Failure log — automation that got built and then deleted

The failure log is one of the most useful documents in the vault. Building something is easy. Acknowledging that you built it, used it, and then deleted it — and understanding exactly why — is where the real learning is.

Automation 1: Auto-summarised meeting notes (deleted at week 6)

Built to automatically summarise all client meeting notes using a fixed prompt. Deleted because the fixed prompt produced uniform outputs that stripped context-specific nuance. The summaries were accurate but not useful — they couldn’t distinguish between a note where the client was agreeing and a note where the client was agreeing-but-clearly-wasn’t. Replaced by: a structured note template where the human does the first-pass tagging before any AI touches it.

Automation 2: Daily briefing digest (deleted at month 3)

Built to aggregate daily industry news into a five-point briefing each morning. Deleted because it was optimising for the wrong thing — it produced a briefing I was reading instead of doing work. The briefing was good; the habit it created was not. Replaced by: a weekly, manual scan of two specific sources. Less automated; more useful.

Automation 3: Proposal section generator (deleted at month 7)

Built to generate first-draft proposal sections from a structured brief. Deleted because the outputs, while plausible, were creating a new editing problem: cleaning up AI-generated text that sounded right but wasn’t quite right took longer than writing from scratch once the brief was clear. Replaced by: using AI for the structure (headings, argument skeleton) only, with human-written copy throughout.

Automation 4: Client follow-up tracker (deleted at month 11)

Built to track open client commitments and generate draft follow-up messages. Deleted because it was applying a uniform follow-up style to relationships that required differentiated handling. The automation was technically correct; the outputs were professionally wrong. Replaced by: nothing. Some follow-ups have to be written by hand.

The pattern across all four: Each automation was deleted not because it was broken but because it was optimising for efficiency in a place where efficiency was not the limiting constraint. The follow-up tracker was fast; the relationship required care. The briefing digest was comprehensive; the habit required selectivity. Build for the actual bottleneck, not the theoretical one.