My AI Coworkers Are Already Here — They Just Need a Conference Room

The conversation around AI is stuck in the wrong frame. People keep arguing which model is best — ChatGPT or Claude, Gemini or Perplexity, OpenAI or Anthropic. The question misses the point. The model isn’t the product. The team is.

Think about how real work gets done. You don’t hire one person who’s mediocre at everything. You build a team. A strategist. An engineer. A researcher. An analyst. A skeptic who pokes holes. A specialist who handles the sensitive stuff in a back room. Each one is brought in for what they’re best at, and the manager — the human — sits at the head of the table running the meeting.

That’s the AI setup I want, and the pieces already exist.

Here’s how I picture a working session. I open my workspace and there are seats at the table. I’m in one. Claude is in another, because Claude writes the cleanest code. ChatGPT is across from me, handling broad reasoning and synthesis. Perplexity has the research chair, sourcing live citations. Gemini is in the corner running multimodal tasks. When the conversation needs a contrarian, I pull in DeepSeek for a second opinion. When the data is sensitive, a local model takes that part offline. We discuss the problem out loud. Each one contributes what it’s strongest at. They critique each other. I steer.

The product comes out of that meeting — code, a strategy doc, a workflow, a customer email — and we test it together. If it breaks, we go back around the table. Iterate. Resubmit. No single model has to be the smartest one in the room, because the room itself is smart.

The technical pieces of this are not hypothetical. Multi-agent orchestration frameworks exist. Memory layers exist. Routing systems that send the right query to the right model exist. What hasn’t been built well yet is the interface — the conference room itself, with persistent context, role assignments, live workflow awareness, and the ability to bring models in and out the way you’d add a colleague to a Teams call.

This is also where enterprise software is about to feel its age. The ERP, CRM, and project management tools most companies live inside were designed for static workflows: rigid menus, forms, screen after screen of configuration. After spending a few hours redesigning systems conversationally with an AI — “move this, rebuild that, connect this API, change that interface” — going back to traditional software feels like switching from a smartphone back to a flip phone. Salesforce spent two decades trying to become the operating layer for the modern business. AI-native environments may simply skip the rebuild and start over.

The winners in this next wave probably won’t be whoever ships the smartest single model. The winners will be whoever builds the best room for the models to work in. That means the best orchestration, the best memory, the best integration with the tools businesses already use, the best operational visibility — knowing which model handled what, what it cost, and whether the answer was any good — and the best collaboration design, so the human at the head of the table can actually run the meeting instead of fighting the software.

There’s a real cost question buried in here, and it’s the part most demos skip past. Running five frontier models in parallel for every task would get expensive fast. But you don’t run them all the time. You route. The cheap, fast model handles the first pass. The specialist gets pulled in only when the problem warrants it. The local model takes anything that shouldn’t leave the building. Done well, the cost curve looks less like burning API tokens and more like staffing a team — you pay for the expert when you need the expert, and the rest of the time the lights are off.

What I’m describing isn’t science fiction. It’s an integration problem. The models are here. The APIs are open. The orchestration patterns are published. The missing piece is someone building the workspace that ties them together in a way that feels less like a developer tool and more like a place where work actually happens.

When that workspace lands, the question stops being “which AI do I use.” It becomes “who’s at the table today.”