Deep dives into projects built with AI pair programming using Claude Code and Qwen.
A practical guide to canarying model upgrades for AI coding agents with eval slices, fallback lanes, budget guardrails, and promotion scorecards so model refreshes do not quietly degrade code quality.
Read More →
A practical guide to freshness guards for long-running AI coding agents using repo fingerprints, file watchlists, context TTLs, and revalidation gates so stale context does not quietly turn into bad patches.
Read More →
A practical guide to versioning MCP servers with compatibility contracts, capability negotiation, deprecation windows, and handshake tests so MCP upgrades stay boring instead of turning into client roulette.
Read More →
A practical guide to tool capability manifests for AI agents using side-effect metadata, cost hints, auth lanes, and planner scoring so planners stop treating every tool as equally safe, cheap, and reversible.
Read More →
A practical guide to backpressure for AI agents using queue budgets, concurrency caps, lease-aware workers, and overload signals so tool-calling systems stop amplifying latency into self-inflicted outages.
Read More →
A practical guide to AST codemods for AI coding agents using syntax-aware transforms, fixture-led verification, and rollout checkpoints so large API migrations stay reviewable instead of devolving into regex cleanup.
Read More →
A practical guide to dependency update lanes for AI coding agents using risk tiers, SBOM diffs, focused verification, and staged promotion so automated upgrades stay fast without turning every version bump into reviewer roulette.
Read More →
A practical guide to environment manifests for AI coding agents using pinned toolchains, seeded fixtures, and verifier snapshots so patches reproduce the real bug instead of fixing a local fiction.
Read More →
Design AI agent tool calls to fail loudly, classify errors, retry safely, and surface human-actionable recovery paths instead of burying broken automation behind vague exceptions.
Read More →
A practical guide to speculative patch generation for AI coding agents using parallel candidate branches, verifier scorecards, and deterministic promotion rules so teams can explore multiple fixes without turning code review into chaos.
Read More →
A practical comparison of hybrid local and cloud AI development workflows using task routing, privacy lanes, local-first defaults, and explicit escalation rules so teams get faster iteration without turning every prompt into a cloud dependency.
Read More →
A practical guide to secrets management for AI agents using brokered access, short-lived credentials, OIDC, scoped injection, and audit trails that reduce blast radius without killing automation.
Read More →
A practical guide to semantic caching for AI coding agents using fingerprinted prompts, repo-aware invalidation, trust boundaries, and fallback verification so cached answers stay fast without becoming quietly wrong.
Read More →
A practical guide to verifier pipelines for AI coding agents using staged checks, AST-aware rules, policy gates, and escalation paths that catch risky edits before they become reviewer cleanup.
Read More →
A practical guide to capability leases for AI agents using short-lived grants, approval-bound scopes, and audit trails so powerful tools stay available without becoming always-on risk.
Read More →
A practical guide to capability leases for AI agents using short-lived grants, approval-bound scopes, and audit trails so powerful tools stay available without becoming always-on risk.
Read More →
A practical guide to stopping cascading failures in AI agents with circuit breakers, budget guards, and cooldown policies around model calls and external tools.
Read More →
How to debug multi-step AI agent failures with trace IDs, span annotations, replay packets, and failure buckets so incidents stop feeling like prompt necromancy.
Read More →
How to route coding-agent tasks between fast and expensive models using task scoring, verification, and escalation rules that reduce cost without quietly lowering quality.
Read More →
A practical guide to why huge advertised context windows still fail in coding workflows, with concrete patterns for bounded context packets, rolling summaries, pinned files, and retrieval refresh loops.
Read More →
A practical guide to shipping AI features safely with shadow traffic, side-by-side scoring, redaction, and promotion gates that expose failures before the model gets permission to act.
Read More →
A practical guide to using AI-generated tests for fast coverage gains without trusting them blindly, including golden paths, mutation checks, flaky-test traps, and review rules that expose weak assertions.
Read More →
A practical guide to hybrid code retrieval for AI coding agents using embeddings, tree-sitter structure, ripgrep fallback, and reranking so prompts stay smaller and edits stay more accurate.
