We run ordinary office tasks through AI tools and write down what happened.
Meeting notes, invoices, spreadsheets, the policy nobody has time to read. Each entry takes one such task, runs it for real, and keeps the record: the exact request, the output, the mistakes, and the sheet we’d use next time.
Written for offices of a few people with no IT department. This isn’t product news, and no vendor pays to appear here.
- The taskOne job a small office really has, and the file or text we used for it. Test material is public or made by us, never a client’s.
- The setupWhich tool, which version or plan, which settings, on which date. Enough for you to repeat it.
- The runThe request as we typed it, word for word, and each follow-up in order.
- The resultWhat came back, checked against an answer we worked out by hand first.
- What went wrongEvery miss, counted. Left-out items, invented details, wrong sums. This part is never skipped.
From the notebook
The notebook is being restarted. The articles that stood here before were summaries of AI product announcements; they dated within weeks and weren’t based on our own runs, so we took them down. Entries will appear below one at a time, each when its run is finished and written up.
Three shelves
Every entry goes on one of them.
- Run logsOne task, run from start to finish, with the misses counted.
- TemplatesRequest sheets, checklists and log tables you can copy, each with a note on how we tried it.
- Setup and safetyAccount and data settings read from the vendor’s own documentation, and what not to paste into a chat box.
What we don’t do
- Report results we didn’t get. If a run wasn’t done, there is no entry.
- Rank tools or tell you which one to buy. An entry shows one run on one day.
- Promise time or money saved. Where we timed something, the entry gives the timings and how few they are.
- Hide how the pages are written. We draft with the help of AI tools. Every article is reviewed by Dlrow Editorial before it is published.