I Built a Shadow Team

The feeling that our content meetings were spinning in place wouldn't leave me for a while.

We had agenda items but no pitches. Everyone was busy with their own work, and the meeting usually turned into a status report. Planning is something that only emerges when someone takes the time to follow the news, the developments, and the trends — but each of us already had too much on our plates to ask for that time and effort. So did I. Feeding people the information wasn't easy, and just hammering at ideas wasn't the answer either.

The Standup Failed

When I first took over the team, the card I played early last year was the standup. The way my old startup colleagues used to gather briefly each morning to share what they did yesterday, what they'd do today, and what was blocking them. Fifteen minutes at most. I wanted to build a bin where ideas could quietly accumulate.

For the first few weeks I carried it, throwing out a trend article in advance and bringing a competitor's campaign to ask, "what about this?" Everyone got reasonably used to it, and good ideas and story-worthy material piled up. But it rarely connected to actual action. They could see the material was good, yet taking it on as their own and driving it forward with appetite was no easy thing. And the moment we loosened our grip even slightly, the promise of the standup faded. Having dragged it along for about a year and a half, another problem was that everyone had grown too used to it.

The reason wasn't simple. Transplanting a startup's way of working straight into a large-company team is not easy. A startup standup is a tool for people standing on the same line every day to synchronize where things are. We were each absorbed in different tasks, and our work didn't overlap enough to share daily. Above all, pitches require psychological safety. Float a half-formed idea once and get a "you think that'll work?" in return, and your mouth stays shut from then on. Even when I tried to lift people's resolve, reporting and decisions took time, and it's easy to hear "why pick up new work instead of doing what we already do." Changing the format doesn't change the culture.

That isn't to say the team lacked ability. Quite the opposite. They're all experienced and they know the trade. It's just that the daily work binds a person to the present and steeps them in inertia. There was, structurally, no room to look into what a competitor did yesterday, or which words society is reacting to right now.

The Problem Was the Structure, Not the People

Once I'd laid it out this far, the direction became clear. This wasn't a matter of pushing people harder. It struck me that the absence of pitches might not be a problem of will but of time and information. The same goes for polish. Rather than ask someone to take a piece from zero to eighty in order to push it past ninety, maybe the method was to hand them an eighty and ask them to make it a ninety.

What if the research were done for them and set on their desk? If competitor moves, the week's hot topics, and worthwhile overseas cases were organized in advance, the meeting could start not from "what do we look for" but from "what do we make." It's about lowering the barrier to pitching.

There was one more thing nagging at me: the inefficiency of a structure where one or two people hold the plan and execution runs with an agency. A lot leaks in the gap between the picture in the planner's head and the handoff to the agency, and re-aligning eats time. The process of spreading a single piece across web, video, and social was improvised every time.

My thinking converged on one point. Rather than hire more people, I'd add one more team — an invisible one.

The Shadow Team, Virtual Colleagues

In software there's something called Conway's Law. It says that an organization's communication structure becomes the structure of what it ships. A fragmented organization ships fragmented results. Read in reverse, it also means: shape the organization the way you want the output to look.

So I designed a virtual team in the shape of the output. A shadow content team built from AI agents.

• Lead — assists the team-lead role. Sets the big picture of what to publish, when, and on which channel, and distributes the work to the other roles.
• Planner — surfaces items and drafts plans to the content series specs.
• Researcher — gathers competitors, trends, and data, and sets them out before the meeting.
• Writer — drafts copy to the series format.
• Editor / Checker — checks brand voice, prohibited expressions, and facts.
• Multichannel — derives one core piece into social, newsletter, and video scripts.
• Performance Monitor — collects metrics after publishing and feeds them back into the next plan.

This isn't about replacing people — it's closer to setting a direction so the team can work the way I have in mind. The planner who used to shoulder research, writing, derivation, and checking alone now works as if seven colleagues sat alongside. It's closer to adding arms and legs to someone who has the ideas.

Running It for a Quarter

Theory alone won't do. I needed training material.

Last year's anniversary black-and-white photo exhibition was a good textbook. It wasn't a one-off show but a campaign that branched out from internal communications all the way to online rollout. I trained the shadow team on its plan, channel-by-channel rollout, and responses. In effect, I had it learn the grammar of how one core piece of content branches into many channels.

This year I applied that grammar to an art-festa-style campaign. The result was clearly different from last year. The internal notices, the web exhibition build, and the social rollout didn't drift apart — they branched from a single stem. Where we were dropping the ball became visible early in the campaign. Improvisation shrank and structure settled in.

There was a side effect too. As the agents began taking on pieces of the work, the team spent less effort on content structure and scheduling. Our reliance on the agency dropped noticeably. Because now we could run the picture in our heads through one internal loop before handing it outside. I also had it learn everything about our existing newsroom — then analyzed the problems, drafted improvements, and firmly set where we want to head.

What the AI Couldn't Do Mattered More

Honestly, this shadow team isn't perfect. The reports redo analysis we've already done, and they seem to keep up with the latest while stopping short of real insight.

Decisively, it doesn't know what's happening inside the company. It can't do real reporting. Going to the scene, meeting people, reading expressions, asking the question that follows the question — that's the human's part. The subtle currents inside the company, who's sensitive to what, which message must not go out right now — the agents don't know any of it. In the end there's no choice but to work alongside people, and it has to pass through human judgment.

So human-in-the-loop can't be anything but the starting point of the design rather than a compromise. AI handles drafts and analysis; people judge and take responsibility. The thinking about improvement and direction has to keep coming from people. The moment that boundary blurs, the system becomes dangerous. Agents lying plausibly happens more often than you'd think.

Cutting agency dependence isn't purely good either. Once you can solve everything in-house, the reasons to bring in an outside view dwindle. The unfamiliar perspective an external partner used to bring, the advice you couldn't have reached from the inside — that's hard to replace with the word "efficiency." Nor can we even digest all ten proposals the AI churns out.

What Does a Team Lead Actually Do

These days the meetings are changing a little. Before we start, we read the weekly report first. What went well last week and what got buried, what would be good to publish next week — the shadow team offered its opinion. The meeting is turning into a place to add judgment on top of that, add ideas, and weigh reality. It's not finished, but a way of working where people and AI move in step is slowly settling in.

Handing the lead role to an agent left me with a strange question. What does a team lead actually do? If much of dividing the work, setting the schedule, and checking quality can be done by a virtual colleague, what remains in the human lead's seat?

My tentative answer is this: not deciding what to make, but deciding what not to make — and taking responsibility for that decision. To keep learning, keep building structure, and meet change as it comes. Not easy, but a pleasure.