03.Build / learn / document
Docs
& notes
Practical notes on product development, engineering, and running a one-person company in collaboration with AI.Browse the docs ↓
Notes from real projects
Field notes
Vibe Coding at 438,000 Lines: How I Keep the Project From Drifting
A StratThread case study on running five applications, eleven shared packages, and 174 database migrations through specifications, explicit state authority, verification, and release gates.
↗ 02 engineering practiceFrontend Jank: Give Service Workers, Web Workers, and React the Right Jobs
A practical guide to reducing repeat I/O with a Service Worker, moving measured CPU work into Web Workers, and isolating React renders before they reach expensive charts.
↗ 03 engineering practiceFrontend Performance: Lessons From a 325-Byte Bundle Failure
A practical client optimization review covering boot-graph budgets, lazy boundaries, stable vendor chunks, batched initialization reads, and event-driven refresh.
↗ 04 engineering practiceFrontend or Backend Agent Tool? Keep Display and Storage Consistent
Operations that affect frontend state and must keep the displayed and persisted result consistent should call existing frontend business actions. Stores connect those actions to the UI.
↗ 05 engineering practiceTool-call optimization: stop using context as a data transfer layer
General tool-call design patterns for input references, backend working storage, compact results, on-demand reads, and asynchronous continuation, illustrated by a market-analysis implementation.
↗ 06 engineering practiceLow tool-call success? Fix these six layers before switching models
A practical Agent optimization review covering tool scope, schemas, repair results, retries, state, and observability before model comparison.
↗ 07 engineering practiceA Shorter Prompt Won’t Fix Your Agent: 10 Problems I Hit
A weak Agent is not always a weak model. Real numbers show how context, tools, state, streaming, and useful system records shape the result together.
↗ 08 engineering practicePrompt Cache Reuse Dropped From 99% to Zero—Here’s Why
Two production traces show a simple rule: when the order at the start changes, prompt-cache reuse can fall from 99% to zero even if the content looks almost the same.
↗ 09 Getting startedInside a One-Person AI Company: What I’m Building Right Now
A current record of the products, AI infrastructure, and GeekCode writing I am working on—including what is not finished yet.
↗ 10 engineering practiceShould You Test Vibe-Coded Software? Use 4 Risk Layers, Not Coverage
Vibe Coding needs tests, but more is not always better. Start with what a change could break, then choose the smallest useful check for that risk.
↗ 11 engineering practiceWhy Vibe Coding Gets Unstable—and the Engineering System I Use
AI can write code quickly, but quick is not the same as safe. Define the promise first, write the code second, then check and record what really happened.
↗ 12 engineering practiceSlow API? Don’t Add Machines Yet—Find the Bottleneck in 5 Steps
The first job in a slow-endpoint investigation is not “optimization.” It is finding out where the time actually goes.
↗ 13 ai perspectivesYour AI Demo Works. Can It Survive Production? 4 Guardrails
A feature that demos well is not necessarily ready to ship. Keep uncertainty at the boundary before asking users to rely on it.
↗ 14 independent developmentSet Up Your First Environment
A practical environment-setup guide is being prepared for this chapter.
↗ 15 independent developmentSaaS or Self-Hosted? 4 Costs Indie Developers Often Miss
How independent developers can weigh hosted services against systems they operate themselves.
↗ 16 independent developmentWhy Every Developer Should Ship an Indie Product: 5 Reasons
A comprehensive guide to understanding independent development, its charm, and value
↗ 17 Getting startedGetting started
An introduction to GeekCode, what it builds, and how this documentation is organised.
↗ 18 ai perspectivesWhat Is MCP? The Missing Link Between AI, Tools, and Data
An introduction to how the Model Context Protocol connects LLM applications with tools and external information.
↗