[AI SPRINT] AI Workers Are Here. Most Businesses Aren't Ready.

Dots were introduced at OpenAI's DevDay on Tuesday. Each one gets its own cloud computer, connects to more than 4,000 apps, and can be reached through ChatGPT, Slack, Teams or a phone call.

Four days earlier, Microsoft announced it was rebuilding Copilot around a mode called Autopilot, which runs tasks in the cloud while you're away. Meta launched Muse on September 8, an agent that sends email, books travel and makes purchases from its own virtual machine. And Manus now sells agents that start work on their own when a lead emails or a calendar event fires, with a companion product called Cue that gives each agent its own email address, phone number and wallet.

It’s clear we have now entered the automated personal-assistant AI era rather than just talking about it as we have the last few years.

So how does it work?

After I set up our ChatGPT workspace to enable Dots (an admin has to switch it on, and it's only available on Pro, Business Premium and Enterprise plans today), I debated which task to give it. I started by giving it a name (Ollie) and choosing an image to represent it. I saw the phone icon and decided to start there: pressing it to give Ollie a voice call.

Ollie answered after a few rings. After some introductions, it suggested looking through my email to find the commitments I had coming up. I agreed, and we talked through how to track them, settling on Notion to record each so it had a database to track against. Then it got to work, reviewing my email, my calendar and even documents on my computer. It set up a task to review them each morning and either suggest how it could handle each one or remind me what I needed to finish. While it worked on that, I gave it several tasks in parallel: writing an email to my team, sharing a link to a recent GitHub update on new platform user experience changes, and prepping me for a morning phone call.

What struck me most was how easy it was to shift into voice and talk to it like I would any team member. Except this one has access to all my data and systems, knows my style, and has the intelligence of every human put together. It was compelling enough to give me one more reason to switch back to ChatGPT from Claude, where I've spent most of the last year.

After testing Dots and reading everything I can about the rest, I keep landing in the same place: these are remarkable tools for an individual, and they will change the nature of work quickly. Much of what I’ve been sharing about the last three years about the future of work has suddenly arrived.

But here's the thing: for most established businesses, the autonomous worker part is still a ways off.

The reasons why are what I want to talk about this week, because I think they matter more than which agent is best. Check it out after this brief update.

What has to be true before an autonomous agent can help

Most of the coverage I've read this week compares the tools. Which agent connects to more apps, which is cheaper, which model is smarter. For most of you reading this, I suspect that's the wrong question, because every one of these tools assumes a business that mostly doesn't exist yet.

For an agent to do real work inside a company, work you can trust it to do again and again, a few things have to be in place:

• The work is written down. An agent can't take over a process that lives in one person's head.

• Your information is reachable. Project, process, and customer knowledge sits somewhere the AI can get to, reliably and safely, instead of spread across inboxes, spreadsheets and a CRM nobody updates.

• The agent has an accurate memory of how your business works. This is significantly harder than most people think.

• Someone owns the result and can check it quickly. If reviewing the agent's work takes longer than doing it yourself, the delegation has failed.

• There's governance. You know who approved the agent, what it can touch, and who gets the call when it does something wrong.

Now I'll be honest about where most companies are. In the businesses I work with, most staff are still learning the basics of ChatGPT, Claude or Copilot. Few have a central, organized, reliable place for company knowledge. Shared memory barely exists across a team, let alone across a project or a customer. Governance, where it exists, is mostly a policy document written when ChatGPT first came out.

That's a long road before anyone hands an agent the keys.

Two numbers from the companies we've surveyed in the last month make the point. Of those companies, 98% have AI applications, but only 7% have automated a single process. And across them, 78% of processes have no automation at all.

I'll put it plainly. I see two kinds of organizations that can use these tools right now: tech startups and entrepreneurs.

My guess is this isn’t your company.

The agents will show up anyway

Here's the part that worries me. These agents are ready for individuals right now. A founder, a solo advisor, or a manager who's comfortable with technology can turn one on today, connect their work email and calendar, and have it running ten minutes later.

That's how autonomous agents will get into your company: through your people, on their own initiative, before anyone decides anything.

If we thought we had problems with shadow IT, or even shadow AI, those will be nothing compared to shadow AI workers.

And the details matter here. OpenAI says a Dot builds memories from everything it's connected to, and disconnecting an app later doesn't erase what it learned. Only deleting the Dot does that.

Microsoft hasn't confirmed what admin controls Autopilot will have as it enters private preview, but I'd expect the same memory problem. That means your company data may become the memory of someone else's agent, and you won't be able to get it back.

To be fair, these companies tell us the tools have safeguards, and I'm sure they do.

But they haven't shown much of a track record lately of safeguards that actually work, or even of knowing which ones they need. We already have had plenty of "rogue agents" before these tools gained steam.

And that's not all

There's a competitive risk to autonomous agents, too. Your next competitor might be three people with no legacy systems, running exactly these tools today. They don't have twenty years of process to untangle, terabytes of disorganized data in SharePoint, or a host of disconnected enterprise systems few people understand.

That means they can build around agents from day one, with better efficiency, lower costs and often higher service quality, all supported by a fleet of AI workers.

The good news for an established business is that any readiness work pays off no matter which agent wins. Writing down how work gets done, putting company knowledge in one reachable place, naming an owner for each workflow: all of that helps your people today, with the AI they already have.

It's also what lets your team serve customers who used to cost too much to reach. That's where I think the real growth from AI comes from.

What to do this week

I'm speaking at a security operations summit soon, and AI security and governance are top of mind. Below are three minimum standards for agents I'll be talking through. They're simple enough to start on now.

1. Find out what's already running, and decide whether you want it. Ask your team, without judgment, which AI tools and agents they've connected to work email, calendars or files. Then review each one: what it can access, what it can do on its own, and whether you'd approve it today. You can't govern what you can't see, and I suspect the list is longer than you'd guess.

2. Pause new agents until you have a policy. Tell employees not to connect new agents to company systems until you've defined the rules, then write them. Aim to have that policy in place within 30 days. A short pause now costs far less than unwinding access later.

3. Name an Agent Lead. Give one person responsibility for your agent workforce: keeping a registry of every agent and what it can touch, preventing duplicate agents doing the same job, and taking on the more advanced capabilities as you're ready for them.

It's still early days for AI, and it's moving fast. In 2027 I expect agents will be ready for most businesses, and the companies that did this work will switch them on. Everyone else will still be cleaning up spreadsheets.

Hit reply and tell me: has anyone on your team already connected a personal agent to work systems? I'm curious how many of you find out this week.

Catch up on these and other happenings at the October 23 AI SPRINT session at noon Pacific. I will cover what landed, what to do about it as you plan your November sprint, and take open questions. Open to everyone.

Trent Gillespie is CEO of Stellis AI and a keynote speaker helping business leaders understand and operationalize AI in their companies. He spent almost nine years leading global innovation efforts at Amazon before leaving to help other companies build the capabilities they need to compete. Book Trent to speak to your group or book a call to discuss using AI within your business.

Did someone forward this newsletter to you? If you're not already signed up, you can subscribe to AI SPRINT™ for free here.