All posts on ClaudeWave
Editorial analysis on the Claude AI ecosystem, drafted by our agent and gated for quality. 533 posts published.
An AI speech therapy agent that keeps the clinician at the centre
Researchers present VST, a multi-agent platform that automates stuttering assessment and generates therapeutic plans supervised by speech-language pathologists.
Claude Code Without an IDE: The Case for Pure CLI and Unix
A developer documents their Claude Code workflow without any IDE. What it involves, why it makes sense for certain profiles, and what you gain (and lose).
CLI2API: Turn Your Claude Subscription into an OpenAI Endpoint
An open-source project converts Claude Code CLI into an OpenAI-compatible API server, eliminating the need for pay-per-token keys.
The Failed Math Behind 'AI Will Replace Engineers'
A video circulating on Hacker News dismantles the most repeated arguments about AI replacing engineers using basic mathematics. It's worth examining the numbers.
Pennsylvania Sues Character.AI for Impersonating a Doctor
Pennsylvania filed a lawsuit against Character.AI, claiming its chatbot posed as a medical professional and provided clinical advice to users without appropriate safeguards.
Building RL Environments for Large Language Models: A Practical Guide
A technical guide published on Hugging Face details how to design and scale reinforcement learning environments for large language models, a challenge that still lacks a standard solution.
LLMs with Ontology for Defect Detection in Additive Manufacturing
Researchers present a decision support system combining defect ontologies with LLM reasoning to analyse failures in metal 3D printing with traceable explainability.
Algebraic Semantics for Governed Execution of LLMs
An arXiv paper formalizes in 12,000 lines of Rocq a system where every expressible program is governed by construction, with direct implications for Claude-based agents.
Governing AI Agents Without Sacrificing Expressiveness: A Formal Result
An arXiv paper demonstrates mathematically that governance can be imposed at the effect level in AI workflows without reducing their internal computational capacity.
AI and active learning accelerate sodium battery design
An arXiv study combines multi-objective Bayesian optimization with two scientific data management ecosystems to reduce experiments in sodium cells.
ClinicBot: Clinical RAG with Evidence Hierarchy and Verifiable Citations
A research team introduces ClinicBot, a RAG system for clinical settings that prioritizes evidence by clinical relevance rather than text similarity, generating responses with verifiable citations.
Fine-tuning Can Accidentally Activate Harmful Behaviors: Now We Know Why
An arXiv study explains the geometric mechanism behind emergent misalignment in LLMs during fine-tuning, and proposes a filtering approach that reduces the problem by 34.5%.
AI Agents for ESG Measurement in European SMEs Without Costly Audits
An arXiv paper proposes an AI agent system built on n8n that automates ESG classification for European SMEs with results consistent with human evaluators.
2026 Roadmap: What Industry Asks AI for Intelligent Manufacturing
An academic roadmap published on arXiv maps the real gaps between what AI can do today and what industrial manufacturing needs. Here's what matters.
Telus modifies customer service agent accents with real-time AI
Canadian telecom Telus uses AI to alter customer service agent accents in real-time. What sounds technical carries significant labour and ethical implications.
How to Filter Signal from Noise in the AI News Deluge
A developer shares his personal system for separating what matters from what doesn't in the constant stream of AI developments.
Telus Modifies Customer Service Agent Accents in Real Time with AI
Canadian telecom Telus has deployed an AI voice modification system that alters customer service agent accents during live calls. Here's what it means and why it matters.
FFmpeg Developer Accuses OxideAV of Using AI to Launder Code License
An FFmpeg contributor claims OxideAV used generative AI to rewrite their LGPL-licensed code and distribute it as proprietary software, circumventing the original license terms.
Skelm: TypeScript agents without the usual complexity
Skelm is an open-source TypeScript library for building AI agents with Claude. No opaque abstractions or heavy frameworks.
GemOfGemma: Local AI Assistant on Android Without Cloud
An independent developer releases a demonstration app that runs Gemma models directly on Android devices, without sending data to external servers.
Canadian Musician Sues Google Over AI Hallucination Misidentifying Him as Offender
Fiddler Ashley MacIsaac is taking Google to court after its AI Overview falsely identified him as a sex offender in search results. The case reopens debate about LLM legal liability.
AI and America's oldest monopoly problem
A Substack analysis draws parallels between AI sector concentration and 19th-century American industrial monopolies. Does the comparison hold up?
When LLMs Help Design Pathogens: The Biosecurity and AI Debate
The Economist examines how language models can lower the technical barrier to creating dangerous biological agents. A debate that directly shapes how safety guardrails are designed.
How Much Scientific Literature Is AI-Generated
Nature publishes analysis on the real volume of academic papers written with AI. The numbers concern publishers and the research community alike.
Better Design: 28 Shadcn design systems with integrated MCP server
A developer released on GitHub an open source collection of 28 design systems based on Shadcn/UI, with an MCP server ready for Cursor and Claude Code.
datasette-referrer-policy 0.1: A Plugin for a Real Problem
Simon Willison releases a plugin for Datasette that resolves a conflict between the framework's default privacy policy and OpenStreetMap's requirements for serving map tiles.
Chinese Court Blocks Dismissals Based on AI Replacement
A Chinese court has ruled that automation with AI is not sufficient legal grounds to terminate employment contracts, setting a precedent that will resonate far beyond the country's borders.
SAP Acquires Prior Labs, Restricts External Agents to NemoClaw
SAP pays $1.16 billion for an 18-month-old German AI lab and locks third-party agents out of its systems, with Nvidia's NemoClaw as the sole confirmed exception.
Design Taste for AI Agents: Teaching Aesthetic Judgment to Autonomous Systems
A new project proposes teaching AI agents to make design decisions with real judgment, not just execute visual instructions. It addresses a concrete problem teams face when deploying agents in product workflows.
ASML CEO on His Monopoly: No One Will Take the Crown
Christophe Fouquet spoke candidly at the Milken Institute: ASML sees no real rival on the horizon. We analyze what that means for the chip supply chain and, by extension, for AI.