2 min read
An Agent Quietly Burned Through My Budget Overnight
A loop ran overnight and the bill arrived later. Hard budget caps, 80%/100% alerts, and ask-first-for-more keep agent spend bounded.
Problems
Practical posts on turning recurring work into controlled Task Machine workflows with approvals, checks, agents on your own machines, and a history you can read.
2 min read
I Can't Trust AI Output Enough to Send It
Agent work looks finished but you redo it anyway. Verifier checks, approvals, and step-level history turn plausible output into reviewable evidence.
2 min read
My Agent Work Is Scattered Across Terminals and Chats
Codex here, Claude Code there, decisions lost in scrollback. One shared inbox and durable tasks give agent work a real system of record.
2 min read
My AI Agent Did Something I Never Approved
An agent acted before you saw it, and trust is gone. Approval-first control routes every consequential action to one inbox before it ships.
2 min read
The Recurring Work Keeps Slipping
Outreach, content, reports, and follow-ups keep sliding to next week. The fix is structural: recurring workflows agents run and you approve.