Insights for AI builders
Tutorials, product updates, and ideas to help you build and ship AI applications faster.
Subscribe via RSS
What Is Behavioral Lock-In? How Persistent AI Agents Create Switching Costs That Data Portability Can't Fix
Persistent AI agents like Conway accumulate behavioral context that can't be exported. Here's why this creates a new kind of lock-in and what to do about it.
The Claude Code Git Worktree Pattern: A Primer for Builders
Git worktrees explained for Claude Code users: what they are, how the pattern works, and when to reach for it on refactors and concurrent feature branches.
Claude Code Split-and-Merge: A Deep Dive Into Sub-Agent Parallelism
How Claude Code's split-and-merge pattern uses isolated context windows across 10 sub-agents to handle codebases too large for a single agent to process.
What Is Claude Mythos? Anthropic's Most Dangerous AI Model Explained
Claude Mythos is Anthropic's unreleased frontier model that found thousands of zero-day vulnerabilities. Learn what it can do and why it won't be released.
What Is the Conway Agent? Anthropic's Unreleased Always-On Background AI Revealed in the Code Leak
Conway is Anthropic's leaked always-on background agent with push notifications, GitHub subscriptions, and a proprietary extension format. Here's what we know.
What Is the Conway Extension Format? How Anthropic's Proprietary Layer Sits on Top of MCP
Conway's .cnw.zip extension format creates a proprietary ecosystem on top of the open MCP standard—similar to how Google Play Services sits on top of Android.
What Is the Gemma 4 Apache 2.0 License? Why It Changes Everything for Commercial AI Deployment
Gemma 4 ships under a true Apache 2.0 license—no custom restrictions, no compete clauses. Here's why that matters more than the model's benchmark scores.
What Is Gemma 4? Google's First Apache 2.0 Multimodal Model With Audio, Vision, and Function Calling
Gemma 4 is Google's open-weight model family with Apache 2.0 licensing, native audio and vision, built-in function calling, and 128K–256K context windows.
How Stripe Ships 1,300 AI PRs a Week: Harness Engineering
Stripe ships 1,300 AI-generated PRs every week, powered by harness engineering — the discipline beyond prompt and context for multi-agent orchestration.
What Is the OpenAI Unified AI Super App? How ChatGPT, Codex, and Browsing Are Merging
OpenAI is building a unified AI super app that combines ChatGPT, Codex, and browsing into one agent-first experience. Here's what that means for users.
What Is Perplexity Computer for Taxes? How AI Handles Federal Tax Returns and IRS Forms
Perplexity Computer's tax module can draft federal income tax returns on official IRS forms, review professionally prepared returns, and build tax dashboards.
Project Glasswing: Anthropic's Internal Pen-Test With Mythos
Project Glasswing turns Claude Mythos loose on Anthropic's own infrastructure to find zero-days before any release. A look at AI-driven internal red teaming.
What Is Qwen 3.6 Plus? Alibaba's 1M Token Agentic Coding Model With Real-World Agent Design
Qwen 3.6 Plus is Alibaba's frontier-level model built for real-world agents with a 1M token context window, multimodal vision, and strong coding benchmarks.
What Is Recraft V4? The AI Image Model Built for Professional Brand Assets
Recraft V4 introduces two model tiers—photorealistic and native SVG vector—designed for brand visuals, logos, packaging, and agency-quality design work.
What Is Recraft V4 Vector? How to Generate Native SVG Logos and Icons With AI
Recraft V4 Vector generates true scalable SVG files—not rasterized fakes—for logos, icons, and illustrations. Here's how it works and when to use it.
5 Claude Code Workflow Patterns Explained: From Sequential to Fully Autonomous
Claude Code supports five agentic patterns: sequential flow, operator, split-and-merge, agent teams, and headless. Learn when to use each for maximum output.
Claude Mythos vs Claude Opus 4.6: How Big Is the Cybersecurity Capability Gap?
Claude Mythos scores 83.1% on cybersecurity benchmarks vs Opus 4.6's 66.6%. Here's what the gap means for AI agents, security teams, and builders.
What Is the Compounding Knowledge Loop in Claude Code? How Your Agent Gets Smarter Over Time
Claude Code's session hooks capture learnings automatically, building a wiki that improves agent answers over time. Here's how the compounding loop works.
What Is the Gemma 4 Mixture of Experts Architecture? How 26B Parameters Run Like 4B
Gemma 4's MoE model activates only 3.8B of 26B parameters at a time using 128 tiny experts. Learn how this delivers 27B-class intelligence at 4B compute cost.
What Is the Gemma 4 Vision Agent? How to Combine a VLM With Image Segmentation
Combining Gemma 4 with Falcon Perception creates an agentic pipeline that counts objects, segments images, and reasons across modalities. Here's how it works.
Is Google Flow Free? Pricing, Credits, and Tiers Explained
Google Flow's free tier is a trial: a few credits and Veo 2 only. Pro is $19.99 and Ultra $249.99. Here's what each tier really gets you.
Intelligence Arbitrage vs Labor Arbitrage: How AI Is Rewriting the Economics of Knowledge Work
AI shifts value from person-hours to outcomes. Learn how intelligence arbitrage replaces labor arbitrage and what it means for your career and business model.
What Is Andrej Karpathy's LLM Knowledge Base? The Compiler Analogy for AI Memory
Karpathy's LLM knowledge base treats raw articles as source code and a wiki as the compiled executable. Learn the architecture and how to build your own.
What Is Gemma 4? Google's First Apache 2.0 Multimodal Reasoning Model
Gemma 4 ships under an Apache 2.0 license with native audio, vision, function calling, and reasoning. Here's what makes it a breakthrough for open-weight AI.