Read More →
A practical guide to choosing 4-bit, 6-bit, 8-bit, and mixed quantization for coding models on consumer hardware, with concrete runtime tradeoffs for Ollama, llama.cpp, and vLLM.
Read More →
A practical guide to shipping database changes with AI coding agents using additive schema changes, dual writes, bounded backfills, and verification steps that preserve rollback paths.
Read More →
A practical guide to planner-executor handoffs for AI agents using task manifests, claim checks, verification gates, and rollback-friendly execution so multi-agent workflows stay understandable under real load.
Read More →
A practical guide to running scheduled AI workflows with clean task payloads, isolated runs, idempotency guards, and approval-aware delivery so cron automation stays useful instead of becoming a haunted queue.
Read More →
A practical guide to isolating long-lived AI agents with session-scoped memory, worktrees, tool lanes, and approval boundaries so one task cannot quietly contaminate another.
Read More →
A practical guide to building eval harnesses for AI coding agents with patch tasks, invariant checks, sandboxed verification, and scorecards that expose bad fixes before they hit code review.
Read More →
A practical guide to using GitHub Actions OIDC for AI agents so CI jobs can assume short-lived cloud roles for deploys, eval runs, and controlled automation without storing long-lived cloud secrets.
Read More →
A practical guide to stopping prompt injection through tool outputs by treating external content as tainted data, using trust lanes, sanitizers, execution guards, and human review boundaries.
Read More →
A practical guide to building context packets for engineering agents with task manifests, repo maps, bounded evidence bundles, and relevance scoring that improve edit quality.
Read More →
A practical guide to approval workflows for AI agents that can write code, open PRs, and call tools without turning every useful action into a security incident.
Read More →
A practical comparison of Ollama, llama.cpp, and vLLM for local AI development, covering setup friction, quantization, API compatibility, and where each runtime actually fits.
Read More →
A practical guide to building schema-first tool contracts for AI agents with JSON Schema, runtime validation, adapter layers, and fail-closed execution paths that keep tool use reliable and reviewable.
Read More →
A practical guide to making long-running AI agent jobs retryable and recoverable with idempotency keys, durable checkpoints, and safer human handoff paths.
Read More →
A practical guide to choosing between stdio, HTTP, and SSE for MCP servers, with deployment patterns, auth tradeoffs, and failure modes that matter in real agent systems.
Read More →
A detailed step-by-step guide to running Claude Code with Unsloth Qwen3.5-35B-A3B on a Mac mini M4 Pro, including model choice, local inference setup, KV cache fixes, tuning, and practical tradeoffs.
Read More →
A practical guide to running multiple AI coding agents safely with Git worktrees, branch-per-task isolation, runtime guardrails, and merge discipline that keeps parallel work fast and reviewable.
Read More →
A practical guide to using Gemini CLI for terminal-first debugging with plan mode, shell tooling, focused context, and repeatable fix loops that keep production bug hunts fast and reviewable.
Read More →
A detailed guide to designing reusable OpenClaw skills with a scheduled repo worker example, including project-memory vs task-engine modes, repo-local state, PR packaging runs, and reviewer-in-the-loop automation.
Read More →'/%3E%3Ccircle cx='990' cy='125' r='150' fill='%2327d3ff' fill-opacity='0.12'/%3E%3Ccircle cx='175' cy='520' r='195' fill='%238d6bff' fill-opacity='0.14'/%3E%3Crect x='105' y='110' width='990' height='410' rx='28' fill='%230f172a' stroke='%2338bdf8' stroke-opacity='0.35'/%3E%3Ctext x='150' y='218' fill='white' font-family='Arial, sans-serif' font-size='48' font-weight='700'%3ESandboxing AI Coding%3C/text%3E%3Ctext x='150' y='290' fill='%2390cdf4' font-family='Arial, sans-serif' font-size='48' font-weight='700'%3EAgents%3C/text%3E%3Ctext x='150' y='376' fill='%23cbd5e1' font-family='Arial, sans-serif' font-size='28'%3EDev containers, read-only mounts, and microVM boundaries%3C/text%3E%3C/svg%3E)
A practical guide to isolating AI coding agents with dev containers, read-only mounts, explicit network policy, and microVMs so they can stay useful without getting a blank check on your machine.
