Engineering deep dives on agents, limits, and local orchestration

Engineering

Why Claude Rate Limits Cost Engineers 3+ Hours a Week

You sit down for a planned three-hour refactor. Claude Code is humming. The agent is two tool calls into a non-trivial migration when the response …

26 July 2026 · 11 min read

Engineering

Local-First AI Coding Tools: A Security Reality Check

"Local-first" is on every AI coding tool's landing page now, and it means almost nothing until you ask the only question that matters: where does y…

22 July 2026 · 8 min read

Engineering

Understanding TPM and RPM Rate Limits in Claude Code

The five-hour and seven-day windows get all the attention. The per-minute ceilings — tokens per minute and requests per minute — are what actually …

17 July 2026 · 9 min read

Engineering

Pre-emptive rotation: switching before the 429

A 429 is a failure mode dressed up as a status code. By the time your client reads it, the request has already been spent. The turn that was mid-fl…

22 June 2026 · 11 min read

Engineering

Pre-emptive rotation: rotate before you hit a 429

A Claude rate limit 429 error doesn't arrive as a warning. It arrives as a post-mortem: by the time your code reads the status, the request has alr…

19 June 2026 · 6 min read

Engineering

Manage agent worktrees without losing work

Agent fleets pile up git worktrees. This guide shows how to see every session tree relative to local main, refresh only when it is safe, and reclai…

17 June 2026 · 2 min read

Engineering

Where your code goes when you run Power Claude

Before a tool touches a codebase, someone eventually asks the only question that matters: where does our code go now? For Power Claude the answer i…

11 May 2026 · 6 min read

Engineering

Why LLM Context Windows Degrade Before Their Limit

A 200K-token context window does not mean 200K tokens of reliable reasoning. Accuracy on retrieval and multi-step tracking starts falling long befo…

27 February 2026 · 11 min read

Engineering

I Tried to Break Power Claude's Rotation. Does It Actually Work?

Does Power Claude rotation actually work, or is "pre-emptive" just a marketing verb? I spent an afternoon pointing parallel Claude Code sessions at…

20 February 2026 · 10 min read

Engineering

How Power Claude's Account Rotator Works: Proxy Core and Watchdog

Power Claude is two pieces of code: a local proxy between Claude Code and Anthropic's API, and a watchdog process that observes session lifecycle e…

14 February 2026 · 9 min read

Engineering

GitLens for AI Sessions: Navigating Claude's Edit History

GitLens taught a generation of VS Code users that git history is a navigation surface, not just a forensic tool. Hover-to-blame, inline diffs, file…

11 February 2026 · 8 min read

Engineering

Editor Blame for AI-Edited Code: Who Changed What and When

Git blame is the most-used code-archaeology tool a developer owns. It tells you who changed a line and when. It does not tell you which Claude sess…

06 February 2026 · 7 min read

Engineering

Detecting and Recovering Crashed Claude Code Sessions

Claude Code does not ship a heartbeat. It also does not ship a way to tell, from outside the editor, whether a long-running session is mid-thought,…

02 February 2026 · 10 min read

Engineering

What data leaves your machine when using Power Claude

The short answer is in the heading: nothing meaningful. The full answer is what this article exists to spell out, because "trust us, it is local" i…

30 January 2026 · 11 min read

Engineering

Why Heavy Sub-Agent Fan-Outs Hit Claude Rate Limits

Sub-agent fan-outs feel like parallelism. To your Claude.ai account, they look like a stampede. Eight Task spawns from a single parent push nine si…

07 January 2026 · 11 min read