AI Agent Teams & Multi-Agent Orchestration

Most work isn't a single task, it's a workflow.

Research leads to analysis. Analysis leads to decisions. Decisions lead to action. And somewhere in between, context gets lost, handoffs break down, and work stops flowing.

We just launched Agent Teams on agent.ai to make that better.

This is the first step towards teams of agents that pass context to each other, build on each other's work, and handle entire workflows from start to finish.

So today, I want to break down:

• What Agent Teams are (and how they work)

• Why multi-agent systems are the future

• The poll results from last week (they're not what I expected)

—@dharmesh

Note: Agent Teams are for agent.ai Pro members ($25/mo), but you can always sign up to agent.ai and use your new credits to try the agents out individually before going Pro.

What Agent Teams Are

Here's a problem I've had for years: meeting prep (I don’t hate meetings — but I do hate unproductive ones).

We all know we should research the company, profile the person, and prepare talking points. But it's a ton of work per meeting. So most of it gets skipped, rushed, or reconstructed from memory afterward.

We built Meeting Intelligence Team to handle this workflow.

Four agents connect to each other: one researches the company, another builds a prospect profile, a third delivers a prep brief to my inbox the morning of, and the fourth generates the follow-up email and CRM notes from the transcript.

We also built Sales Prospecting Team around the same idea. Four agents that connect list building, research, qualification, and outreach -- so the output of one becomes the input of the next.

It’s just the start -- and we have many more agent teams coming soon -- but what’s most unique about what we’re building is that these aren't just collections of agents. They're agents that connect, share context, and build on each other's work.

No single agent can handle an entire workflow alone. But a team of agents, each handling a step and passing the baton, can.

This is the start of something bigger (and something I’ve been raving about for some time now): multi-agent orchestration.

Why Multi-Agent Systems Are the Future

I've been talking about this for a while now -- both in this newsletter and on LinkedIn -- but it's worth repeating: multi-agent networks are the future.

Single-purpose agents are great, and we've made incredible progress with them. But the future of agents lies in their ability to discover and collaborate with other agents to accomplish a goal.

The vision I keep coming back to: Agents will declare their capabilities. They'll specify their inputs and outputs. They'll build "experience" over time. And in the same way that LLMs support tool calling today, agents will have access to networks of other agents, invoking them to handle tasks they can't do themselves.

For example, this is already happening in the software developer space.

I've been a professional software developer for 30+ years. Throughout that time, we've worked at higher and higher levels of abstraction. But AI coding agents might represent the most significant leap we've seen.

Instead of writing individual functions and classes, developers are orchestrating fleets of agents. We're becoming architects and conductors rather than hands-on implementers.

The best developers will be those who can effectively communicate goals to AI systems, review generated code, and know when human insight trumps AI efficiency.

Last Week's Poll Results (They Surprised Me)

Last week, I asked everyone who reads this newsletter: Have you #UninstalledChatGPT and moved to Claude?

The results:

• 35% said "Yes, I've moved to Claude"

• 35% said "Not yet, but I'm considering it"

• 30% said "No, I'm sticking with ChatGPT"

What surprised me wasn't that 35% moved -- that tracks with the download and uninstall data.

What surprised me was that another 35% are considering it.

That's 70% of respondents who are either gone or thinking about leaving.

This is a real shift in sentiment from ChatGPT to Claude in a very short period of time.

Will it stick? I don't know. But it's a data point worth watching, and since last week, it doesn’t seem to have slowed down.

Competition is good. It keeps companies sharp. And right now, the competition is fiercer than it's ever been.

—Dharmesh (@dharmesh)