'/%3E%3Ccircle cx='990' cy='120' r='150' fill='%2327d3ff' fill-opacity='0.12'/%3E%3Ccircle cx='180' cy='520' r='190' fill='%238d6bff' fill-opacity='0.14'/%3E%3Crect x='105' y='110' width='990' height='410' rx='28' fill='%230f172a' stroke='%2338bdf8' stroke-opacity='0.35'/%3E%3Ctext x='155' y='220' fill='white' font-family='Arial, sans-serif' font-size='50' font-weight='700'%3ERepo Playbooks for%3C/text%3E%3Ctext x='155' y='292' fill='%2390cdf4' font-family='Arial, sans-serif' font-size='50' font-weight='700'%3EAI Coding Agents%3C/text%3E%3Ctext x='155' y='378' fill='%23cbd5e1' font-family='Arial, sans-serif' font-size='28'%3EAGENTS.md, CLAUDE.md, and rules that reduce drift%3C/text%3E%3C/svg%3E)
A practical guide to using repo-specific instruction files like AGENTS.md and CLAUDE.md to make AI coding agents safer, more consistent, and much easier to review.
'/%3E%3Ccircle cx='990' cy='120' r='150' fill='%2327d3ff' fill-opacity='0.12'/%3E%3Ccircle cx='175' cy='520' r='190' fill='%238d6bff' fill-opacity='0.14'/%3E%3Crect x='105' y='110' width='990' height='410' rx='28' fill='%230f172a' stroke='%2338bdf8' stroke-opacity='0.35'/%3E%3Ctext x='155' y='220' fill='white' font-family='Arial, sans-serif' font-size='54' font-weight='700'%3ECursor Rules for%3C/text%3E%3Ctext x='155' y='292' fill='%2390cdf4' font-family='Arial, sans-serif' font-size='54' font-weight='700'%3EMonorepos%3C/text%3E%3Ctext x='155' y='378' fill='%23cbd5e1' font-family='Arial, sans-serif' font-size='28'%3ELayered guardrails, scoped context, and faster verification loops%3C/text%3E%3C/svg%3E)
A practical guide to using Cursor effectively in large monorepos with layered rules, scoped context, repo maps, verification loops, and guardrails that stop AI edits from wandering across the codebase.
A practical guide to reviewing AI-generated pull requests with scoped diffs, invariant checklists, CI guardrails, and fast human approval loops.
'/%3E%3Ccircle cx='980' cy='120' r='145' fill='%2327d3ff' fill-opacity='0.12'/%3E%3Ccircle cx='180' cy='520' r='190' fill='%238d6bff' fill-opacity='0.14'/%3E%3Crect x='105' y='110' width='990' height='410' rx='28' fill='%230f172a' stroke='%2338bdf8' stroke-opacity='0.35'/%3E%3Ctext x='155' y='220' fill='white' font-family='Arial, sans-serif' font-size='50' font-weight='700'%3EPrompt Caching for%3C/text%3E%3Ctext x='155' y='292' fill='%2390cdf4' font-family='Arial, sans-serif' font-size='50' font-weight='700'%3ETool-Heavy AI Agents%3C/text%3E%3Ctext x='155' y='378' fill='%23cbd5e1' font-family='Arial, sans-serif' font-size='28'%3EStable prefixes, compaction, and lean retrieval for lower latency%3C/text%3E%3C/svg%3E)
A practical guide to reducing latency and cost in tool-heavy AI agents with stable prompt prefixes, prompt caching, context compaction, and retrieval boundaries.
