The hard part of a self-service agent platform isn't the agents
Gartner expects 40% of enterprise apps to ship AI agents by end of 2026. After building AgentHive at JFrog, I think generating the agent is the easy part.
LB/ WRITING
Blog
Dev tools, design, AI, and some stuff. Not in that order.
00 ENTRIES
Boost compressed 212M tokens out of my agent sessions. Here are the receipts
Boost wraps every CLI my agents touch. Across 964 sessions it compressed 212M tokens (63.4%) and saved $1,060. The three pillars and what they actually do.
Why every Ghostty tab said 'Claude Code' (and the 30-line zsh hook that fixed it)
Every Ghostty tab said 'Claude Code'. Two configs and a 30-line zsh hook turned the tab bar back into useful navigation across 4-pane workflows.
Claude Code hooks: the half of Claude Code nobody uses
Skills feel like capability. Hooks are enforcement. Inside: the full event reference, 10 real hooks (one cuts tokens 73%), and the enterprise patterns nobody writes.
ForgeCode vs Claude Code: which AI coding agent actually wins?
Claude Code's speed was driving me and my team crazy. I tried ForgeCode. The speed was real. Now I'm running both.
Cursor 3 shipped parallel agents, but is any of it new?
Cursor 3's parallel agents existed in v2 via worktree config. Users report $2,000 bills in two days. The real story is cost opacity and unsolved context sharing.
Undercover mode, decoy tools, and a 3,167-line function: inside Claude Code's leaked source
Anthropic shipped a source map in Claude Code v2.1.88, exposing 1,700+ files, undercover mode, decoy tools, and a 3,167-line function. Then they DMCA'd 8,100 repos.
I found my database password in a Claude Code session file
How finding leaked credentials in a Claude Code session file led me to build Claudoscope, a macOS menu bar app for session analytics, cost tracking, and secret scanning.
NO POSTS MATCH THIS FILTER