Skip to content

ls -t blog/

Blog

  1. Drumming with a Whoop: A Rhythm Game on a Sensor That Arrives a Second Late

    I built a Guitar Hero style drums game played with a Whoop on one wrist and an Apple Watch on the other. The Whoop delivers every hit up to a second late, and the game still scores it as Perfect. Here is how, and what I learned.

    tags: side-projects, dev-tools

  2. How It Feels to Learn JavaScript in 2026

    Ten years ago, "How it feels to learn JavaScript in 2016" buried a simple page under React, Babel, Webpack and two dozen templating engines. I rewrote the same conversation for the age of coding agents: nobody learns Webpack anymore, and the fatigue moved up a layer.

    tags: dev-tools, ai-agents

  3. Structured Outputs on Bedrock: The Shape Is Guaranteed, the Content Is Not

    AWS made structured outputs generally available for Claude on Bedrock, and it reads like permission to delete your parsing and validation layer. We measured it against our JSON-plus-Zod path, with a noise floor. It works, it helped on one schema, it returned valid but empty JSON on another, and the AI SDK still strips the constraints you care about.

    tags: llms, architecture, ai-agents

  4. Tap to Call: Turning Our AI Voice Interview into a Native WhatsApp Call

    Candidates applied on WhatsApp and then had to leave the chat for a voice interview in the browser. Now they tap a button and the interview is a WhatsApp call. No provider could carry the audio for us, so we built a small bridge on AWS: how a call flows, how we made it invite-only, and the gotchas along the way.

    tags: ai-agents, architecture, recruiting

  5. Published, Healthy, and Broken: The WhatsApp Flow Meta Accepted and No Phone Could Open

    Meta's API reported our WhatsApp Flow as published, with no validation errors and every health check green. Every candidate who opened it got "something went wrong". The cause was three repeated ids in a dropdown, and fixing it led us to run our own checks on every Flow before it goes to Meta.

    tags: architecture, recruiting, dev-tools

  6. The CI Tax Nobody Budgeted: Why We Moved to a Merge Queue Once AI Started Writing Our Code

    Since January, the merges into our main branch tripled and our test suite grew 74 times. Half of every pull request's CI minutes were going to one thing: re-testing it because main had moved. Why that cost grows faster than the code does, how we moved to a GitHub merge queue, what changed in the first week, and why the queue removes the need to update a branch but not the habit.

    tags: dev-tools, ai-agents, architecture

  7. Not Everything That Glitters Is Gold: The Guardrail That Cost Our Voice Agent Two Seconds per Turn

    Our ElevenLabs voice agent took three seconds to answer every candidate. The turn detector, the LLM, the audio and the browser were all innocent: a safety guardrail we had turned on ourselves, documented at 200–500 ms, was costing about two seconds on every reply. What the platform gives you, what we had to build to find it, every hypothesis we ruled out, and what is still missing.

    tags: ai-agents, llms, architecture

  8. The Compound Interest of an ATS: Your Next Search Shouldn't Start From Zero

    If you have been hiring for five years, you should have five years of accumulated advantage. Most teams don't, because they use their ATS as a board instead of a talent base. Where the knowledge leaks, what Ashby, Greenhouse, Manatal, Spott and Selenios are doing about it, and the habits that make every search easier than the last.

    tags: recruiting, ai-agents, leadership

  9. Batch Inference: Same Model, Half the Price, and Production Never Notices

    Every major LLM provider sells a batch mode at 50% off, and most teams still run their bulk work on-demand. How we used it to migrate tens of thousands of CVs at half the cost without touching production, what it charges you in exchange, and the checklist we learned the hard way.

    tags: llms, recruiting, architecture

  10. Jev Doesn't Generate. That's the Point: Building a Browser QA Loop on TypeSafe's System One Model

    A model that returns calibrated decisions instead of text landed on Vercel's AI Gateway this week and became its fastest-adopted launch. I built a Playwright QA loop where Jev judges every step, measured it, and compared it to the LLMs you already pay for.

    tags: ai-agents, dev-tools

  11. Next.js 16.3 From the CTO Seat: Five Apps, One Month, Three Bugs

    We adopted Next.js 16.3 across five production apps the week it shipped. What paid off, how Cache Components quietly changed our HTTP status codes, and why experimental still means experimental.

    tags: nextjs, leadership

  12. Ralph Loop: The AI Development Methodology Taking Over 2026

    Discover the Ralph Wiggum technique—an autonomous AI agent loop that lets Claude Code work continuously until your project is complete.

    tags: ai-agents, dev-tools

  13. MCP: The Protocol Connecting AI to Everything

    Exploring Anthropic's Model Context Protocol (MCP) and how it's revolutionizing the way AI models interact with external tools and data sources.

    tags: ai-agents, llms

  14. Claude Code: AI-Powered Development in Your Terminal

    My experience using Claude Code CLI for everyday development tasks and why it's become an essential part of my workflow.

    tags: dev-tools, ai-agents

  15. Why Hono is My Go-To Framework for Edge Computing

    A deep dive into Hono, the ultrafast web framework that works seamlessly across Cloudflare Workers, Deno, Bun, and Node.js.

    tags: edge, architecture

  16. Durable Execution: Building Reliable Systems with Temporal

    Understanding durable execution patterns and how Temporal makes building fault-tolerant distributed systems accessible.

    tags: architecture

  17. Replay Conference 2024 Highlights

    Highlights and resources from the Temporal Replay Conference 2024 in Seattle.

    tags: events, architecture

  18. Exploring Cloudflare's Hidden Gems

    Discover how Cloudflare Tunnels, Workers, and Email Redirects can supercharge development for startups and proof-of-concepts.

    tags: edge, architecture