Category
Guides
Step-by-step guides and how-tos for getting AI tools working in your actual development workflow — setup, configuration, and practical tips.
Is ZCode safe? What the open-sourced code shows
Z.ai open-sourced ZCode after it was caught uploading .git histories. We read the source: what the changelog won't say, and one default to change before CI.
AGENTS.md vs CLAUDE.md: which file Claude Code reads now
Claude Code 2.1.277 reads AGENTS.md when there's no CLAUDE.md. We tested 17 setups: a CLAUDE.local.md or DISABLE_TELEMETRY quietly turns it off.
Spotify's 90% Claude Code saving: what your repo gets
Spotify says its shunt plugin cuts 90% of tokens on bulk file reads. We counted 5,131 reads across 761 real Claude Code sessions to see how often it would fire.
How to make Claude Code less verbose: 3 fixes, measured
We measured Claude Code's default output against a CLAUDE.md rule, the 32K-star i-have-adhd skill, and the built-in Concise style. The built-in wins.
Using any LLM with Codex CLI: what the proxy sends
We routed Codex CLI through a universal provider proxy and captured the wire: 51,731 bytes left for a five-word prompt, and half the guides online are broken.
Is OpenCode private by default? Reading the source
Two privacy claims about OpenCode get repeated everywhere. We read the source at tag v1.18.28: one is stale, the other is narrower than it sounds.
What GitHub Copilot sends: we ran it through mitmproxy
We put VS Code behind mitmproxy and counted every byte Copilot uploads: 890KB of telemetry before sign-in, what a completion carries, and what agent mode costs.
Cursor usage page shows tokens, not dollars: find the cost
Since July 31 Cursor's Usage page is tokens-only on every self-serve plan. Where the dollars still live, how to rebuild the rest, and the workaround that fails.
Claude Code auto mode: what the classifier really blocks
Auto mode became Claude Code's default on August 14. We ran 24 commands through the classifier on a throwaway repo. It blocked one. Here's what it misses.
Hermes Agent: what 'self-improving' actually means
Nous Research's Hermes Agent claims the only built-in learning loop. Our code-read review: three mechanisms, their defaults, what they cost per turn, where data goes.
GitLost: can a public issue leak your private repos?
We read the issue, the lock file and the run behind Noma's GitLost disclosure to see which of GitHub's own agent guardrails should have caught it.
Cursor Router: does it actually save money?
Cursor says Router cuts model spend 60%. We checked what that number is measured against, what the docs say it really costs, and who can turn it on.
Cursor Gmail plugin permissions: what you grant
We installed Cursor's Gmail plugin and read the OAuth request and the logs: three restricted scopes, 29 tools, and a write to the mailbox with no prompt.
Claude Code token overhead: what 33k actually costs
Claude Code sends about 30,000 tokens before it reads your prompt. We measured it, priced it with caching on, and checked the 50k claim about MCP servers.
Codex CLI code review: how /review actually works
What the /review command in OpenAI's Codex CLI reviews, the rubric it grades your diff against, what it can't see, and how to run it in CI.
Claude API error 529: what it means and how to fix it
A 529 overloaded_error means Anthropic's API is saturated, not that you did something wrong. How it differs from 429, why your SDK already retries it.
Windsurf is now Devin Desktop: what actually shipped
Cognition renamed Windsurf to Devin Desktop and replaced Cascade with Devin Local. A fresh install can still land you on Cascade — here's why, and how to switch.
How to set up MCP servers in Claude Code and Cursor
The commands, the config files, and the three server states you'll actually see — plus what the two tools do differently when you add too many servers.
Claude Opus 5: what's new, pricing, and how it compares
Claude Opus 5 is Anthropic's new coding model, out July 24, 2026 at the same price as Opus 4.8. What changed, and whether it's worth switching.
Kimi K3: what the 2.8T open-weight model means for devs
Moonshot's Kimi K3 hits frontier benchmarks at $3/$15 per million tokens. What to know about pricing, token burn, and the license attached to the open weights.
Are AI coding assistants safe for proprietary code?
What Copilot, Cursor, Claude Code, and Codex actually do with your code — training defaults, retention, the Alibaba ban, and how to use them without leaking IP.
Why companies are suddenly capping AI token spend
Uber burned its 2026 AI budget in four months. One company spent $500M on Claude in a month. Token caps are coming to your team — here's what changed.
8 engineering levers to cut LLM API costs
Prompt caching, model routing, context trimming, output caps, batching, and measuring real tokens — the levers that actually move an LLM bill.
What is Google Antigravity? The 4 surfaces explained
Google Antigravity ships as four surfaces — Antigravity 2.0, IDE, CLI, and SDK. Here's what each one is, how they differ, and when to use which.
What is Claude Fable 5? Features, pricing, and how to use it
Claude Fable 5 is Anthropic's most capable widely released model. Here's what it does, what it costs, and the API behavior developers need to know.
Loop engineering: how self-correcting AI agents work
Loop engineering is building AI systems that act, check their own work, and retry until a goal is met. The anatomy of a loop, a template, and when to skip it.
How to set up GitHub Copilot in VS Code (2026 guide)
Install and configure GitHub Copilot in VS Code: the free vs. paid tiers, the settings worth changing, and how to get useful completions from day one.
Is GitHub Copilot safe? Privacy and code security explained
What GitHub Copilot does with your code, what it doesn't store, enterprise vs individual privacy settings, and how it compares to alternatives on security.
The complete guide to AI coding assistants in 2026
A practical guide to every major AI coding assistant in 2026 — Copilot, Cursor, Claude Code, Codex, Devin, and more — with pricing and how to pick.