Top all-time · #2078
Philipp Schmid
Personal blog · philschmid.deaiprogramming
Philipp Schmid's personal blog at philschmid.de covers ai and programming. Since 2023, 38 different Hacker News readers have linked it 50 times, most often “Agents 2.0: From Shallow Loops to Deep Agents”. It is still writing: the last post was on 2026-09-28, and you can follow it by feed.
Personal Blog of Philipp Schmid Staff Engineer at Google DeepMind.
Most-linked posts?
- Agents 2.0: From Shallow Loops to Deep Agents5 people · 5 mentions · first linked 2025-10-12
- The Rise of Subagents3 people · 4 mentions · first linked 2025-09-15
- MCP CLI3 people · 3 mentions · first linked 2026-01-09
- How to fine-tune open LLMs in 2025 with Hugging Face3 people · 3 mentions · first linked 2024-12-20
- Why (Senior) Engineers Struggle to Build AI Agents2 people · 2 mentions · first linked 2025-11-27
- The new skill in AI is not prompting, it's context engineering2 people · 2 mentions · first linked 2025-06-30
- The importance of Agent Harness in 20262 people · 2 mentions · first linked 2026-01-05
- Google Gemini CLI Cheatsheet2 people · 2 mentions · first linked 2025-07-24
- Extended Guide: Instruction-tune Llama 22 people · 2 mentions · first linked 2023-07-26
- Function Calling Guide: Gemini 2.0 Flash2 people · 2 mentions · first linked 2025-03-05
What Hacker News says?
If you are looking for something that actually explains most concepts behind it, Karpathy’s series will teach you what you want to know. (Mentioned elsewhere) If you are looking for command line tools to fine tune and evaluate models on known datasets, this article is a good take! https://www.philschmid.de/fine-tune-llms-in-2025
Comment linking /fine-tune-llms-in-2025
on “SmolGPT: A minimal PyTorch implementation for training a small LLM from scratch”Submitted The new skill in AI is not prompting, it's context engineering as a story · 915 points
HN discussionI'm working on Conductor https://github.com/skanga/Conductor Conductor is a LLM agnostic framework for building sophisticated AI applications using a subagent architecture. It provides a robust platform for orchestrating multiple specialized AI agents to accomplish complex tasks, with features like LLM-based planning, memory persistence, and dynamic tool use. It provides a robust and flexible platform for orchestrat…
Comment linking /the-rise-of-subagents
on “Ask HN: What are you working on? (October 2025)”Submitted Mini-R1: Reproduce DeepSeek R1 "Aha Moment" as a story · 191 points
HN discussionI read a paper called "The Rise of Subagents" by Phil Schmid at https://www.philschmid.de/the-rise-of-subagents and thought it was an incredibly powerful architectural pattern for running AI agents with complex tasks. So, I decided to build a practical implementation of this system with a central Orchestrator that manages a fleet of implicit or explicit Subagents. Each subagent is a specialized, isolated AI agent de…
Comment linking /the-rise-of-subagents
on “Ask HN: What are you working on? (September 2025)”Submitted How Deepseek R1 Was Trained as a story · 31 points
HN discussionI've only just skimmed this blog post but if I'm reading correctly FunctionGemma can work just like what's intended here, a "contextless" tool router. Going one level up you as a developer have a choice how much context you want to provide to the model. Philipp Schmid wrote a good blog post about this, titling this "context engineering". I like his idea because instead of just blindly throwing stuff into a model's c…
Comment linking /context-engineering
on “FunctionGemma 270M Model”Submitted Fine-tune classifier with ModernBERT in 2025 as a story · 19 points
HN discussionWhat's surprising about this is how sparsely defined the rewards are. Even if the model learns the formatting reward, if it never chances upon a solution, there isn't any feedback/reward to push it to learn to solve the game more often. So what are the chances of randomly guessing a solution? The toy Countdown dataset here has 3 to 4 numbers, which are combined with 4 symbols (+, -, x, ÷). With 3 numbers there are 3…
Comment linking /static/blog/mini-deepseek-r1/tensorboard-r1.png
on “Mini-R1: Reproduce DeepSeek R1 "Aha Moment"”Submitted From PDFs to Insights: Structured Outputs from PDFs with Gemini 2.0 as a story · 16 points
HN discussionGist is 1. Gather training data 2. Format it into JSONL or Hugging Face Dataset format 3. Use Axolotl or Hugging Face peft to fine-tune 4. Export model to GGUF or HF format 5. Serve via Ollama https://adithyask.medium.com/axolotl-is-all-you-need-331d5de... https://www.philschmid.de/fine-tune-llms-in-2025 https://blog.devgenius.io/complete-guide-to-model-fine-tunin...
Comment linking /fine-tune-llms-in-2025
on “I tried vibe coding in BASIC and it didn't go well”Submitted Why (Senior) Engineers Struggle to Build AI Agents as a story · 10 points
HN discussionit can be almost anything like your HN comments or some corporate wiki, then get colab pro 10$ month or some juicy gaming machine and fine-tune that using eg this tutorial https://www.philschmid.de/instruction-tune-llama-2 but https://www.reddit.com/r/LocalLLaMA/ is full of different fine tuned models.
Comment linking /instruction-tune-llama-2
on “Azure ChatGPT: Private and secure ChatGPT for internal enterprise use”Submitted From PDFs to Insights: Structured Outputs from PDFs with Gemini 2.0 as a story · 7 points
HN discussionUse SageMaker: https://www.philschmid.de/sagemaker-llama-llm
Comment linking /sagemaker-llama-llm
on “Azure ChatGPT: Private and secure ChatGPT for internal enterprise use”Submitted Agents 2.0: From Shallow Loops to Deep Agents as a story · 4 points
HN discussionPass^k and not Pass@k (see https://www.philschmid.de/agents-pass-at-k-pass-power-k ). Would be a great twofer to see the code used to run the benchmarks as examples.
Comment linking /agents-pass-at-k-pass-power-k
on “Show HN: New Agentic AI Framework in CNCF”Submitted MCP CLI as a story · 3 points
HN discussionSubmitted The Rise of Subagents as a story · 3 points
HN discussionSubmitted Memory Overview for AI Agents as a story · 3 points
HN discussionSubmitted Zero to One: Learning Agentic Patterns as a story · 3 points
HN discussionSubmitted Function Calling Guide: Gemini 2.0 Flash as a story · 3 points
HN discussionSubmitted LLaMA 2 – Every Resource you need as a story · 3 points
HN discussionSubmitted Foundational Models Became Superhuman in Bash as a story · 2 points
HN discussionSubmitted Agents Manage Other Agents: Four Subagents Patterns in 2026 as a story · 2 points
HN discussionSubmitted Practical Guide on how to build an Agent from scratch with Gemini 3 as a story · 2 points
HN discussionSubmitted Agents 2.0: From Shallow Loops to Deep Agents as a story · 2 points
HN discussionSubmitted Agents 2.0: From Shallow Loops to Deep Agents as a story · 2 points
HN discussionSubmitted AI Agent Benchmark Compendium as a story · 2 points
HN discussionSubmitted Agents 2.0: From Shallow Loops to Deep Agents as a story · 2 points
HN discussion
