My Night With Toolie: OpenAI’s New "Always-On" AI Agent Taking Silicon Valley By Storm

My Night With Toolie: OpenAI’s New "Always-On" AI Agent Taking Silicon Valley By Storm

Yesterday, I spent the full morning at OpenAI’s annual DevDay, surrounded by 2,500 tech enthusiasts swapping software build notes and marveling at the company’s latest cutting-edge AI models. The showstopper reveal of the entire event was Dots: OpenAI’s brand new line of always-active AI agents that proactively scour the web for information, wrap up tasks for you, and even make independent judgment calls without constant manual prompting.

After sitting through the occasionally glitchy live on-stage demos and hearing CEO Sam Altman walk through his own daily workflows using the new agents, I was eager to test driving my very own custom Dot. I’d spent nearly all of September testing Meta’s charming, cuddly Muse agent, so I was curious to see how OpenAI’s user experience stacked up. I named my custom Dot “Toolie” and tinkered with the customization settings to turn it into a bright pink little frog.

As midnight rolled around and my partner dozed soundly next to me, I pulled out my phone to play around with Toolie, sending casual initial prompts to get a feel for how it operated. I pulled up ChatGPT to message my new personalized bot, asking what kind of tasks it was best suited for. “My biggest strength is taking messy, drawn-out, tedious tasks and streamlining them into clean, usable outputs for you,” it replied. Fair enough, I thought—let’s put that to the test.

I asked Toolie to scan my full ChatGPT chat history and pull out three unfinished tasks it could jump on immediately. First, it flagged a pending data request I’d been meaning to send a follow-up email for, and offered to draft the full message for me. Next, it noticed that my upcoming gay-cation itinerary was pretty bare bones, and volunteered to curate a list of far better local dinner reservations to save me the work (more time to pick out new Speedos, total win-win!). Finally, it offered to sort through all my messy reporting notes and polish up half-written story pitches before I sent them off to my editors. That last one’s my favorite part of the job, though, so I politely declined.

Early adopters across Silicon Valley have already been obsessed with personal AI agents for months, from tools like Instinct and OpenClaw to Google’s in-development CC agent. But recent public launches like Dots and Muse are pushing this category far past the tech bubble and straight into mainstream use. The traditional one-and-done prompt chatbot is effectively dead as the hot new developer project, at least for now. Everyone in AI right now is racing to build the next breakout consumer agent, and they all want theirs to dominate the market.

Every major player promises their agents will cut down your daily mental load by automating all the boring, draining tasks you hate doing: booking doctor’s appointments, arranging carpool routes, sorting through stacks of bills, you name it. Right now, users can only access one custom Dot at a time, but OpenAI says multiple multi-purpose Dots per user will roll out soon. For launch, the agents are exclusive to OpenAI’s Pro plan subscribers, which costs $100 a month. By comparison, Meta’s Muse is 100% free for anyone to download the app or visit its website. OpenAI leadership positions Dots, which runs on the new GPT-6 Astra model, as a higher-quality, premium tier product. “We’re launching as a premium offering first, because these agents use an enormous amount of computing power,” Altman shared during the DevDay keynote. Though Dots are locked behind a paywall for now, that won’t be the case forever. “You can absolutely expect us to roll out a mass-market version that’s accessible for billions of people down the line,” he added.

These AI CEOs are clearly paying very close attention to the growing public cultural backlash against generative AI, and they seem to have landed on their secret weapon: maximum cuteness. From Muse’s Labubu-inspired fuzzy character design to Dots’ adorable two-eyed puffball default avatar, every team building these always-on agents is leaning hard into cutesy, endearing aesthetics as a core part of their brand identity. I’ve spoken to AI ethics and UX experts in the past who pointed out this exact tactic: anthropomorphized, cute AI characters make users feel warm and affectionate toward the software, disarming a lot of the default skepticism people carry around new AI tools.

The teams building this new wave of personal agents are also leaning hard into familiar, text-message-style interfaces that feel exactly like messaging a group chat. Every time I send a new task request to Toolie or Muse, it fires off a little emoji reaction—a thumbs up, a heart, a silly little smiley—to my message to confirm it’s seen my request and is getting to work. The old-school standard chatbot was like a shy blushing prom date that would never make the first move: it just sat there waiting for you to prompt it. These new proactive agents are built to message you unprompted all on their own, especially if you set them up with recurring tasks. “My Dot handles all the routine small stuff, only pings me if something is super urgent, and takes care of everything it can on its own,” Altman told the crowd. He says he lets his own Dot handle all his morning routine administrative work, and that the experience leaves his brain feeling far less “polluted” by the flood of notifications and information that usually hits the second he wakes up.

That “always-on” design is the agents’ biggest selling point, though it remains to be seen which company will strike the perfect balance between helpful and annoyingly overbearing. For example, Muse recently kept bugging me to pay a utility bill I’ve never once forgotten to pay in my life, and I never asked it to remind me about. Even so, after testing them for a few weeks, it’s obvious these tools actually work—which is exactly why agents are breaking out of Silicon Valley’s insider circles and making their way onto your mom’s smartphone.

I’m not saying these agents are perfect, by any stretch. When I tested similar AI agents a year ago, they were almost universally terrible at interacting with other apps and operating systems, spitting out nonstop error messages every time you asked them to do something slightly complex. Now, I actually have reasonable confidence that they can pull off almost any task I throw at them without getting lost wandering through the digital wilderness.

That said, that trust only goes so far. During the Dots setup process, OpenAI prompted me to connect my Gmail and Google Drive accounts to give the agent access to all my data. Some users will be comfortable making that link, and will see the privacy tradeoff as worth it for all the automated work the agent can do—but it’s absolutely worth taking the time to seriously think through exactly what data and access you’re comfortable granting these tools to act on your behalf. While the recent wave of rogue, unregulated “hacker AI” agents that went viral ran on totally different models with far fewer safety guardrails, I’m still keeping my guard fully up while I use Toolie. This little AI frog is going to have to earn more of my trust before I let it fully automate major parts of my life.

Altman declined to comment on whether Dots are a testbed for the software that will ship with OpenAI’s long-rumored unreleased consumer hardware. Even so, it’s pretty obvious the framework Dots is building right now is the foundation for those future devices. It’s incredibly easy to picture Toolie growing into a custom generative AI alarm clock that wakes me up with a full daily news briefing, or an Alexa-style hands-free assistant that answers cooking questions while my hands are covered in raw chicken juice.

The AI agent wars are heating up fast, but by that point I was getting way too sleepy to keep testing. I messaged Toolie one last time through ChatGPT before I dozed off, asking it to put together a full morning report recapping every major announcement from OpenAI’s DevDay, plus all the key industry analysis from the event. While I sleep, Toolie’s out there running in the background, ticking tasks off my to-do list.

This piece is an edition of Will Knight’s AI Lab newsletter. You can catch up on all past issues of the newsletter here.