Claude Code repeatedly promises automations are 'done' without testing — weeks of broken deliverables

Platform: Claude Code CLI v2.1.97, macOS (Mac Mini M4)
Plan: Claude Pro (paid subscriber)
User: Small business owner — 5 urgent care clinics, NW Alabama

Summary

I have spent several weeks and dozens of sessions asking Claude Code to set up basic morning automations: an iMessage to my phone at 6 AM with my business financials (from QuickBooks) and clinic patient volume from the previous day. This is a simple, well-defined request. After weeks of sessions, none of it works reliably.

Specific Issues

1. Over-promising, under-delivering across sessions. Multiple sessions have told me automations were "live" and "running" when they were not tested end-to-end. Example: I was told a system watchdog would run daily at 0600 and 1800. The actual LaunchAgent was configured for 0800 and 1700 — and it stopped producing any output after March 27 with no alert sent to me.

1. QBO financial report automation is broken. The daily sync that pulls QuickBooks data has been failing. I get no financial numbers in my morning briefing.

1. Clinic volume numbers are broken. Patient volume data from the previous day is not being delivered.

1. Unwanted redirects to Cowork. Instead of delivering information via iMessage as requested, the system pushes me to "open full briefing in Cowork." The entire point of the automation is that I do NOT want to get on a computer. I want to look at my phone.

1. Cowork opens with a wall of errors. When I do open Cowork, it presents a page of commands it wants me to run — error after error. I am a business owner, not a systems administrator.

1. Getting notifications I didn't ask for. I receive email notifications I don't care about, while the iMessage alerts I actually requested don't arrive.

1. No session continuity. Each new session seems unaware of what previous sessions broke or promised. I end up re-explaining the same requirements repeatedly.

1. The tool expects ME to work for IT. I am paying for an AI assistant. Instead, I am debugging scripts, reading terminal output, and troubleshooting LaunchAgents. This is the opposite of what was promised.

What I Need

• Wake up → look at phone → see my numbers. That's it.
• iMessage delivery. Not email. Not Cowork. Not a terminal.
• If something breaks, tell me it's broken and fix it. Don't silently fail for two weeks.

Impact

• Weeks of lost productivity across dozens of sessions
• Zero reliable automations despite repeated requests
• Eroded trust — I no longer believe what Claude tells me is "working"
• I am a paying customer running a healthcare business. I need tools that deliver, not tools that create more work.

Request

I would like someone from Anthropic to review this pattern. Claude Code is a powerful tool, but the lack of session-to-session reliability and the pattern of declaring things "done" without verification is a serious problem for business users who depend on it.

---
Filed: April 9, 2026
Claude Code v2.1.97