Your collaboration tools were built for a team with no AI on it

Look at the tools your team collaborates in every day. The @mention. The comment thread. The approval chain. The shared doc with nine cursors blinking in it. The Monday standup where everyone says "no blockers" (but you know at least two people are definitely lying).

Every one of them was designed around the same assumption: everyone on the team is a real-life, human person.

That assumption held up for decades. Then AI arrived, and most of us did the obvious thing. We bolted it on. One person, one AI, one chat window, and a lot of copy-paste in between. Each of us got faster on our own, and the team around us barely changed. Optimizely's agentic experimentation report found the same pattern across 900 companies: AI's impact is stuck at the individual level.

Collaboration is only difficult when you don't have the right tools for it, and the tools we have were never built for this brand-new type of team.

Collaboration feels hard because the tools were built for a different team

Software has spent decades getting very good at one thing: helping people work with other people. Comments, mentions, permissions, version history. All of it assumes a human is on the other end of every handoff.

That mismatch explains something odd about how companies talk about AI and headcount. The usual story is that AI makes each person more productive, so you need fewer of them. There's a second logic hiding underneath: every extra person adds coordination costs. More handoffs, more meetings, more "just circling back." If collaboration is friction, fewer collaborators starts to look like efficiency.

Change the starting point and the math changes with it. Marketing teams have more work than people. Good marketers are scarce, and every new hire needs real time to get up to speed. The work that never gets done isn't waiting on a smaller team, but waiting on capacity nobody could hire.

Which is why the "AI takes jobs" framing misses the point. The shift is real, but what it opens up is bigger than what it replaces.

Who's on the marketing team now then?

Human-agent collaboration comes in more shapes than people expect. Two marketers editing the same brief. A marketer handing a first draft to an agent. A workflow where one agent pulls the data, another checks it against brand guidelines, and a person approves it before anything ships. A Virtual Teammate working through a goal you gave it at the start of the quarter.

All of that is collaboration — and agents don't all play the same role, just like people don't.

Here's what that looks like on an ordinary work task. Lauren Hammarstedt, who runs product for Optimizely Agent Platform, was building the roadmap deck for an internal Ask Me Anything call. Mark, Optimizely's AI, pulled information from a bunch of different places to put the first version together. In the old flow, she'd then tag Alex Whitney on the slides that needed his input and ask what tools were landing by the end of the quarter.

This time, the work stayed in one place. One shared artifact, one shared thread. A Virtual Teammate in product marketing takes the first pass, Alex adds his part, someone reviews it, and nobody has to stitch together outputs from five private chat windows.

First things first: Agents need to know your goals

Goals come before shared skills, shared context, or anything else you give an agent. Without them, every interaction starts with you. You open the window, explain the task, make the request. Which means the work you never get around to asking for never happens.

That's the gap most people feel today. The AI is capable. It's just waiting on you to start.

Think about how any new hire works. Every role has a fixed core, the things that have to happen no matter what, plus everything that lands on the desk unannounced. An agent should be able to carry both: anchored to the work that matters, flexible enough to handle the chaos.

Once goals are in the picture, the relationship flips. You stop being the person who has to ask for everything.

Now, for step two: Agents need to know your values

Goals tell an agent what to work on. Values tell it how to behave while it's working. They're different from goals, and different from brand guidelines too.

Same task, opposite right answer, depending on the company. Optimizely builds in public. We published our own marketing AI benchmark in the open, and a marketing agent working here should know that. A more discreet company would want the exact opposite.

Values can be personal, too. When Shafqat first set up his own Chief of Staff Virtual Teammate, he told it up front that they were operating with a wartime mindset: prioritize ruthlessly, and treat everything outside the top priority as noise.

Then there's the smaller stuff: the colleague who knows everything, like which topics to steer around in a meeting, which jokes land, and what happened last time. That's company history, personality, and judgment rolled into one sentence, and it's exactly the stuff an agent or Virtual Teammate has to pick up.

It has to feel natural, or nobody will stick with it

All of this falls apart if working with agents feels like filling in forms. The bar is a colleague who's paying attention.

Getting to stage that Lauren is talking about here takes more than a smart model. Memory, personality, speed, and the experience itself all have to work together — otherwise the whole thing turns into a robotic, monotonous chore. And who wants that?

Less busywork, more of the good kind of collaboration

Think about how much of the old way of working went into everything around the decision. Pulling context from five places. Copy-pasting across tabs. Filing a ticket. Waiting. Rebuilding the same analysis into a slide deck. All of it before anyone could say what to do next.

Agents (and their connectors) take on that cognitive load. People get to the decision faster, and spend their time on the parts of collaboration that need them: judgment, debate, and the final call.

And here's the part that surprises people: more AI can mean more collaboration, not less. When an agent handles the follow-ups and the first drafts, the humans on the team have more room to actually talk to each other.

No job disappeared in that story. Nobody had been hired to do that work in the first place. It just stopped eating weekends, late nights, and family time.

What to do with all of this

If your team is still collaborating like it's all humans, start here:

The team changed, so the tools should too

This is what we're building with Optimizely's Agent Platform: shared org and brand context, agents and Virtual Teammates working alongside your people, workflows where a person approves at the moments that matter, and shared artifacts teams can work on together.

Collaboration got a reputation for being slow and painful, mostly because the tools were built for a team that didn't have AI on it. Whilst yours definitely does now.