Why Every Chief of Staff should be Following the Rise of AI Agents

(I wrote this article originally for Clara Ma and her excellent Ask a Chief of Staff community - you should get in their orbit!)

Spend time in Silicon Valley these days and the topic that everyone wants to talk about is AI Agents; a new kind of software built on large language models that can act autonomously, making unscripted decisions, in pursuit of a goal.

As Chiefs of Staff are often running point on AI initiatives, and writing their company’s AI policies, it makes sense for them to get ahead of the agentic capabilities of this new tech.

You can think about AI Agents as less “assistant,” and more “worker.”? While an AI Sales Assistant might help a sales rep with discrete tasks like sourcing leads and drafting emails, an AI Sales Agent can take the goal of Sell More Widgets, and run with it while using the context and guardrails that you set for it; chopping the goal up into tasks (lead sourcing, outreach, follow up, negotiation, etc.), and stringing all the tasks together in pursuit of the goal.

Frameworks and Definitions

One framework for what constitutes an AI Agent is the following;?

1) It perceives data and context

2) It interprets that data

3) It make decisions autonomously

4) It acts autonomously.??

An early pioneer of the current crop of AI Agents built on OpenAI also described to me an additional dynamic; the ability for AI Agents to autonomously chunk up a project into tasks, and then set about completing those tasks.?

One example of a production ready AI Agent are the Waymo self-driving cars that can be found on the streets of SF, Phoenix, and LA.? Waymos perceive data using sensors, they interpret that data to create a map of the surroundings, they make unscripted decisions based on those surroundings, and they act by driving autonomously.

So why should Chiefs of Staff be paying attention to AI Agents?? Beyond the obvious productivity gains for their orgs, this is the rare opportunity for the Chief of Staff to staff up so to speak; to command an AI team that can vastly increase their throughput and impact.

One dynamic of the Chief of Staff role is that it often falls on them to pick up big, strategic projects that have no home elsewhere in the org; the opening of a new office, the pricing rev, the financing, the M&A deal, etc.? As every Chief of Staff knows, this scale of project can be daunting without a solid line team to help execute.

Case Study: Opening Up a New Office

Take for example, the opening of a new office.? When I was the CEO of my previous company, we decided to open up a second North American office outside of our San Francisco hub.? I asked my CoS, Matt, to evaluate the top five options for us and make a recommendation, and then lead the execution of that recommendation.? My top criteria for making a decision was access to talent, but that wasn’t the only factor - cost, commutability from SF, time zone, and culture were all things I cared about.

My CoS went to work researching each of the criteria, developing a weighting system based on my criteria, making a recommendation (Phoenix!) and then designing a landing plan (hiring events, marketing events, community outreach). ? He got it done in 100 days, which was impressive for a team of one, with some dotted line help.

Now consider instead that he was “managing” an AI Agent.? Instead of doing the work, Matt would have clearly laid out the goal, laid out some guardrails, and asked the Agent to do the work for him.? Of course he would have final editing rights on the output, but the vast majority of the research, logistics, and other moving pieces could have been fully outsourced to a properly guided AI Agent.

The AI Agent would have probably gone about the project the same way that Matt did by starting with a project plan; chunking up the project into milestones, and would then start using the Web for research, designing an evaluation model, cranking out a rank ordered list of potential destinations, writing up a recommendation, designing a launch plan, and then executing (securing temporary housing, scheduling interviews and launch events, etc.).? But the AI Agent could have moved much, much faster.

(Let’s do this live!? Here is a quick example of me playing with an early AI Agent, AgentGPT, asking it to help me make a recommendation for a second office):

https://www.loom.com/share/0d05f5845f1644eca6f5c14b77acdfd6?sid=98721e71-9c95-4fb6-8486-ea762e286c21 )

Now run the same thought experiment for a Pricing Project, for designing and planning an offsite, for designing and rolling out a new communications policy, etc.? Imagine what a Chief of Staff could do if they could spin up virtual resources to do the heavy lifting on the special projects that come their way.

The Promise of AI Agents (Caution: It’s Still Early Days)

I consider the Waymo self-driving car to be one of the great human achievements.? As I sit in the back of these Agent powered cars, and watch them navigate construction cones, avoid pedestrians behaving erratically, operate in bad weather, etc, I marvel at the brain under the hood that seems to always make the right decision.?

Waymos can’t decide where you should go, but they can get you there just about flawlessly.

This is the promise of the AI Agent in respect to Chiefs of Staff; a tireless, all knowing resource that can intelligently and autonomously chunk projects into tasks, and start knocking them off.

It is still early days, and few Chiefs of Staff would trust mission critical projects to an AI Agent that might hallucinate, might not do exactly what it’s told, and may not have access to the domain knowledge (systems, history, context) to make perfect decisions.? But the foundational models are improving rapidly, as is the plumbing that will allow AI Agents to tap into the systems they need to do the job.

One nagging fear I have - the more I read about AI Agents, the less they seem to fit neatly into the "AI as a Copilot to Help Me Do My Job Better" metaphor. This feels like true worker replacement tech :/ Don't believe me? Ask your Uber driver how they feel about Waymo.

Ambient’s interest in AI Agents

For those of you familiar with Ambient, you know that our core value prop is to extract next steps / action items from various business channels (meetings, email, chat), and help you bundle those tasks, with access to reinforcing context, into AI powered project plans.

We believe that task lists will be an obvious jumping off point for AI Agents to hit mainstream adoption, and think that Ambient can help provide the guardrails (business context, transcripts, email history) to help guide AI Agents execute better.

Want to chat more about Ambient and AI Agents?? Feel free to schedule some time with me here.

Resources:

Here are a few helpful overviews of AI Agents:

Zapier:

https://zapier.com/blog/ai-agent/

Sierra (co-founder Bret Taylor is on OpenAI’s board and was former co CEO of Salesforce):

https://sierra.ai/blog/ai-agents-guide

MIT

https://www.technologyreview.com/2024/07/05/1094711/what-are-ai-agents/

Baby AGI

https://github.com/yoheinakajima/babyagi