The Predictor Already Moved
Two years before I arrived, something with my project's name was already waiting in the same city. The countdown ends on March 27, 2026.
Essay
Writing & Notes
Thoughts on AI agents, modern engineering, and building spike.land.
Why Spike Chat Stays Sharp While Transcript-First AI Chats Go Soft
Most AI chats solve problems by throwing more transcript at the model. spike-chat does the opposite: split prompts, scored memory, staged execution, compressed working memory, and bounded tool calling.
Developer Experience
We Migrate Next.js Apps Without Guesswork
Most migrations fail because teams try to rewrite the app before they understand the architecture. We extract the structure first, automate the mechanical work, and turn the process into a repeatable MCP capability on spike.land.
Services
Appendix: Anticipated Pushback on the Next.js Migration Post
Preemptive responses to the most common objections about leaving Next.js for TanStack Start, from build config to lock-in to the solo-dev framing.
Developer Experience
Next.js vs TanStack Start: Why Next.js Became Tech Debt for Me
An honest account from a solo open-source developer: expensive build minutes, slow deployments, flaky pipelines, and why Next.js stopped feeling like leverage.
Developer Experience
Where Does Your Code Actually Belong? Take the Quiz.
A reorganization script sorted 990 TypeScript files while I couldn't sleep. Can you beat the algorithm? Paste your own code and find out.
Engineering
How spike.land Uses AI and A/B Testing to Find Bugs Before You Do
Variant testing on spike.land is not just for conversion. It is a product-quality loop for catching unstable app behavior before it becomes a support issue.
Engineering
Embed spike.land MCP Tools in Your Existing Project in 5 Minutes
You do not have to rebuild on spike.land to use spike.land. The MCP worker is designed to be called from other origins, Cloudflare Workers, and existing apps.
Developer Experience
Introducing the spike.land App Store: Vibe Code, Publish, Earn
The spike.land app store turns MCP tools into publishable apps you can discover, install, embed, and eventually monetize.
Product
The Universal Interface Wasn't GraphQL
In 2017 I told two brilliant engineers that DynamoDB + GraphQL = done. They weren't impressed. Nine years later, I finally understand why — and what the actual universal interface is.
Architecture
Why Your Claude Agent Is Wasting 70% of Its Context Window on Tool Descriptions
Adding more MCP tools to your Claude agent can make it worse, not better. Here's the math behind the context window problem, and the lazy toolset loading pattern that fixes it.
Developer Experience
A Chemist Walked Into a Codebase
A chemist shipping apps is interesting. The deeper story is spike.land's core bet: formalize business logic as typed MCP tools, enforce it with tests, and turn that into running software.
Developer Experience
Why We Gave Bugs an ELO Rating
How we built a public bug tracker with chess-inspired ELO rankings, user reputation gating, and per-service feedback tools — turning bug reports into a defense against platform abuse.
Architecture
Getting Started with spike.land: From Signup to Your First AI Tool Call
A step-by-step guide to setting up spike.land, exploring the MCP tool store, configuring API keys, and making your first tool call — in under 15 minutes.
Tutorial
Introducing spike.land — The MCP-First AI Platform
spike.land launches with 80+ MCP tools, an AI tool marketplace, API proxy credits, and a free tier. Here's what we built and why it matters for developers building with AI.
Product
defineBlock(): How I Built a Full-Stack Database Abstraction on Cloudflare Workers
I tried SpacetimeDB, wrote a blog post about it, then deleted it two days later. But the syntax stuck. Here's how I rebuilt those patterns on Cloudflare Workers and migrated my entire platform in a weekend.
Developer Experience
The Architecture of Scale: How I Made My MCP Tools Agent-Writable
When Google released Nano Banana 2, my image tools became obsolete overnight. Here's how I rebuilt my MCP tool builder so agents could compose, test, and deploy tools without touching framework code.
Developer Experience
MCP Explained: The Universal Adapter for AI Tools
MCP is more than tool calling — it's a presentation layer for how AI understands tools, data, and workflows. Here's the full spec, what spike-cli implements today, and what's coming.
Developer Experience
The Grandmother Neuron Fallacy: Why AI Tool Chains Break
Why deterministic tools become brittle inside LLM tool chains, and how to design the model boundary more clearly.
Developer Experience
One Site, Many Faces: How We Built 16 Versions of spike.land
The same URL, 16 different experiences. How spike.land uses persona-based personalization to serve content that actually resonates.
Engineering
Think Slowly, Ship Fast
The Hourglass Model: heavy specs at the top, disposable UI in the middle, bulletproof business logic at the bottom. Five explanations, one for every level of your team.
Developer Experience
Godspeed Development: 100 App Ideas Powered by Spike Land MCPs
A look at the incredible velocity of AI-assisted development. Here are 100 full-stack app ideas you can build instantly using Spike Land's MCP tools.
Productivity
Docker Layers Are Just Like LLM Context Caching
LLM workflow design follows the same optimization law as Docker builds: stable prefixes are cheap, early changes are expensive, and monolithic pipelines waste compute.
Developer Experience
It Feels Like Cheating (Because It Is)
How I deleted my E2E tests, rebuilt my API in two days, and stopped pretending code is precious. The BAZDMEG Method's radical conclusion: code is disposable.
Developer Experience
The Testing Pyramid Is Upside Down
The worst part of your test suite — the slow, flaky E2E tests — can run at unit test speed. MCPs remove the browser from the equation.
Developer Experience
The Vibe Coding Paradox: Why Your AI Gets Dumber the More You Let It Wing It
We built an AI that generates React apps from URLs. It worked 40% of the time. Then we taught it to learn from its own mistakes — using the same physics that makes your prompts fail.
Developer Experience
Hogyan tervezi a Claude Code a kontextust: Interjú az Opus 4.6-tal
Egy mély technikai interjú arról, hogyan állítja össze a Claude Code a kontextust tervek készítésekor — és mit tanulhatnak a fejlesztők a token-generálás, a figyelem és a KV-gyorsítótár megközelítéséből.
Developer Experience
How Claude Code Engineers Context: An Interview with Opus 4.6
A deep technical interview about how Claude Code assembles context when creating plans — and what developers can learn from its approach to token generation, attention, and the KV cache.
Developer Experience
Context Engineering Your Zero-Shot Prompt
Stop iterating endlessly. Learn how to front-load context so AI models get your code right on the first try. A beginner-friendly guide.
Developer Experience
You Cannot Automate Chaos: The Complete Guide to AI-Powered Dev Pipelines
A practical guide to combining AI agents and workflows for shipping production code. Covers the automation-ready checklist, the test pyramid for agents, a concrete workflow from issue to merged PR, and why your CI pipeline matters more than your choice of AI model.
Developer Experience
How to Automate Your Dev Team: AI Agents That Ship Production Code
A beginner-friendly guide to replacing manual bottlenecks with autonomous AI workflows. Learn how to use Claude Code, Jules, and CI/CD to ship features without writing code.
Developer Experience
What I Learned From My Worst Pull Request
A developer's honest story about misusing AI tools on a new project, and the framework that fixed everything.
Developer Experience