'/%3E%3Ccircle cx='980' cy='120' r='150' fill='%2327d3ff' fill-opacity='0.12'/%3E%3Ccircle cx='180' cy='520' r='190' fill='%238d6bff' fill-opacity='0.14'/%3E%3Crect x='105' y='110' width='990' height='410' rx='28' fill='%230f172a' stroke='%2338bdf8' stroke-opacity='0.35'/%3E%3Ctext x='155' y='220' fill='white' font-family='Arial, sans-serif' font-size='52' font-weight='700'%3ETracing AI Agents%3C/text%3E%3Ctext x='155' y='292' fill='%2390cdf4' font-family='Arial, sans-serif' font-size='52' font-weight='700'%3Ewith OpenTelemetry%3C/text%3E%3Ctext x='155' y='378' fill='%23cbd5e1' font-family='Arial, sans-serif' font-size='28'%3EDebug tool calls, latency, tokens, and multi-step failures%3C/text%3E%3C/svg%3E)
A practical guide to using OpenTelemetry and OpenInference to trace agent workflows, inspect tool calls, track token spend, and debug multi-step failures without drowning in flat logs.
'/%3E%3Ccircle cx='995' cy='120' r='145' fill='%2327d3ff' fill-opacity='0.12'/%3E%3Ccircle cx='185' cy='515' r='190' fill='%238d6bff' fill-opacity='0.14'/%3E%3Crect x='105' y='110' width='990' height='410' rx='28' fill='%230f172a' stroke='%2338bdf8' stroke-opacity='0.35'/%3E%3Ctext x='155' y='220' fill='white' font-family='Arial, sans-serif' font-size='52' font-weight='700'%3ELocal LLM Dev Environments%3C/text%3E%3Ctext x='155' y='292' fill='%2390cdf4' font-family='Arial, sans-serif' font-size='52' font-weight='700'%3EFast, Cheap, and Private%3C/text%3E%3Ctext x='155' y='378' fill='%23cbd5e1' font-family='Arial, sans-serif' font-size='28'%3EOllama, Open WebUI, coding clients, and smart fallbacks%3C/text%3E%3C/svg%3E)
A practical guide to building a local LLM development environment with Ollama, Open WebUI, coding tools, quantization choices, prompt/version hygiene, and smart fallbacks when local models are not enough.
'/%3E%3Ccircle cx='985' cy='130' r='145' fill='%2327d3ff' fill-opacity='0.12'/%3E%3Ccircle cx='175' cy='520' r='190' fill='%238d6bff' fill-opacity='0.14'/%3E%3Crect x='100' y='110' width='1000' height='410' rx='28' fill='%230f172a' stroke='%2338bdf8' stroke-opacity='0.35'/%3E%3Ctext x='150' y='225' fill='white' font-family='Arial, sans-serif' font-size='52' font-weight='700'%3EPractical RAG Evaluation%3C/text%3E%3Ctext x='150' y='295' fill='%2390cdf4' font-family='Arial, sans-serif' font-size='52' font-weight='700'%3EThat Catches Real Failures%3C/text%3E%3Ctext x='150' y='380' fill='%23cbd5e1' font-family='Arial, sans-serif' font-size='28'%3EScore retrieval, grounding, and regressions separately%3C/text%3E%3C/svg%3E)
A practical guide to evaluating RAG systems by separating retrieval from generation, building grounded test sets, tracking the right metrics, and running regression checks before users find the bugs.
'/%3E%3Ccircle cx='1015' cy='125' r='145' fill='%2327d3ff' fill-opacity='0.12'/%3E%3Ccircle cx='180' cy='525' r='190' fill='%238d6bff' fill-opacity='0.14'/%3E%3Crect x='105' y='110' width='990' height='410' rx='28' fill='%230f172a' stroke='%2338bdf8' stroke-opacity='0.35'/%3E%3Ctext x='155' y='225' fill='white' font-family='Arial, sans-serif' font-size='54' font-weight='700'%3EAgent Memory Strategies%3C/text%3E%3Ctext x='155' y='295' fill='%2390cdf4' font-family='Arial, sans-serif' font-size='54' font-weight='700'%3Efor Production%3C/text%3E%3Ctext x='155' y='380' fill='%23cbd5e1' font-family='Arial, sans-serif' font-size='28'%3EWorking memory, write policies, retrieval, and drift control%3C/text%3E%3C/svg%3E)
A practical guide to designing memory for AI agents with working memory, episodic summaries, semantic facts, write policies, retrieval rules, and debugging loops that stay useful over time.
