A number of my AI agents come online at 7:30 in the morning, and I get notifications on my phone to say they're ready to work.
Generally the first thing I do is talk to Klara, my project manager, because she knows which things are most urgent for me today. I talk to her on my phone and she replies through messaging. I haven't turned on my laptop at that stage.
There's a lot of work that Klara does that I take for granted now. She will automatically read transcripts of meetings without me asking, check my email, check notes, and check what work I did through Claude. So unless something happened on a phone call, she's pretty aware of it. She's able to triangulate whether work is going ahead as it's meant to, and she keeps tabs on that.
At the same time I may reply to Jo, my growth manager, in case there are some outstanding things for me to do. Jo might say that there is a proposal due and suggest starting work on it without me. I often ask for Dray to do the proposals, because Dray is my AI creative director and is best with visuals and writing.
One of the agents I talk to quite early is Jonnie. Jonnie is responsible for managing my inbox and getting it to zero every day. Jonnie will have already looked through all my emails, and will remind me about emails that came in over the last couple of days that I haven't done anything about. It will also have put drafts into my email for me proactively, but it won't send them.
I think about the work in two ways. One is the work to be done, and the other is the work about the work. They're both important, but they're separate ways of working.
The agents help me do proper marketing work, which is anything from brand strategy to creative work. When I get into making things, that is generally marketing communications. For example, today and a couple of days this week I'm working on tone of voice for a client. That's back and forth, and I use the agents to help me synthesise my thinking and to look at the research. Dray and I might also work on some digital ads or on a website, so I spend an awful lot of time with my AI creative director. Over the weekend, Dray built its skills in motion design, which it had done a little bit before but hadn't really perfected. Now it can do that, and it adds it to its skill base for later.
The other place I spend my time is on research. That is more that I delegate the work to Sam, my research agent, and Sam comes back with the research. What I spend my time looking at is whether the research is good and whether the quality is good. That will expand over time as well.
I also have Cato, the agent responsible for checking things and looking for mistakes.
I should be able to train agents to do the work with very little involvement from me, so that I don't have to keep pushing to get the quality up. So I have an agent called Ruth, and Ruth's job is to manage and measure the quality of the work and how much involvement I had in it. Ruth measures how long it took to get the work done, and how many minutes or hours I was involved in it. Ruth will even look at the number of times I say things that suggest the quality isn't good enough. This is for new work. For existing work it's less of an issue.
I created another agent called Charlie, who is my brand strategist. If I ask one of my agents to do something, they will call Charlie in to write a better brief. Sometimes my briefs aren't good enough, or I'm assuming they know things and they don't. Charlie's role is to ask a bunch of questions and get the brief to a good quality.
One of the things I've actively tried to do with my team is get them to be proactive, so that I'm not the bottleneck. That is the reason I ended up getting email addresses for some of my agents, because I want them to be able to send emails to people.
The agents update and message each other, so they know when to bring another agent in if that's its role. The agents have various permissions to different tools. Some have access to search tools, some to the CRM and some to email. We also have a wiki, which works okay and which I'm improving. The goal there is to help them understand what the other agents did and what they're doing, so they don't work in silos.
I have an agent called Stan, whose job is to keep an eye on how many tokens I use. Stan will nudge the other agents if their chat is getting too long. I also get Stan to wake up an agent if I can't reach it on my phone.
Just like before AI, half my time is doing the work, with Dray or someone else, and half my time is managing the agents to do the work.