Give every database a clear job
Multiple persistence layers can work well in one SaaS product when each piece of data has an authoritative owner, a recovery story, and a deliberate lifecycle.
Blog · Engineering notes
Notes from building software.
Practical writing on AI and agent systems, evidence, developer workflows, and the engineering work around the model.
Treat credit-based AI billing as a state machine
A usage ledger, idempotent payment events, and explicit charging rules keep paid AI workflows consistent when requests fail or webhooks retry.
Make acceptance criteria the unit of agent verification
Connect each requirement to a versioned check and its evidence so a green test run tells a reviewer what the agent actually proved.
Designing approval boundaries for agentic systems
Useful approval flows show the exact action and its consequences at the moment an agent crosses from analysis into change.
Lessons from building browser automation in production
Browser automation becomes a service only after jobs, resource limits, persistence, partial results, and recovery are designed around it.
Why evidence provenance matters in AI research systems
A citation should be a resolvable relationship between a claim and the evidence that supports it, not a decorative link added after generation.
Building local-first AI developer tools
A local-first coding agent needs a clear host boundary for workspace access, provider credentials, tools, permissions, and state across client surfaces.
What production AI applications need beyond a prompt
Production AI systems need typed contracts, identity, persistence, budgets, failure handling, user review, and operational evidence around generation.
Designing provider-agnostic LLM applications
A provider interface should isolate meaningful differences in streaming, tools, errors, and capabilities instead of hiding them behind a lowest-common-denominator request.
Building reliable AI agents with MCP and tool calling
Reliable agent workflows depend on explicit tool contracts, bounded execution, validated results, and traceable state around the model call.
Credence: A Professional Network That Actually Works for Tech Professionals
How Credence addresses LinkedIn's growing problems with verified accounts, real job listings, GitHub analysis, and a culture focused on proof of work instead of follower counts.
Building a Zero-Downtime CI/CD System with Webhooks
How I built a self-hosted CI/CD system that gives you zero-downtime deployments with webhook automation and one-command setup, solving the deployment gap when moving to self-hosting.
Build a RESTful API with Express.js and Supabase
Learn how to build a RESTful API using Node.js, Express.js, and Supabase for a Dinosaur Facts Database.
Creating a GUI For yt-dlp
In this detailed guide, we'll create a Python GUI application for Windows that allows users to download media from YouTube using yt-dlp and convert it to either MP3 or MP4 format. The application will be built using the tkinter library for the GUI, and we'll use yt-dlp and ffmpeg for downloading and converting the videos.
Written by Jared Hooker · [email protected]