'/%3E%3Ccircle cx='1010' cy='140' r='150' fill='%2327d3ff' fill-opacity='0.12'/%3E%3Ccircle cx='160' cy='520' r='200' fill='%238d6bff' fill-opacity='0.14'/%3E%3Crect x='105' y='110' width='990' height='410' rx='28' fill='%230f172a' stroke='%2338bdf8' stroke-opacity='0.35'/%3E%3Ctext x='155' y='225' fill='white' font-family='Arial, sans-serif' font-size='54' font-weight='700'%3EClaude Code Workflows%3C/text%3E%3Ctext x='155' y='295' fill='%2390cdf4' font-family='Arial, sans-serif' font-size='54' font-weight='700'%3Efor Large Refactors%3C/text%3E%3Ctext x='155' y='380' fill='%23cbd5e1' font-family='Arial, sans-serif' font-size='28'%3ESpecs, checkpoints, and verification loops that scale%3C/text%3E%3C/svg%3E)
A practical guide to using Claude Code for large refactors with specs, checkpoints, verification loops, and guardrails that keep big changes understandable.
'/%3E%3Ccircle cx='960' cy='120' r='150' fill='%2327d3ff' fill-opacity='0.12'/%3E%3Ccircle cx='180' cy='520' r='190' fill='%238d6bff' fill-opacity='0.14'/%3E%3Crect x='110' y='110' width='980' height='410' rx='28' fill='%230f172a' stroke='%2338bdf8' stroke-opacity='0.35'/%3E%3Ctext x='165' y='235' fill='white' font-family='Arial, sans-serif' font-size='56' font-weight='700'%3EHarden OpenClaw%3C/text%3E%3Ctext x='165' y='305' fill='%2390cdf4' font-family='Arial, sans-serif' font-size='56' font-weight='700'%3Eon a VPS%3C/text%3E%3Ctext x='165' y='390' fill='%23cbd5e1' font-family='Arial, sans-serif' font-size='28'%3EKeep the assistant useful without giving it a blank check%3C/text%3E%3C/svg%3E)
A practical guide to hardening OpenClaw on a VPS with least-privilege credentials, network restrictions, approval boundaries, cron hygiene, and operations checks that actually matter.
'/%3E%3Ccircle cx='980' cy='120' r='140' fill='%2327d3ff' fill-opacity='0.14'/%3E%3Ccircle cx='180' cy='520' r='180' fill='%238d6bff' fill-opacity='0.16'/%3E%3Crect x='120' y='120' width='960' height='390' rx='28' fill='%230f172a' stroke='%2338bdf8' stroke-opacity='0.35'/%3E%3Ctext x='170' y='250' fill='white' font-family='Arial, sans-serif' font-size='54' font-weight='700'%3ESecure MCP Server%3C/text%3E%3Ctext x='170' y='320' fill='%2390cdf4' font-family='Arial, sans-serif' font-size='54' font-weight='700'%3EDesign Patterns%3C/text%3E%3Ctext x='170' y='395' fill='%23cbd5e1' font-family='Arial, sans-serif' font-size='28'%3EPractical guardrails for tool-calling agents%3C/text%3E%3C/svg%3E)
A practical guide to building MCP servers that stay useful under real agent workloads without turning into a security liability. Covers tool boundaries, approval flows, validation, session isolation, prompt-injection defenses, and observability.
How I built a collection of 8 browser games (Snake, Pong, Tic Tac Toe, Memory, Breakout, Flappy, 2048, and Sudoku) using Claude Code and Qwen for AI pair programming. Explores game loops, collision detection, AI opponents, and the collaborative development process.
How I built an AI-powered document-to-quiz generator that converts PDFs and textbooks into chapter-level quizzes. Covers document parsing, question generation, quiz engines, and using AI for educational technology.
How I built a comprehensive mortgage cost calculator that models the true cost of homeownership. Includes financial formulas, amortization schedules, data visualization, and AI-assisted TypeScript development.
How I built a comprehensive stock tracking and analysis tool with technical indicators, portfolio management, and visualizations. Covers API integration, RSI, MACD, Bollinger Bands, and AI pair programming for financial tools.
More writeups coming soon...