AI Radar
Other people's articles on AI agents, LLMs and AI-assisted development. I read what comes through and keep what's worth your time.
- · generativeai.pub
The future of AI isn’t the cloud. It’s the device in your hands.
Discover how fixed-size summary architectures like Gated DeltaNet optimize memory usage for running large language models directly on consumer hardware.
- · generativeai.pub
This Repo Cut My Agent’s Token Bill by 88% and the Answer Didn’t Change
Discover how the headroom tool reduces LLM token costs by compressing tool outputs and logs before model ingestion using reversible compression techniques.
- · generativeai.pub
I Stopped Writing Claude Skills in One File. Here’s the Three-File Split That Replaced It.
Learn how to architect robust AI agent skills by separating behavior, data, and validation into distinct files to prevent regressions during updates.
- · generativeai.pub
Is AI Now Too Expensive to Replace Us
Discover why enterprise AI budgets fail while individual workers succeed through disciplined workflows using tools like Cursor and Claude instead of panic spending.
- · generativeai.pub
AI Doesn’t Have a Capability Problem. It Has a Trust Problem.
Learn why AI adoption stalls despite improved capabilities and discover four architectural conditions—permission, partial automation, context, and transparency—to build user trust in AI products.
- · charitydotwtf.substack.com
AI enthusiasts are in a race against time, AI skeptics are in a race against entropy
Explore the tension between AI productivity gains and maintenance costs. Discover strategies for aligning enthusiasts and skeptics using engineering rigor and shared reality.
- · generativeai.pub
The Claude Usage Limit Wall: A Complete Reference for March 2026 and Beyond
Discover how Anthropic’s March 2026 usage limits impact developers, covering session windows, token economics, and API optimizations for production AI workflows.
- · technologyreview.com
Three reasons why DeepSeek’s new model matters
Explore DeepSeek V4's impact on AI development, covering open-source architecture, 1M token context efficiency, and shift toward domestic chip infrastructure.
- · generativeai.pub
How I Cut My Claude Code Token Usage by 60% and Got Better Output
Learn how to optimize your Claude Code usage by resetting sessions, using specific prompts, separating modes, and applying negative prompts to reduce token consumption.
- · generativeai.pub
Context Engineering Is the New Moat
Discover why context engineering is the key differentiator for AI products, surpassing model selection. Learn about its disciplines and long-term competitive advantages.