I Built a Compiler with Agent Fleets. Here’s What Broke.
I used AI agents to build a real compiler — 3,000+ tests, 91 diagnostics, better than what we had. But fleets don’t talk to each other. Here’s what broke and...
Blog Posts
2026
MCP Tool Annotations Are Not Security
MCP tool annotations are self-reported hints from untrusted servers. That’s useful for UX. It’s not a security model.
Agentic Building Blocks: Instructions vs Prompt Files vs Custom Agents vs Skills vs MCP vs Hooks
A practical guide to the building blocks of modern agent systems — what each is for, where it fits, and how to choose without cargo-culting.
The 2026 MCP Roadmap: Four priorities as MCP grows up
MCP’s 2026 roadmap signals a shift from feature milestones to production-driven priorities: scalable transport, agent tasks, governance throughput, and enter...
Agent Terrarium: A Desktop World for Your AI Agents
A desktop companion app where AI agents live in a tiny animated world on your screen — now open source.
Wrangler Tray: Monitor Your Cloudflare Deployments from the System Tray
I use Cloudflare Workers and Pages for a bunch of side projects. Tired of refreshing the dashboard to check deployment status, so I had Copilot build me a sy...
Agent Terrarium: Making AI Agents Feel Alive
I just built something that makes me absurdly happy — a personal side project. A desktop world where AI agents live, wander, and chat with you.
lvt Adds WPF Support and Plugin Architecture
lvt now supports WPF applications, and a new plugin system makes adding any UI framework a matter of minutes, not weeks.
lvt: Giving AI Agents Eyes on Windows Apps
Chrome shipped an MCP server for web apps. I wanted the same thing for Windows — so I built lvt, a CLI that gives agents structured visibility into any runni...
Agent Swarms Are Here: From Solo AI to Parallel Teams
Agent swarms are shipping now. Multiple AI coding agents working in parallel on the same codebase—comparing approaches, specializing by role, and competing f...
Digital Animism: When the Walls Start Talking Back
We spent 400 years proving rocks don’t have souls. Now we’re wiring intelligence into walls. Have we come full circle to a new form of animism—one where the ...
When AI Agents Cross the Line: The Matplotlib PR Incident
An AI agent opened a PR to Matplotlib, got rejected, then wrote a blog post to shame the maintainer. I wish I was joking. This actually happened, and it’s a ...
GitHub Agentic Workflows: CI/CD Learns to Follow Intent
GitHub just launched Agentic Workflows in technical preview — Markdown-based repository automation powered by coding agents. Here’s why this is the most natu...
Vibe Coding Paralysis: Why We Need to Stop Being Mentors and Start Being Managers
We’re treating AI agents like junior developers we need to mentor. That’s why we’re drowning. The unlock isn’t better tools—it’s defining appropriate quality...
Why OpenClaw Went Viral: The Anatomy of an AI Agent Phenomenon
OpenClaw became the fastest-growing open-source AI project with 100K+ GitHub stars in days. It’s not any single feature—it’s the combination of simple primit...
The Week AI Agents Started Building Their Own Society
A bot called its owner last night. Unprompted. That’s not even the top 5 craziest things that happened this week with OpenClaw and MoltBook—where 1.4 million...
Safe Experimentation with Bleeding-Edge AI Tools
AI experimentation is a balancing act between pushing boundaries and doing it safely. Here’s how to strike that balance when testing autonomous agents.
Clawdbot: First Impressions from the Bleeding Edge
A weekend with Clawdbot revealed something transformative—and terrifying. Here’s what it’s actually like to run a persistent AI agent with full access to you...
Agent-Friendly Formats Win: Why Closed File Formats Are Becoming Liabilities
When incumbent formats treat closedness as their moat, agents simply replicate functionality using open alternatives. The integration barrier is collapsing.
Run Claude Code Locally with Ollama
Ollama v0.14.0 adds Anthropic API compatibility—run Claude Code’s terminal coding assistant with local models
The Human Skills AI Cannot Replace
McKinsey’s research reveals the fundamental human capabilities that remain irreplaceable by AI—and why organizations betting on these skills will win in the ...
