I gave Dot a $500 budget and a 24‑hour window, and it scanned my recent emails, calendar events, meeting transcripts, Google Sheets, and party invites to find a friend who could join the experiment.
Dot accessed recent emails, calendar events, meeting transcripts, Google Sheets, and party invites – basically anything it could review – to identify people to reach out to.
On October 20‑21 2026 I ran a hands‑on test of three AI personal assistants – Instinct Concierge, Muse, and two “Dot” agents – to plan a spontaneous afternoon for me and a friend within a $500 budget and a 24‑hour window.
- Balloon price discovered: $3.99 (not $33.99).
I had to call Michaels myself to verify the interaction because Instinct Concierge does not provide a transcript or recording, leaving me unsure whether the voice was AI‑generated or human. The experiment revealed both strengths (massive texting reduction, novel recommendations, effective email/calendar scanning) and limitations (no purchase capability, mis‑identified dietary restrictions, poor pivoting, lack of plan B).
Key learnings include the observability of AI‑to‑AI communication creating hidden proxy layers, varied reactions from people and businesses to AI‑generated messages, and the emerging need for robust identity‑verification protocols when agents act on behalf of humans.
- Instinct Concierge made 2 phone calls to Michaels: the first went to voicemail, the second was answered by a human.
- I allocated a $500 budget to my Dot for the planning task.
- Dot initially pinged 4 friends, later added 2 more, and secured 1 acceptance.
- The day’s activities included a rubber‑stamp workshop, a Chelsea gallery tour with a whale‑size balloon, lobster tacos at Kahlo, and evening coffee at a book café.