I Set Up My AI Team. Now What? Give It One Routine With a Human Approver

I Set Up My AI Team. Now What? Give It One Routine With a Human Approver

This article is a companion to my video I Setup My AI Team. Now What?

Can't see the player? Watch the video on YouTube.

You connected the agents. You linked your tools. You asked a few questions in the chat box and got decent answers. Then the setup sat there waiting for you to think of the next thing to ask.

That is where a lot of us land after building an AI team, and it is where I found myself too. My answer is to give the team one routine: a trigger that starts the work, a draft the agents produce, and a person who decides what happens next. Once one routine works, you have something you can repeat, teach, and build on.

In the video, I walk through the routine I am setting up for my own content. When a new video goes up on my YouTube channel, the agents draft a blog post from it, and I review and approve it before anything is published. This post is part of testing that routine, so treat what follows as a working process, not a finished system with a track record.

Why the chat box is not enough

A chat box only works when you show up with a request. Every task starts with you remembering it, framing it, and checking the answer. That is useful, and it still leaves you as the bottleneck.

A routine changes who carries the work. The agents start when something happens, follow steps you already agreed on, and hand you a decision instead of a blank page. For a small team, that is a capacity question. The hours you spend re-prompting the same task every week are hours you are not spending with your people, your partners, or your mission.

What a successful routine looks like

The routine I show in the video has three parts:

  • Video posted. A new upload is the signal that work should begin.

  • Blog drafted. The agents turn the video into a draft post on my website and save it as a draft, not a live page.

  • You approve. A human reads the draft, fixes what needs fixing, and decides whether it gets published.

The third step is the point of the whole design. The agents take on the repetitive work. A person stays accountable for what goes out under my name.

Choose your trigger: event-based or time-based

Every routine needs something that starts it. In the video, I separate two kinds of triggers.

An event-based trigger starts the work when something happens, such as a new video, a form submission, or a new file in a folder. My video-to-blog routine is event-based.

A time-based trigger starts the work on a schedule, such as every Monday morning or the first day of each month. That fits recurring reviews, reports, and check-ins, where the work is due whether or not anything new happened.

If you are unsure which one you need, ask yourself what has to be true before the work makes sense. If the answer is "something new arrived," use an event. If the answer is "it's that time again," use a schedule.

Ask for a plan before you ask for a build

When I set this up, I did not start by telling the agent to build the routine. I started with a prompt asking for a plan. A plan shows you what the agent thinks the steps are, which tools it expects to use, and where it is guessing. That is the cheapest moment to catch a misunderstanding, before anything is connected or running.

I also dictate a lot of my answers when I brief agents. Talking through what I want makes briefing the agents faster.

Connect your tools, then check what the connection can actually do

Agents reach your other tools through MCP servers and APIs. In the video, I connect the tools this routine needs, including my website and blog builder.

A connection does not give an agent full access to a tool. Each MCP server or API exposes a specific set of actions. Your agent might be able to create a draft post but not upload an image, or read a list but not edit it. Before I trust a routine, I check the limits of every MCP and API it depends on, so I know which steps the agent can complete and which still need a person.

From a preparedness point of view, this is the same habit I would recommend for any plan. Know where the gaps are before you depend on the system, not after it fails in the middle of real work.

Automation still needs a person who keeps it running

A routine does not run itself forever. Tools change, connections break, and drafts sometimes miss the point. A resilient workflow pairs the automation with a named person whose job includes keeping the routine running: reviewing the output, noticing when something is off, and adjusting the instructions.

For most small organizations, that person is someone already on the team with a clear role as the routine's facilitator and approver. Write that role down. If the routine matters, someone should be accountable for it, and everyone should know who that is.

Make it reusable

Once a routine works, I ask Hermes, the agent I used for planning, to turn it into a skill. A skill captures the steps so the agent can repeat the routine, and so you can adapt it for a similar task later without starting over.

In the video, I also show a bonus step: giving agents their own email inboxes with Agent Mail. That helps when a routine needs to send or receive messages, but it is not where I would start.

Your decision this week

If you already have an AI team set up, decide whether you will keep using it only through the chat box or give it one routine with a human approval step. I recommend the routine, starting small.

Here is a bounded way to begin:

  • Pick one task that repeats at least every week or every month.

  • Decide whether it starts on an event or on a schedule.

  • Ask your agent for a written plan before it builds anything.

  • List the tools the routine touches, and check what each connection can and cannot do.

  • Name the person who approves the output before it goes anywhere.

Run that one routine a few times before you add a second. If you want more background on how the pieces fit together, read my post on building an AI-powered operating system and my earlier roundup of digital solutions for streamlined operations. For the setup that comes before this video, watch FuseBase Work + Hermes: 3 Steps to Unlock Your AI Team.

If you want help setting up a routine like this for your organization, Next Workflow works with purpose-driven organizations on their technology hub, automations, and AI agents for Less workarounds. More Mission.

I'm sending so much love, hope, and intentional action to you.

Related Posts

All Aces, Inc. helps purpose-driven leaders close the gap between what their organizations stand for and what their systems deliver, using the Integrated Resilience Framework.

NextWorkflow, an All Aces brand, helps purpose-driven agencies, firms, and nonprofits take routine work off their plates so they can deliver on what they stand for. It starts with a workflow analysis, then puts the right tools and services in place, from websites to AI assistants and online workspaces.

Next Leadership Development, a sister nonprofit, and its Black Resilience Network accelerate health, wealth, and cultural resilience in Black communities before, during, and after disasters.

All three were founded by Dr. Atyia Martin, Boston's first Chief Resilience Officer.