5 BI staffers who put AI agents to work share what worked and what went wrong

I downloaded Instinct two weeks ago and chose a very simple task to start with: Remind me to call my mother (she worries).

My whoa moment: Instinct jumped to my aid when I told it I needed a new torque wrench for my bike. After a simple prompt, Instinct came back with four options: a cheap one, a mid-priced option, and two higher-end versions — all with links and a pro/con list.

In the two weeks I’ve been using it, Instinct has created an itinerary for a weekend trip I’m taking next year, booked tickets for my daily commute into London, and secured me a refund on an $80 subscription I forgot to cancel after the free trial.

My yikes moment: This past weekend, I took a trip to Oxford, one of Britain’s most notoriously complex places to drive. Oxford is famed as a warren of one-way streets, traffic cameras, and tolls.

I asked my Instinct agent to pay the city’s congestion charge after I drove in. Instinct obliged, cross-checking my car’s model against the different emissions tiers for the charging zone, and sending me a confirmation that it had paid for two days at £4 per day.

As I touted how easy it all was to my AI-skeptical partner, she pointed out that the cost was £5, not £4. So I checked, and it turned out Instinct got it wrong — it had paid a similar but separate charge for a smaller area of the city. The charge was nonrefundable, so I was forced to swallow both my own pride and an £8 tax for trusting the robots. Lesson learned, even if Instinct tried to find a silver lining.

Some people use their AI agents to save money. I decided to do the opposite and ask it to tell me how to spend my money.

My whoa moment: I’m a K-pop fan, and my favorite stars — Jeonghan and Joshua, from the group Seventeen — opened preorders for their new album “Dreamscape” on Wednesday morning.

I didn’t have the time or energy to parse through all the listings for “preorder bonus” gifts; things like limited-run photocards and polaroids that go to fans who place their orders early. Collecting limited edition K-pop merchandise can be a bit of a blood sport, so I wanted a solid strategy.

I asked Instinct to create a “buy plan” to maximize my fan experience. It took three minutes to surface a link to exclusive signed copies available for early orders on the band’s US website. The agent also gave me a road map for purchase quantities and the optimal time to make those orders.

Instinct saved me two hours with one prompt. Without it, I would have missed the window to purchase the US-only autographed albums.

I did receive a bit of judgment from the agent — it asked me not once but twice if I’d set a cut-off budget for my buy plan. I had not, and did not intend to set one!

I haven’t plucked up the courage to give the agent my credit card details yet, though.

My yikes moment: I was less impressed during my first 24 hours of using Instinct, when I added it to my Notion dashboard and asked it to do an overhaul. It ended up clearing out my goals calendar and schedule, and replacing it with a new — but blank — template.

Instinct’s WhatsApp-based chat feature is intuitive and consumer-friendly. But for optimal results, you’ve got to be precise in your prompts, and lay out exactly and unambiguously what you want the agent to do.

The other day, I asked Instinct whether tickets were available for a 6:30 p.m. showing of “Spider-Man: Brand New Day” at a nearby theater.

My whoa moment: We were on our way when Instinct texted that there wasn’t a 6:30. It was 6:15, “which starts in about 8 minutes.” Thank goodness for trailers — and for a speedy driver (which wasn’t me). Instinct said online sales had ended, so “your only shot” was the box office. It also started checking backup films.

My yikes moment: It suggested a 6:50 “Spider-Man” at another theater, but said tickets “were glitching,” and gave me the theater’s number. Obviously, because no one died, I wasn’t going to call. The theater’s site showed no “Spider-Man” that night. I told Instinct, and 16 minutes later, it said the movie “likely got pulled back.” By then, we had our tickets at the original theater. “Enjoy the movie!” Instinct said.

My whoa moment: Most of the tasks I wanted it to do for me were related to the morass of messages and paperwork from my kids’ school and activities. However, it got stuck on things like logging into the clunky school site app or being unable to check out on the Girl Scouts website. But finally, a breakthrough — I had put off the task of printing out and signing a form for after-school care. I asked Muse to fill it out for me from my email. I sent it a photo of my signature on a Post-It and used a photo of my daughter’s handwritten name (she was also supposed to sign it) as her signature (morally questionable, sure). It successfully “signed” it and emailed it back to the school.

My yikes moment: It offered to RSVP “no” to an invite in my email that happened to be on the same day as the Meta Connect conference. It suggested I could reply to the person and explain that this was “the most important event on my beat” — not totally wrong per se, but certainly weird to say in an email, and quite flattering to Meta.

My whoa moment: I had one simple dream: outsource the stupidly fiddly work of finding books and placing library holds to an AI. I mapped Muse to my iPhone’s Action button and wired everything up by granting Muse access to my Libby account and my accounts at my two local libraries.

Now, when I ask Muse to look for a book, it checks Libby for a digital version and sends it to my Kindle if it’s available immediately. If not, it compares the wait times on Libby and at my libraries, then places a hold on the copy that will get to me faster. Pressing one button to delegate a small slice of library bureaucracy feels slightly godlike.

My yikes moment: Amazon has blocked Muse, so I discovered that I can’t connect it to Goodreads, the place that I keep track of all the books I want to read, and which Amazon owns. It was a reminder that a personal agent is only as useful as the platforms that allow it in. The minute this stuff becomes genuinely convenient, somebody puts up an anti-bot wall.