An assistant working at my desk
Should you try Dot, or any other assistant? Absolutely, try it out. Managing personal life, from ordering groceries, requesting refunds, cancelling subscriptions, researching trips, staying on top of emails I might have missed and more, has been wonderful.
It’s an assistant working at my desk. I can text or message Dot and have it use computer use on my laptop to really do whatever I want.
Dot, Duet, Muse and Orca
Of the assistants, I haven’t actually used anything except Dot, OpenAI’s ChatGPT agent. From what I can gather, Dot is great for computer use, Duet for coding and development work, and Muse (the Muse from Meta) for a consumer-level personal assistant. Dot is the only one I’ve had working at my desk.
Orca is an IDE focused on agentic development. It has a great CLI that makes it work well with Dot. Orca runs Claude Code, but it’s agnostic; it could run Codex sessions. I use Dot and Orca every day.
Setting up the assistant at my desk
For the assistant at my desk, the setup is what matters.
I have two computers. One computer acts as an Orca server and the other as a client. This allows me to maintain my personal laptop as the server and manage sessions on that. I pair the two through the Remote Orca Servers settings, which are still in beta: the laptop advertises itself as a server and generates an access link, and the other machine adds it. That access link lets the other computer run commands on the laptop, so treat it like a password.
The issue I currently have with Dot is that it can only use one computer at a time. So my personal laptop (my beefiest one) runs Amphetamine with Power Protect so that I have an always active machine. An assistant can’t work at a desk that’s asleep.
I’ve considered getting a Mac mini to do this, and likely will purchase a Mac mini for my wife.
The assistant at my desk did Halloween
Every Halloween we have a barbecue on the driveway. We cook brats, pork belly and more as trick-or-treaters come by. We watch a movie and have a good time.
For this Halloween I had Dot put together a Costco same-day cart based on a simple prompt I entered. It did so, and gave me the price. Before pulling the trigger, I recalled Uber Eats had a promotion for Costco delivery, so I had it cross-check it. Uber Eats was about $15 cheaper, so we ordered there. I never had to check either cart or check out manually.
Unfortunately the shopper was pretty bad, so I needed to try and claim a refund. I told Dot to do it. It did so, sent pictures to Uber Eats, and we received a credit back.
This was pretty convincing.
Daily updates from the assistant at my desk
Having Dot give me daily updates for my work and personal projects, based on Orca, GitHub, email, Slack and more, has been wonderful.
I maintain a private, hosted ChatGPT project site that has the project stuff on it, which Dot updates, and it sends me a daily message. The project site is a great hack. It allows me to stay on top of things and answer questions in a nice, easy-to-review, five-minute context.
I no longer wonder if I’m missing things.
How far I trust the assistant at my desk
Dot confirms before buying. I haven’t given it hard financial access either. It does have email and text permissions, and it can send and archive. I sandbox my coding agents, too.
Why I want more than one assistant at my desk
I use Dot for managing personal and work right now, and I really wish I could have more than one Dot so I could nicely separate context. I wouldn’t even need separate VMs or connections, but having Dot do both things seems to affect its context and memory negatively.
It just seems to forget some things, especially around how to access different stuff. It’s not as much a problem with Dot as it is a problem with me wanting to operate in separate contexts.
I’d like to be able to keep those separate, though early indications from providers (like grokbot) seem to indicate that it’s too expensive for them.
Next for the assistant at my desk
I’m going to have Dot file a protest against my property tax rate here soon.