Carbon Chooses, Silicon Scales
AI may become better than humans at many forms of thinking without replacing humanity. The real question is whether the intelligence extending us will remain accessible to everyone.
Posts in
AI
23 posts
AI Feels Faster. The Data Is Less Certain.
AI coding tools feel transformative, but the productivity evidence is mixed. Part 2 of the LogCTL series on AI-native engineering examines the gap between perceived speed and measured outcomes, and why review is becoming the new bottleneck.
AI-Native Software Engineering Isn’t About AI
AI-native software engineering is not defined by how much AI a company uses. It describes a development model in which AI is treated as a primary participant in software construction rather than an occasional assistant.
The Future of Software Engineering Is an Operating Model
AI-native engineering is not defined by removing humans from software development. Part 5 of the LogCTL series examines how roles, teams, economics and accountability change when implementation is no longer the dominant activity.
The Model Is Only One Part of the System
A useful AI-native workflow requires more than a better prompt. Part 3 of the LogCTL series maps the practical method: specifications, context engineering, agent workflows and the harness that enforces permissions, boundaries and stopping conditions.
When Agents Scale, Coordination Becomes the Product
One agent can misunderstand a requirement. Several agents can misunderstand it in different ways. Part 4 of the LogCTL series on AI-native software engineering maps coordination, multi-agent orchestration, layered verification and governance at scale.
The Model Is Not the Development Environment
I spent 12 million input tokens on a small monorepo change. The problem was not the model. It was the development environment around it.
AI Speeds Up Coding, Not Engineering
An opinionated workflow for building good-enough software with AI coding agents without surrendering architecture, judgment, or security.
I Tried Too Many Coding Agents. Here Is What I Would Actually Pay For
A subjective comparison of Claude Code, Codex, Cursor, OpenCode, and the models inside them, based on ordinary application work rather than benchmarks.
I Couldn't Become a 10x Developer
The pursuit of becoming a 10x developer through context-switching across six projects, multiple languages, and AI agents — and the quiet realization that 1.5x is honest work.
Your AI Agent Should Be a Folder
After testing agent frameworks for personal automation, I realized the best personal AI agent is not a platform—it is a well-organized directory of Markdown, scripts, logs, and state.
We Abused POST for Years. HTTP QUERY Is the Apology.
Complex read-only queries have been hiding inside POST /search endpoints for years. RFC 10008 adds an HTTP QUERY method — safe, idempotent, and built for body-heavy reads. Here is how it changes API design.
Why Don't We Build Data Centers in Antarctica?
Cooling is expensive, Antarctica is cold, so why not? Because data centers are factories, not hot rooms — and factories need power, fiber, roads, spare parts, workers, permits, and boring infrastructure that Antarctica refuses to provide.
Why Don't We Build Data Centers in Space?
Free solar power, no land disputes, no cooling towers — space sounds like the perfect home for the cloud. But space takes away almost everything a data center actually needs: easy cooling, maintenance, upgrades, networking, and repair.
I Automated an Entire Course Production Pipeline. Then I Decided Not to Use It.
I built an AI pipeline that turned storyboards into JSON, generated assets, narration, slides, presenter video, and captions. It almost worked, but the experiment taught me why course production is authored, not merely manufactured.
AI Doesn't Need to Become a Better Doctor. It Needs to Become a Better Historian.
When AI spends 30 minutes patiently building a complete medical history and a doctor gets 10, the comparison isn't about intelligence — it's about information. Here's why we should be building AI historians, not AI doctors.
Stop Cloning the Same Repo Again and Again. Git Worktrees Exist.
Stop cloning the same repository for every context switch. Git worktrees let you work on multiple branches simultaneously without WIP commits, broken dependencies, or mental clutter.
Why I Created an Ultra-Simple MySQL Backup Tool
Backups shouldn't need dashboards, agents, or cloud tokens. Just a boring script, cron, and a compressed dump copied to another VM. Here's the philosophy behind mysql-backup.
Human-in-the-Loop Is a Lie We Tell Ourselves Before Clicking "Approve All"
Human-in-the-loop governance sounds responsible, but tired humans blindly approving every prompt isn't governance. The real future is programmatic boundaries for machines and accountability for humans.
Learning in an AI-Driven World is a Paradox
AI has fundamentally changed the feedback loop between learning and building. I build faster than ever, yet some days I feel like I know less than ever. Here is what that paradox actually means for developers.
I Tried Building My AI Agent With Existing Agent Frameworks. Then I Chose the Boring Deterministic Way.
After evaluating multiple AI agent frameworks, I found that deterministic automation with an LLM inside beats magic autonomy for real operational work.
Not Everyone Needs Frontier LLMs for Coding
Frontier models are brilliant, but many everyday coding tasks run just fine on smaller models — if your workflow is clear. Stop buying the hype and choose tools for your actual use case.
Why Use LangGraph for Your First Agentic AI Project
LangGraph forces you to learn state management, tools, routing, and human-in-the-loop approval — the fundamentals that survive framework churn in the fast-moving agentic AI space.