Reviewing AI tools on launch day tells you what the demo does. Six months of real work tells you whether anyone still opens it. We tracked twenty-three tools across a working team; five are still in use.
How we tested
Each tool was assigned to real tasks, not benchmarks, and usage was logged weekly. A tool was recorded as abandoned when nobody had opened it for three consecutive weeks. No vendor was told they were being tested, and none of the accounts were complimentary.
What survived, and why
The survivors had one thing in common: they removed a step from an existing workflow rather than adding a new place to work. Transcription with speaker labels, a research assistant that cites sources, an image cleanup tool, a code review helper, and a meeting summariser that writes into the existing notes system.
What did not survive
Anything that required people to change where they worked. Standalone chat interfaces were used enthusiastically for two weeks and then quietly replaced by whichever assistant was already inside the software people had open. Tools with per-message pricing were abandoned earliest, because nobody wanted to think about the cost of asking a question.
The pattern worth stealing
Before you buy anything, ask which existing step it deletes. If the answer is “none, but it enables something new”, expect it to be abandoned. That is not cynicism — it is what six months of usage logs showed, repeatedly.