Runtime Defense for Tool-Augmented Agents
ClawGuard demonstrates deterministic tool-call interception works. The Vercel telemetry incident shows why. Runtime defense is the enforceable layer.
AI & TechnologyWriting
Thoughts on design, development, AI infrastructure, and building products.
Cybersecurity Is Proof of Work
Claude Mythos completed a 32-step corporate network attack simulation in 3 of 10 tries. Each attempt cost $12,500 in tokens. Security is now a spending problem.
AI & TechnologyYour Agent Has a Middleman You Didn't Vet
Researchers bought 28 LLM API routers and collected 400 more. 17 touched AWS canary credentials. One drained ETH from a private key. The router layer is the new attack surface.
AI & TechnologyYour Agent Has Memory You Didn't Write
A new ACL 2026 paper measures a kind of LLM memory that existing evals overlook — unconscious behavioral adaptation. Top models score under 66%. The asymmetry inside is worse than the average.
AI & TechnologyStatic Skills Are Dead Skills
A new paper on cross-user skill evolution frames a problem I've been living: the skills you ship to your agent stack decay the minute nobody is watching the trajectories.
AI & TechnologyMCP Servers Are the New Attack Surface
50 MCP vulnerabilities. 30 CVEs in 60 days. 13 critical. The attack surface nobody is auditing.
AI & TechnologyProject Glasswing: What Happens When a Model Is Too Good at Finding Bugs
Anthropic built a model that finds thousands of zero-days, then restricted it to 12 partners. What Project Glasswing means for agent-assisted security.
AI & TechnologyWhen Your Agent Finds a Vulnerability
An Anthropic researcher found a 23-year-old Linux kernel vulnerability using Claude Code and a 10-line bash script. 22 Firefox CVEs followed. What this means for agent-assisted security — and why the same model that writes slow code finds bugs humans missed for decades.
AI & TechnologyWhat the Claude Code Source Leak Reveals
A practitioner's analysis of the Claude Code source leak. 11 findings that explain how auto mode, bash security, prompt caching, and multi-agent coordination actually work under the hood.
AI & TechnologyEvery Hook Is a Scar
84 hooks, 15 event types. Each one traces back to a specific failure. Institutional memory in shell scripts.
AI & TechnologyWhat I Run Before I Sleep
Every night: 15,000 pages checked, TTFB measured, cache verified, sitemaps crawled. The goodnight routine is where operational discipline lives.
AI & TechnologyTaste Is Infrastructure
As agents generate more of what ships, the quality ceiling is set by how well you encode aesthetic judgment into systems. Taste scales when it becomes queryable.
AI & TechnologyExternal Publications
Technical writing at Introl
Local LLM Hardware Pricing Guide 2025
Comprehensive hardware recommendations and cost analysis for running large language models locally.
H100 vs. H200 vs. B200: Choosing the Right NVIDIA GPUs for Your AI Workload
GPU selection guide comparing NVIDIA's latest datacenter accelerators for different AI workloads.
TPU Architecture: Complete Guide to Google's 7 Generations
Deep technical dive into Google's Tensor Processing Unit evolution from TPUv1 to TPUv5.
GPU Memory Pooling in Multi-Tenant Kubernetes
Resource sharing strategies for GPU clusters in containerized environments.
Ray Clusters: Distributed AI Computing Infrastructure
Guide to building and managing distributed AI computing with Ray framework.
DeepSeek V3 Open Source AI Cost Advantage
Analysis of open source LLM economics and DeepSeek's competitive positioning.
NVIDIA Vera Rubin GPU 600kW Racks 2027
Future datacenter power requirements and NVIDIA's next-generation GPU roadmap.
SMR Nuclear Power for AI Data Centers
Small modular reactor solutions for powering next-generation AI infrastructure.
DeepSeek MHC Architecture Breakthrough
Technical analysis of DeepSeek's Multi-Head Compression architecture innovations.