Meet some of the agents that keep Vaul on track

As I write this, our Marketing Project Manager agent is letting me know that it’s struggling. A request in our marketing channel was only partly done the last time it picked it up, and our Agent Hub has already told me what went wrong. Checking in on our agents is how I start my day now that we’ve built several agents that are live, in production, helping us run our own operations . 

Vaul is a small, lean team, and everybody wears more than one hat. When we started building agents for ourselves, we filled the gaps where we needed more support: marketing, sales, internal operations and project management, plus the admin that comes with each of them. 

We thought it might be useful to share some of what we learned  throughout this project. Turns out, getting them to do useful work is only one small part of the journey. Knowing what to ask them to do, recognising when that work is incomplete, and having what both they and you need to put it right, is just as important.

Meet some of our agents

Our Marketing Project Manager keeps our marketing and editorial activity up to date, refreshes analyses, and circulates meeting agendas before our weekly catch-up, so every meeting starts with the latest  information.

Our Internal Operations Coordinator makes sure that what we agree to in a meeting happens. Most of our work gets decided in conversation with each other, and before agents we relied on each other to write up the outcomes and make progress visible to everyone else. Now our agent does it. It reads every transcribed meeting as it comes in, files what was agreed, moves tasks along, assigns tickets to the right people and updates comments and Slack channels, so everyone can see where the work stands without having to ask (or they can at least ask it directly and not worry about disturbing someone else).

Our Sales Process Agent keeps our sales pipeline up to date and researches the background of any incoming leads, so we can walk into a first conversation prepared.

Between them, they keep us on track. To the point that we’ve massively reduced any manual admin: the updates, the filing and the follow-ups that used to depend on one of us remembering now happen, which leaves us free to spend our time on the work we’re good at.

The structure we never had time to build

Before the agents, client work always came first, and our own internal  ways of working got whatever time was left over. Most of what we knew lived in people’s heads, available only if you asked the right person at the right time.

That is no longer the case. Our regular meetings are now set up for us, with current information circulated ahead of time, so we arrive knowing what we need to discuss. The agents flag decisions still waiting to be made, and the vague “we should think about this” items that would otherwise drift. They also hold us accountable: when an action has been agreed but nobody has picked it up, they say so, and ask who is taking it on and by when.

None of this is complicated. It’s the structure every team knows it should have and few small teams get round to building, because client work comes first. 

Keeping an eye on the agents

Part of my day now looks like managing a team, though a different kind of team. Every morning I open the Vaul Agent Hub and see which agents are working, which are healthy and which are struggling, what each did last and what it’s doing next. I know what each one can access and, as importantly, what it can’t. I can also see what each one has cost us so far this month.

However, work doesn’t always go to plan. Building agents within a custom harness has usually been a technical job, and when one broke, you went back to the technical team to find out what had happened. We’ve made that part easier for ourselves. The agent’s cards tell me in plain English what went wrong, and the “Fix it” button lets me put it right without needing to be technical. That makes my job much easier, and much faster.

What’s in a name

For months I talked about our “meeting agent”, and I’m not sure anybody at Vaul knew what that meant beyond a vague sense that our calls were being recorded. That’s changed since we’ve invited our agents into our Slack channels and have articulated a simplified role in how they help our teams. Now, in meetings, people say “the marketing agent will  handle that” knowing that the tasks we discussed will get captured and actioned. People talk to the agents in Slack just as they would a colleague. 

That being said, interacting with our agents doesn’t then make them ultimately accountable. Each of our agents has a named person at Vaul who owns it, answers for what it does, and is responsible for improving the agent in a timely manner. The agents may take on the admin, but  people still make the decisions.

What we learned doing it

None of this arrived fully formed. We’ve built, rebuilt, renamed and rethought along the way, and we’ve learned what makes an agent worth using and trusting.

We’re a small team, and some of this is easier for us than it would be across an organisation of several hundred people. The patterns we’ve run into, though, are the same ones we see in the organisations we work with, many of them much bigger than us. That’s what the series that follows is about.

It starts with why building the agent is the easy part. From there, we’ll explore  why people trust the agents they helped shape and what changes when “my agent” becomes “our agent”, why we start narrow and let agents earn more responsibility, when it makes sense to build your own, and finally what to measure when counting agents and tokens tells you little.

If you’re building your own agents, or wondering whether you should, I hope it saves you a few of the detours we took.

Next in the series: Building the agent is the easy part.