There's a phrase that's been floating around product circles for decades. "Don't ship your org chart." It's generally credited to Steven Sinofsky, the former Microsoft executive now at a16z. The idea is simple: when products end up shaped by the internal structure of the company that built them rather than the problems of the people using them, the customer pays for it.

You've seen this. The Samsung Galaxy S7 shipped with two email apps, two music services, two photo apps, and two messaging apps. Not only the company, but the entire industry org chart was painfully apparent. Ravi Mehta, former CPO at Tinder, called out Zoom's settings page as "the very definition of shipping the org chart. Somewhere, two teams probably couldn't agree, putting customers in the position of making decisions the organization failed to." The customer should never be privy to internal politics, but when "org charts" are shipped, how products get made takes priority over the problems customers want solved.

Many agentic AI companies are falling for a similar trap: They outwardly promote specialized sub-agents as part of their product experience. Or, enable their customers to create "agents" for every action they want automated. Our industry is now at a point that we're "shipping our agentic org chart" and losing sight of the customer problem.

Every agentic AI company is built on a stack of specialized agents. A research agent, a planning agent, an execution agent, a budget agent, a pacing agent. The architecture reflects how engineers decompose problems into AI systems that actually work. They generally do this to ensure accuracy, specialized agents get really good at a specific aspect of the customer experience and context window optimization. A collection of these specialized agents with a master agent is called multi-agent orchestration, which is a way of building agentic AI products that almost everyone has adopted.

There is nothing in principle wrong with this, the mistake — in my opinion — is when that architecture becomes the product. The result is that you've handed the customer your agentic org chart and called it a feature.

In agentic media buying, think about how a media manager actually works. Before AI, no one sat down and said, "let me invoke my budgeting skill, now let me call my pacing skill, now let me consult my creative rotation skill." That's not how people work. A media manager opens a campaign, looks at what's happening, and makes decisions. The goal is the best outcome for the client. The cognitive mechanics underneath are invisible. They don't need names and they definitely don't need their own interfaces.

When you surface the agent layer, you force customers to think like your engineering team. You've replaced a job that required one focused person with a coordination problem, and it creates a suboptimal customer experience.

At Gigi, we made a deliberate call on this early. Our customers don't need to know what's happening under the hood. When Gigi builds a campaign, manages a budget, or monitors pacing, there is orchestration happening. Sub-agents get invoked. Decisions get made. None of that is the customer's problem. Their problem is the outcome.

We also made a language decision that we think matters. We ask our customers to assign "tasks" to Gigi. (note: it's nice to see that Claude has adopted this as well).

Task assignment, in our view, rather than sub-agent creation, is a more elegant and apt descriptor of our approach to assigning work to AI. Marc Andreessen said it plainly on Lenny's Podcast in January: "The job is not actually the atomic unit of what happens in the workplace. The atomic unit of what happens in the workplace is the task. A job is a bundle of tasks." That framing is exactly right. When you assign Gigi a task, you hand off a unit of work. Gigi is on budgeting. Gigi is on pacing. The work is accounted for. You move on.

Tasks are also how AI capability actually progresses. The work best suited for AI today is the low-level, repeatable execution layer. As the agent proves itself, the scope of what you can hand off expands and you graduate from execution work to strategic work.

If you build around agents instead of tasks, that graduation never happens cleanly. You're always managing the roster instead of managing the outcome. The org chart becomes the product, and the customer is stuck running it. Exposing more agents gets in the way of a clean product experience. The best AI products will be the ones where customers barely notice how it works, only that it does.

Note: This was originally published as part of our Cherry Picked newsletter. Subscribe here ➡️ https://cherry-picked-gigi.beehiiv.com/subscribe