastian Raschka”
Google DeepMind Releases Gemma 4 Open Models #1 𝕏 Google DeepMind launched Gemma 4, a family of Apache 2.0–licensed open models you can run on your own hardware for advanced reasoning and agentic workflows. Also covered by: @Sebastian Raschka , @Simon Willison , @Philipp Schmid , @Jeff Dean , @Google DeepMind , @Demis Hassabis , @Demis Hassabis , @Sebastian Raschka #2 𝕏 Qwen unveiled Qwen3.6-Plus, a next-gen multimodal agentic model with smarter, faster coding execution, sharper vision reasoning and a 1M-token context window by default via API, all while maintaining top-tier general performance.
“#2 𝕏 Philipp Schmid shows how @Kimi_Moonshot, @cursor_ai, and @trychroma all train vertical agentic models via RL using a strong base model, production harness, and outcome-based rewards.”
Today's top 10 insights for PM Builders from X and Blogs. #2 𝕏 Philipp Schmid shows how @Kimi_Moonshot, @cursor_ai, and @trychroma all train vertical agentic models via RL using a strong base model, production harness, and outcome-based rewards. K2.
“#6 𝕏 Philipp Schmid : Starting April 1, the Gemini API billing tiers get monthly spending caps that pause the API once reached (resuming next month or upon upgrade), with faster automated tier upgrades.”
#6 𝕏 Philipp Schmid : Starting April 1, the Gemini API billing tiers get monthly spending caps that pause the API once reached (resuming next month or upon upgrade), with faster automated tier upgrades. You can also set per-project spend limits directly in AI Studio.
“Also covered by: @Demis Hassabis , @Philipp Schmid , @Google AI , @Google AI , @Sundar Pichai , @Sundar Pichai”
#1 𝕏 Google DeepMind launched Gemini 3.1 Flash Live, an audio model that delivers more natural conversations with improved function calling for more useful, informed interactions. Also covered by: @Demis Hassabis , @Philipp Schmid , @Google AI , @Google AI , @Sundar Pichai , @Sundar Pichai
“#15 📝 Philipp Schmid released Lyria 3 Pro and Clip in Google AI Studio and the Gemini API, offering prompt-driven full-song generation (minutes at $0.08 each) and 30-second clips ($0.04 each).”
#15 📝 Philipp Schmid released Lyria 3 Pro and Clip in Google AI Studio and the Gemini API, offering prompt-driven full-song generation (minutes at $0.08 each) and 30-second clips ($0.04 each). Also covered by: @Google DeepMind , @Demis Hassabis , @Demis Hassabis , @Logan Kilpatrick #16 📝 Surge AI Blog Riemann-bench: A Benchmark for Moonshot Mathematics - Riemann-bench is a verifiable benchmark of extreme-tier mathematical problems designed to test frontier models; current top models score under 10% on these challenges.
Related
An AI coding assistant/editor that can use dynamic context across models and MCP servers to reduce token usage. Useful for AI PMs thinking about agentic workflows, context management, and efficiency.
Developer and writer known for hands-on AI and tooling tutorials. Here he provides a Docker-based walkthrough for running OpenClaw locally.
A developer platform company behind Sandbox at Vercel. Relevant to AI PMs because it is positioning infrastructure for agentic workflows and automation.
Google DeepMind is presenting the Interactions API beta, positioned as a unified interface for Gemini models and agents. For AI PMs, it signals continued investment in agent infrastructure and product surfaces for 2026.
An AI researcher mentioned for sharing transformer residual connection improvements. Relevant to AI PMs because model architecture advances affect capability and training stability.
Technology company behind Gemini and related AI initiatives. Mentioned here through Jeff Dean's comments on personalized learning.
Google's AI model family referenced as a tool for personalized education. Useful to AI PMs as an example of applied model use in learning products.
Google’s AI development studio for building and monitoring Gemini-based apps and workflows. In this newsletter it’s highlighted for dashboard improvements that make usage and performance easier to inspect.
A protocol for connecting tools to AI agents; the newsletter contrasts bulky MCP setups with lighter skill-based integrations.
Google’s research organization, cited for a method to help small models match large-model performance on intent extraction. Relevant to PMs interested in cost-efficient model architectures and mobile understanding.
Google leader and AI researcher cited for discussing personalized learning with AI models. Relevant to education product use cases and model applications.
CEO and cofounder associated with Google DeepMind and AI research. Here he is referenced teasing a robotics collaboration involving Gemini Robotics.
A company building developer tools and frameworks for LLM applications. It is implied by the mention of Harrison Chase and a product progress bar/pricing discussion.
Autonomous or semi-autonomous systems used here in sales and coding workflows. The newsletter highlights their role in replacing human SDR tasks and orchestrating complex tasks.
A Gemini model variant used here to power agentic workflow examples and multi-agent systems. It is relevant to AI PMs as an example of frontier model capability enabling more complex automated workflows.
A beta API associated with Gemini that supports multimodal inputs including images, PDFs, CSVs, and custom data via Deep Research. Useful for AI product teams building multimodal workflows.
A streamlined, high-speed multimodal model optimized for low-latency text and vision tasks. AI PMs would care about its performance-cost tradeoffs, on-device suitability, and throughput gains.
A W3C-backed browser extension that exposes website functionality to MCP-capable agents. It lets developers register site functions as structured tools in the browser.
Google’s video generation model with updates to portrait mode, visual consistency, and higher-resolution upscaling.
Google’s search product used as a grounding source in AI Studio. The newsletter notes hosted grounding tools for building citation-backed apps.
Google AI Edge Gallery is a Google tool for showcasing and running on-device AI experiences at the edge, including offline use cases.
An agent-focused tool or environment that supports agent skills. The newsletter highlights compatibility with Gemini CLI, Claude Code, and OpenCode.
An open-source command-line tool for dynamic discovery of Model Context Protocol servers. It is described as reducing MCP token usage and improving AI agent tool interactions.
Stay updated on Philipp Schmid
Get curated AI PM insights delivered daily — covering this and 1,000+ other sources.
Subscribe Free