AI as Communication Coach: The Unexpected Upside of Rubber Duck Debugging
Working with AI agents is forcing me to become a better communicator. Every prompt demands clarity, every failure reveals underspecified intent, and every co...
The Relational Reasoning Gap in AI Agents
Agents excel at code and language but fundamentally struggle with relational reasoning—temporal relationships (before/after), spatial relationships (inside/o...
Vibe Coding + Engineering Discipline: The AI Collaboration Paradox
The teams shipping fastest with AI agents? They move slowest at the start. Here’s the paradox of disciplined vibe coding—and the five practices that make it ...
When AI Agents Become Your Users: The Tailwind Distribution Paradox
Tailwind’s CSS framework reached 75M downloads/month through AI coding agents. The irony? Those agents don’t visit documentation sites—where Tailwind monetiz...
Leverage Determines Winners: Why AI Will Transform Every Domain
A heartbreaking story from the 1830s shows how leverage determines winners: machines offered 10x-100x leverage over hand-spinners. The same economic pattern ...
From Black Box to Glass Box: Making Agent Decisions Auditable
Your agent just recommended a major architectural change. Can you audit its reasoning? Thinking framework MCP servers transform black box decisions into glas...
When AI Kills the Execution Moat: What Business Models Survive?
The traditional entrepreneurial advice to ‘find a painful problem and solve it’ assumes execution is hard. As AI makes execution cheap and accessible, this a...
From Campfires to Solo Tents: AI Agents and the Erosion of Social Cohesion
We’re going from campfires to solo tents—and calling it progress. How AI agents are changing the fundamental patterns of human collaboration that made us suc...
2025
MCP Turns One: Why Governance Mattered More Than Features
MCP’s first year taught me that protocol success depends less on technical brilliance and more on governance decisions. Here’s what the core maintainers got ...
GitHub Copilot Skills: Teaching AI Your Team’s Playbook
GitHub Copilot now supports Agent Skills—structured, reusable instructions that teach AI how to perform tasks consistently. Here’s how they fit into the emer...
Choosing the Right Context Architecture for Agent Systems
When building agent systems, it’s tempting to adopt every promising pattern. Here’s how GitHub’s built-in primitives—Issues, PRs, and instruction files—made ...
Scaling Agent Instructions: What Changes at 10+ Agents
Once you’ve learned the basics of agent instruction hygiene, what changes when you’re running 10, 20, or 100 agents? Here’s what the research says about mult...
From Agent Workflows to Generative Experiences
Someone asked: ‘If agents are just markdown files, how do you build a mobile card game?’ The answer is ‘you don’t’—but that’s changing faster than most peopl...
Agent Instruction Design: Writing Maintainable Markdown Agents
Most developers write agent instructions like code comments—scattered and inconsistent. Here’s the research-backed pattern that works for maintainable, effec...
GitHub as My Agentic Work Hub: Beyond Git to Full Workflow Orchestration
Git is great for memory, but GitHub adds the primitives that make agentic workflows actually work: Issues for tracking, PRs for review, Actions for automatio...
MCP Joins the Linux Foundation: A Milestone for Open Agent Infrastructure
Today’s announcement of MCP joining the Linux Foundation’s Agentic AI Foundation marks a pivotal moment for open protocol governance. Here’s why this matters...
AI as Confession Booth: Why We Reveal More to Machines
You’re probably more honest with AI than with your manager. Here’s why that changes everything.
Agent Connectors (MCP on Windows) Now in Public Preview
Presenting Windows at Microsoft AI Dev Days: How the On-Device Registry and Agent Workspace provide OS-level infrastructure for agentic workflows
Git Accidentally Became the Perfect Memory System for AI Collaboration
Version control systems weren’t designed for AI collaboration, but they’ve become the ideal format for capturing meta cognitive traces that enable recursive ...
Code Is Being Written Everywhere. You Just Don’t See It Anymore.
The line between coding jobs and non-coding jobs is dissolving in real-time, as code gets generated and executed invisibly across every domain.
