AI has a teamwork problem
We gave AI to individuals, but most work happens in teams. Andrew Filev introduces Hops and explains why shared context changes how a company works.

By Andrew Filev, founder and CEO of Hops
In 1987, the economist Robert Solow wrote that you could see the computer age everywhere but in the productivity statistics. Companies had bought PCs by the million and put one on almost every desk, and the economy wasn't moving noticeably faster.
I think we're living through the rerun. Almost everyone I know uses AI every day, and plenty of them can finish a task in minutes that used to take an afternoon. Yet when I ask founders and executives how much faster their company moves than it did a year ago, the answer is usually 10 to 20 percent. That's well short of 2x, and of what the individual gains would suggest.
The usual diagnosis is that the models aren't good enough yet, or that people haven't learned to use them. I think that's the wrong diagnosis. The models are remarkable. The problem is that we gave AI to individuals, and most of the work that matters happens in teams.
We're in the sneakernet era of AI
Before office networks, people moved files between computers by copying them to a floppy disk and walking it over to a colleague's desk. Engineers called it sneakernet. It worked, in the sense that the file arrived.

Look at how AI work moves inside a company today. Someone gets a great analysis out of a chat session, copies it into a doc, pastes it into a message and explains the context again to the person who picks it up. That person opens their own session and starts from zero. The AI that helped with step one knows nothing about step two, and every handoff goes through a human clipboard. (I've done plenty of that walking myself.)
The PC era didn't pay off when the PCs got faster. It paid off when they were networked and companies rebuilt how work moved between people. I believe AI will follow the same path.
Collective intelligence was always the point
I've spent the last 20 years building software to help teams work together, and the lesson I keep relearning is that a good team is smarter than any person on it. One person knows the customer, another knows the numbers and a third remembers why we stopped doing it the old way. The value comes from connecting those pieces.
Personal AI works against that. Each person's assistant learns what that person tells it, and the knowledge stays in their sessions. Ten people with ten private assistants means ten separate sets of context, so the team gets faster individuals and the same slow handoffs between them.
The question I care about is what happens when AI joins the team instead of sitting beside each person on it.
Why we rebuilt from the ground up
The obvious answer is to add a bot to the collaboration tools we already use. We looked at that seriously, and we quickly realized it only takes you so far.
Once an AI works for a team and not for one person, hard questions show up. Whose permissions does it act on when three people ask it things in the same thread? What can it see, and how do you make sure it doesn't share something confidential with the wrong person? Can it work inside the actual document while two teammates edit it, or does it hand back a block of chat text someone has to paste in? Does it get better as the team uses it, or does every conversation start cold?
You can't answer those well from outside an existing product. The answers live in the data model: how access works, how documents are structured, how a change by a person and a change by the AI show up in the same file. That's why existing collaboration vendors struggled for years to offer useful AI agents.
So we rebuilt chats, docs, tables, slides and tasks from the ground up, with people and AI working in them together, in real time, under the same access rules.
That's Hops™. Your team and AI agents work in the same chats, docs, tables and slides, in real time, building on each other's work.

How the company brain compounds
The part that surprised me most, once we ran our own company in Hops, was how quickly the work started to compound.
Jen leads marketing for us, and for a long time every ad, email and landing page came to her for review. She spent her days fixing copy when her time was needed for strategy. So she showed Hops our marketing voice, and that went well past word rules: which message leads, what each audience cares about, which examples to use. Now anyone on the team can ask Hops to draft or check copy and get some of her judgment along with it, without waiting on her.
When David, our designer, created our mascot, he showed Hops how it should be used. Now anyone can ask Hops to add the mascot to a one-pager, and it comes out right.
What one person teaches Hops becomes something the whole team can do, and each job the team hands over makes the next one faster. That's what I mean by a company brain: the team's accumulated knowledge and expertise, available to everyone who needs it, with the right permissions.

From workspace to operating system
Once your team and its AI work in one place, the next steps follow. Connections bring in the systems you already run, from your CRM and drive to 2,000+ other apps, so Hops works with live data and nobody has to paste it in. Automations hand Hops the recurring jobs. Apps let a team build custom tools for itself, such as a tracker, a dashboard or an intake form, on the same data.
Take a customer renewal. Hops pulls the account's history from the CRM, past conversations and the contract, and drafts a renewal brief in a shared doc. The account manager reviews it, and the team saves the approach as a skill. From then on, an automation prepares a brief for every renewal two weeks before it's due, and a small app shows the team which accounts need attention.
People stay in charge of that process. Hops gathers, drafts and checks, and the team makes the calls.
Put those pieces together and the workspace stops being a place where you talk about the work and becomes the place where the work runs. Knowledge, data, recurring jobs and custom tools sit in one system that the people and the AI share. I think of that as an operating system for the business in the broadest sense.
The obvious objection
I can hear the pushback. "Andrew, the big AI labs will add team features to their assistants."
They tried, but that's hard to bolt onto a personal assistant, for the same reason networking was hard to add to a standalone PC. Permissions, shared documents, live editing, approvals and a memory the whole team builds are properties of the environment the team works in.
Besides, different models are good at different things: one is the best programmer, another writes better prose, a third generates the videos for our socials, and a fourth does routine work for a fraction of the cost. An assistant from one lab gives you that lab's models. Hops routes each job to the model that fits it, so your team gets the best results for its budget. That matters most when AI use grows from a few people to the whole company, and the bill grows with it.
The punchline
The PC paid off when we connected it to the network. AI will pay off when we connect it to the team.
We've spent the last few years making AI smarter. Now it's time to make teams smarter. Hops is open to everyone today at hops.ai, so bring your team.
Author
Andrew Filev
Founder & CEO, Hops
Andrew Filev is the founder and CEO of Hops, where people and AI agents work together in the same chats, docs, tables and slides. He previously founded Wrike, the collaborative work management platform he led for 17 years as it grew to more than 1,000 employees and 20,000 customers before its $2.25B acquisition in 2021. He also founded Zencoder, an AI coding agent platform. His writing on productivity and the future of work has appeared in Forbes, Entrepreneur, Inc. and The New York Times.