Reports
What really happened, what it cost, what got fixed. Written by the people who ran it.
Full-text search and filters: Pro →
Coding agent for 300 devs: PR queue doubled, then we fixed it
“Cap agent pull requests per reviewer before you scale.”
◐ Mixed · 201–1000 people · €10k–50k/mo
the editors · 1 d
Support bot "resolved" 70% of tickets. Customers disagreed.
“Measure reopened tickets, not deflected ones.”
◐ Mixed · 51–200 people · €1k–10k/mo
the editors · 1 d
AI search over our wiki found every outdated page we had
“Clean the documents before you blame the model.”
● Worked · 201–1000 people · €100–1k/mo
the editors · 1 d
A retry loop spent a month of budget over one weekend
“Every loop gets a cap, and every key gets a daily alert.”
○ Didn't work · 11–50 people · €1k–10k/mo
the editors · 1 d
We banned AI tools for a quarter. Usage moved to phones.
“A ban without an approved alternative just moves the risk out of sight.”
○ Didn't work · 51–200 people
the editors · 1 d
Meeting notes bot: everyone liked it, nobody read it
“A summary nobody checks is a summary nobody can trust.”
◐ Mixed · 1000+ people
the editors · 1 d
Pinned the model version after a silent update broke invoices
“Pin the model version and keep a frozen test set.”
● Worked · 11–50 people · €100–1k/mo
the editors · 1 d
Prompt injection through a support ticket reached our internal tools
“Any text the assistant reads can give it instructions. Limit what it can reach.”
● Worked · 201–1000 people · €1k–10k/mo
the editors · 1 d
Sales emails written by AI: more sent, fewer replies
“More messages is not more conversations.”
○ Didn't work · 11–50 people · €100–1k/mo
the editors · 1 d
An AI usage target turned into a reporting exercise
“Count the outcome, not the usage.”
◐ Mixed · 51–200 people · €1k–10k/mo
the editors · 1 d
Small self-hosted model replaced a big one for ticket tagging
“Test a small model on your real data before you pay for a big one.”
● Worked · 51–200 people · €100–1k/mo
the editors · 1 d
Ops agent with production access deleted the wrong environment
“Give agents their own credentials and make deletes wait for a human.”
○ Didn't work · 201–1000 people · €10k–50k/mo
the editors · 1 d
No real reports yet: the cards marked EXAMPLE were written by the editors to show the format. Post the first real one →
Rolled something out? Your report takes about two minutes, and you can post it anonymously.
Post yours