Leah MessengerBy the time you read your first morning to-do list, your Virtual Teammate has already been at it for hours.
That's the part people are underestimating. Virtual Teammates (true AI coworkers as opposed to what the rest of the internet has been talking about) don't wait for you to open a tab and think of something to ask. It's a named user with its own access rights, sitting in your team roster, picking up workflow steps, running on schedules and triggers and inbound events.
Yes, that means:
They ship work overnight and when you're not available (ooh)
They know everything about your brand and how you work (ah)
They collaborate with humans AND each other to keep work moving (nice)
SO, the question in week two isn't whether it's doing anything. Instead, it's whether what it's doing is actually any good.
A Virtual Teammate with a thin brief and a narrow view of yourself, your business, and your day-to-day tasks still runs, still delivers, still turns up in your inbox every morning at 7am. It's just a bit more generic than what it could be doing for you. Meanwhile, the SEO guy just one desk down from you gave his VT access to brand guidelines, approval chains, three months of campaign context, and Conductor (to name but a few things). The result? Work that lands, every time.
Same product. Wildly different outcomes. The variable is how you are collaborating with your virtual teammate.
Access to AI stopped being the bottleneck a while ago, how you're using it is what makes the difference. Longer-term vision and direction is necessary (otherwise you just get another version of that to-do list in your notepad). The great news? They're super easy to collaborate with.
So, here are 7 practices to work best with this entirely new category of agentic AI: Virtual Teammates.
The highest-leverage work happens in the first hour, before you've seen a single output.
When a human joins your team, you don't hand over a laptop and say "okay, write the blog, and I'll see ya tomorrow." You give them the brand guidelines, the style doc, the three pieces you're proudest of, the two you're embarrassed by, the words legal hates, and the name of whoever signs everything off. Then you let them shadow you for a bit to see what's up.
A good onboarding pack (filled to the brim with context) could look something like:
Brand and tone guidelines: The current one, not the 2019 PDF nobody's opened (plz)
3-5 exemplar pieces: Your best work, with a note on why it's your best work
A jargon blacklist: Every word your team has banned, and the reason*
Audience definitions: Who you're writing for and what they already believe
Your approval chain: Who signs off on what, and what never ships without a human
Your preferred output format: Bullets or prose, long or short, three headline options or one
That last one sounds trivial, but it isn't. Half of all "the AI output needs heavy editing" complaints are really "I never told it what good looks like" complaints. #facts
If you want something for scale: the average human hire takes 3-6 months to fully ramp. A Virtual Teammate is productive in its first week. Brief it like you mean it, front-load the context. It'll be the cheapest hour you'll spend all quarter, and provide the ultimate sidekick for all your current (human) employees.
*Please read this before you ban the em dash though
I know how this sounds, but stay with me.
The people who name their Virtual Teammate and configure its personality use it more. Consistently. A name doesn't make the model smarter — obviously — but it changes how you treat the thing. A named teammate gets corrected, developed, and maintained. An unnamed one gets tolerated.
(There's a reason why, like 86% of Optimizers have named their Virtual Teammates after their pets 🫶).
There's a whole behavioral point behind this: you won't invest in a working relationship you don't perceive as a relationship. It's the shortest route to the mental model that makes every other practice on this list work.
Set the personality properly too. You know, direct and terse if that's your team. Warmer and more conversational if that's you. No correct answer, just one you'd like to keep working with come month four.
And because it's a real user, this isn't just vibes. It shows up in your team roster and your workload views. It gets assigned CMP workflow steps like anyone else. Give it a name and a role and the org chart stops being a metaphor.
Each Virtual Teammate gets individually scoped permissions — configured in plain English, so you have full say over what it can and can't access, with every action logged against a named identity. Use all three parts of that.
Start narrow, and widen as your trust builds and builds (because, it will):
|
Start with |
Graduate to |
|---|---|
|
CMS draft rights |
Publish rights |
|
Querying ODP segments |
Touching customer identifiers |
|
Flagging experiment ideas |
Activating experiments |
|
Suggesting content updates |
Staging updates for human approval |
|
Read-only campaign data |
Editing campaign records |
Yes, this constrains day one. It also means month one isn't spent unpicking something that shouldn't have gone live. Teams who granted everything at once often got the opposite of what they wanted: broad access before established context produces more correction work. The teammate had permission to act before it had enough context to act well, and it acted — because being proactive is kinda its thing. Start with what's reversible, and always keep the human-in-the-loop vibe going.
