A reflective analysis on whether AI tools truly boost productivity or just create a sense of performative busywork in software development.
Microsoft announces MAI-Thinking-1 and MAI-Code-1-Flash LLMs, with low active parameters and claims of performance, but training data issues remain.
Microsoft announces MAI-Thinking-1 and MAI-Code-1-Flash LLMs, with details on parameters, licensing, and training data.
A plain English guide explaining how LLMs like ChatGPT actually work, covering prediction, vectors, embeddings, and common misconceptions.
A developer reflects on using Claude Code, noting less coding but more testing and understanding of AI-generated code.
A critical analysis of AI-generated content duplication on Hacker News, exploring the theory of collective hallucination in tech media.
Explores the concept of AI gateways, their benefits, and how they decouple AI clients from LLM backends for improved management, cost control, and compliance.
Explains mathematically why traditional lakehouses fail AI agents and introduces the agentic lakehouse concept.
Explores using AI tools like Claude and Codex to write high-quality code slowly by finding and fixing bugs in PRs.
A reflection on a 30th college reunion, discussing AI anxiety and nostalgia for a student-built multiplayer Tetris game called BattleTris.
Explores 'vibe coding'—building software by prompting LLMs without reviewing generated code, its benefits, risks, and distinction from agentic programming.
Martin Fowler shares insights from a retreat on software development's future with agentic programming, LLM code porting, and legacy modernization.
Explores using an LLM to interview humans for context gathering, document creation, and review in complex tasks.
Learn how to use TLA+ with LLMs to model and verify system correctness through a classic bean puzzle example.
Explores the deeper purpose of code beyond machine instructions, emphasizing its role as a conceptual model and shared vocabulary in software development.
Analysis of how AI delegation in long workflows corrupts content, with solutions inspired by linters and CI for LLM oversight.
Explores token frugality (Tokensparsamkeit) for coding assistants, offering methods to reduce LLM token usage.
Summary of a fireside chat at Sequoia Ascent 2026 discussing AI agents, Software 3.0, and the shift to agentic engineering.
Structured-Prompt-Driven Development (SPDD) workflow for governable, reviewable, and reusable LLM-assisted code changes.
A quote from Romain Huet confirming OpenAI won't release a separate GPT-5.5-Codex model, as coding capabilities are unified.