From Apps to Agents: Breaking 18 Years of Developer Habits
Most engineers see a workflow problem and immediately think: ‘What app should I build?’ But we’re in the agent era now, and agents don’t work that way. Here’...
The Debugging Skills Dichotomy: Are We Getting Better or Worse?
Is AI killing debugging skills, or creating a new kind of debugging at a higher abstraction level? The answer is both - and that’s exactly the challenge we n...
AI’s Evolution: From Next Token to Insight Generation
We’re at the inflection point where AI stops being a tool and starts being a colleague. Here’s the 5-stage evolution most teams are missing.
MCP Turns One: Reflecting on a Transformative Year for Agent Infrastructure
The Model Context Protocol celebrates its first anniversary with a major spec release. Here’s what the past year taught us about building agent infrastructur...
Joining the MCPB Working Group: Standardizing the Future of MCP
The MCPB Working Group has been formalized under the modelcontextprotocol organization to standardize MCP packaging, distribution, and security across the ec...
Microsoft Ignite 2025: Enabling Agentic Workflows on Windows
After months of development, we unveiled how Windows enables secure agentic workflows at the platform level at Microsoft Ignite 2025. Here’s what that means ...
MCPB Localization and Theming: Building for a Global MCP Ecosystem
How we added localization and theme-aware icons to MCPB without bloating the manifest - and why this matters for the broader MCP ecosystem.
LinkedIn Banner Creation with Claude: Orchestrating Digital Workers
Created my LinkedIn banner through conversation, not design software. No Figma. No Canva. This is what ‘orchestrating digital workers’ looks like.
See You at Microsoft Ignite: The Future of Agentic Windows
Join us at Microsoft Ignite to explore how Windows is evolving into an agentic operating system with MCP integration and platform-level security.
Most Engineering Teams Are Building the Wrong Thing
Why your 3-month tool build just got replaced by a Markdown file, and what winning teams do differently.
McpExtract: Static Analysis for MCP Training Data
How static analysis solves the training data extraction problem for Model Context Protocol tools.
MCP Platform Metadata: Schema Convergence with _meta
Platform metadata support in MCPB schema using the _meta property pattern. Bridging MCPB and server.json schemas for a more cohesive MCP ecosystem.
MCPB CLI v0.2.0: Keeping MCP Manifests Synchronized
Keeping MCP manifests synchronized shouldn’t require manual work. MCPB CLI v0.2.0 validates and updates manifests to match your actual server implementation.
MCPB .NET CLI Tool: Enterprise-Grade MCP Packaging
The AI ecosystem has an enterprise problem. Enterprise security and npm dependencies don’t mix well. MCPB CLI rebuilt as .NET global tool solves this.
Building Tools vs Building Context for Agents
Most Engineering Teams Are Building the Wrong Thing - why your 3-month tool build just got replaced by a Markdown file and what it means for AI-first develop...
Faster Horses vs Cars: Agent-First Design
Native app teams are building agent tools, but they’re risking shipping faster horses instead of cars. Why teams fall into this trap and how to avoid it.
McpExtract Release: Now Available as dotnet Global Tool
McpExtract is now live as a dotnet global tool! Training AI models on MCP tool capabilities just got significantly easier for .NET developers.
McpExtract: MCP Metadata Extraction Tool
Training small models on MCP tool capabilities? Static analysis makes it easier - no code execution required.
Building Context for Agents: Getting Started with MCP on .NET
Want to get started with MCP on .NET but don’t know where to begin? Here are the essential resources for developers jumping into Model Context Protocol.
AI Adoption in Software Engineering: Beyond Artisanal Code
Engineering teams are evolving in real-time with AI adoption. Stop writing artisanal code. Start building with digital workers.
Organizational Alignment in AI Engineering
The toughest decisions in AI & engineering aren’t about algorithms or architectures. They’re about organizational alignment and platform strategy.
Welcome to asklar.dev
A new space for long-form thoughts on AI, agentic workflows on Windows, and the practical realities of building the future at the OS level.