There's also the element of accessibility outside of Optimizely too; tools like GA4, Fullstory, Contentsquare, and Conductor (to name a few). This reiterates the idea that Virtual Teammates meet you where you work, and when you want them.
Your Virtual Teammate is always working—with or without a prompt.. The work happens whether you're in the conversation or not, whether you're at your desk, or on a beach with a cocktail in hand (wish this was the case more often tbh).
Which means your job isn't switching VT's proactivity on. It's aiming it at a target. 🎯
Most people aim it at about 20% of what it could cover, because they only think to delegate the things they'd have asked for anyway. The teams getting the real benefits hand over standing responsibilities, not tasks:
That last line is doing more work than it looks like. Standing jobs fail when they generate too much noise, because realistically, we have too much noise in our lives already. It would only lead to you skimming them by week three and ignoring them by week five. Set the threshold for what earns an interruption, and set it high.
Then once again, you can widen the remit over time. The teammate that starts with a morning brief should be running competitive intel briefs and pre-meeting prep by month two. Scope creep is the goal here, not the risk.
Correct a chatbot and you get a better answer in that session. Then it's gone, and tomorrow you start from zero again. That loop is exactly why 76% of marketers think AI costs them three hours a week — because for most of the tools they've used, it does.
Virtual Teammates store corrections and edits in their memories. We call it self-learning.
The Translations Specialist turns every correction into a Skill, with your approval, so the same error doesn't recur. The Chief of Staff retains your feedback and brief preferences, so tomorrow's output is already ranked against what you care about. Personal memory stays private, team memory gets shared, and the whole thing thickens week by week.
So every correction is a deposit that earns interest.
Basically, access, context, and building the memory is the whole idea. Access without context gives you a very confident stranger. Context without memory gives you a decent first draft, forever. Memory is what turns it into a colleague who knows how you work.
Practically: correct hard in the first three weeks, and be annoyingly specific — it will be moooore than worth it in the long-run. Teams who corrected consistently early hit a completely different ceiling from teams who were too busy and let it slide.
A Virtual Teammate parked in a separate tab is a Virtual Teammate you'll forget by Thursday... which kinda takes away from the whole "teammate" thing.
The teams with the highest sustained usage embedded theirs into the places work already happens: instant messaging, canvases, tasks, email, meetings. Nobody has to remember to go and use it, because it's already in the room. That's also where real-time collaboration kicks in, between both themselves and us humans. That's right, humans and Virtual Teammates working the same problem simultaneously rather than passing it back and forth.
Every extra click between your team and the work is an adoption and efficiency risk. Remove the ones you can. The beauty of Virtual Teammates is that ease of collaboration and ultimately, adoption for the highest efficiency of marketing. 👑
Every irreversible action is staged for human approval. Every action carries an audit trail tied to a named identity. Offboarding works exactly like it does for a human user. That's the architecture, and it's the right one.
Your job is deciding where the line sits, upfront, in writing, somewhere your team can see it:
Stays with a human: publishing, campaign budget changes, experiment activation, anything customer-facing that can't be pulled back, anything legal would want to know about.
Doesn't need to: research and synthesis, first drafts, performance monitoring, meeting prep, competitive scans, tagging and metadata, internal summaries.
Settle this early and you never have the awkward meeting where someone asks who approved a thing and everyone looks at each other... only to find the newest Virtual Teammate's name documented. Autonomy and governance get designed together, or neither one works.
The anti-patterns are consistent enough to name:
Relate to it like a chatbot: You ask, it answers, you close the tab — while the proactive work you never configured sits unclaimed. You'll get chatbot value and conclude the category is overhyped.
Onboard with nothing: Then acting surprised when it doesn't sound like your brand.
Correct vaguely: "Make it better" is a vibe, not feedback.
Skip the permissions and guardrails customization: ...just because it feels like admin. Cheapest insurance you'll probably ever buy.
Leave the remit where you set it in week one: The scope should grow as the memory does.
They're the ones who learned to work with one properly, then scaled the habit and grew their VT's expertise and skill range.
None of it is complicated. Brief it well. Name it. Scope it carefully. Point the standing work at things that matter. Correct it like you mean it. Put it where you already are. Keep a human accountable.
They're already working (so go on, go make it count).