We Built an AI Teammate We Can Phone and Delegate To

Last week my phone rang between two meetings, on my actual number. It was a member of my team, telling me what had come in overnight and asking if I wanted anything handled. There was no person on the other end. The call was from Hermes, an AI teammate we built for ourselves at Dabster Labs. It calls me, I call it, and we talk it through, out loud, in Greek.

Status, up front and honest: Hermes is our own internal AI teammate, not a product you can buy. It is live on our own systems, it does real work every day, and it now calls our real phone number and holds a proper conversation. We are not calling it finished, but the parts described here work today. We built it on ourselves first, on purpose.

What it does every day, before we open the laptop

Hermes is not just something we phone. It runs a set of standing jobs on a schedule, the kind of morning research and admin that used to eat the start of a founder’s day. On a normal day it:

• reads our CRM and sends a digest of the open tickets and leads that need attention
• drafts the day’s LinkedIn post for us to review
• writes a short morning brief on what moved in AI and automation
• drafts outreach emails for a market we are working
• mines small-business pain points as a quiet signal feed for where we can help
• enriches our prospect list with names and verified contact details
• runs a weekly strategic brief and a quarterly competitor snapshot
• audits its own call quality, so it keeps itself honest

Every one of those used to be a slice of someone’s morning. Hermes runs them before we start, on a schedule, and messages us only what is worth reading. Dozens of these scheduled runs are already logged, and it has been running without a babysitter.

One real call, from ring to done

The standing jobs run on their own. The phone is for the things that come up in the moment. Here is one call, start to finish:

1. My phone rings, or I call Hermes. We talk, in Greek.
2. I ask what is open. It checks our live CRM and reads the open tickets and leads back to me. Real data from our real system, not a canned reply.
3. I tell it, out loud, to handle something real. On one call I asked it to build a client onboarding form for a profession we were targeting, put the right images in it, and get it ready to send.
4. I hang up. The heavy work does not happen on the call. Hermes hands the task to a fuller agent, which goes off and does it after we finish talking.
5. A little later, a message lands: the task is done, with a short status update.

In practice it works like phoning a colleague and handing off a job. It is not a person, and we are not pretending it is. It just behaves like one for the specific things we built it to do.

That is the part people underestimate. It does not just fetch an answer or draft a paragraph. It builds the actual thing. A custom onboarding form, the images that belong in it, ready to go out, and then it delivers it. We hand it a widening range of work this way, from research and drafts to finished forms and onboarding kits with their media. The rule of thumb we use is simple: if we can describe the outcome we want, it is usually worth handing to Hermes.

We built the cheap version first, then earned the upgrade

We did not start with a polished phone line. First we made Hermes answer on a voice channel that was already free to us, a clever hack with close to no cost, just to prove the idea was worth anything at all. It was rough: it talked one way at a time, you could not interrupt it, and the audio was not smooth. But it worked well enough to show the loop was real, so we kept going.

Once it had proved itself, we moved the calling onto a proper voice line. Now Hermes calls our actual phone number and the conversation is clean. That path is the whole point, and it is exactly the judgment we bring to a client’s budget: build the cheap, useful version first, prove it earns its keep, and pay for the polish only once it has. We did not buy the smooth phone line on day one for a tool nobody had validated. We earned our way to it, and we would steer a client the same way.

The point is not that it talks

The phone call is the part people remember, but it is not the point. Plenty of things can talk. What matters here is that a set of real jobs runs every morning without anyone starting them, and that you can hand extra work to software the way you hand it to a colleague, by voice, in the middle of your day, and it comes back done. The voice is the door. The work is the room.

What an agent like this could do for your business

To be clear, this next part is the shape of the idea for your business, not a result we have delivered for a client. It is the reason we built Hermes in the open. Picture a standing agent that, before you arrive, has already read your inbox and your CRM, drafted the replies that can be drafted, flagged the two things only you can decide, and lined up the day. Then, between meetings, you phone it, or it phones you, and you say handle this one, and it does. Same pattern as Hermes, pointed at your systems instead of ours.

Where it stands

• Status: our own internal AI teammate, live, doing real work daily, now calling our real phone number. Not for sale.
• What it runs: a schedule of standing jobs (CRM digest, morning brief, content drafts, outreach, prospect research and enrichment, weekly and quarterly briefs) plus voice delegation on demand.
• The path: we prototyped the calling on a free channel for close to nothing, then upgraded to a proper voice line once it had earned it. Cheap to prove, polished where it counts.
• Outcome: we are still putting a clean hours-saved number on it and we will not invent one. What is plainly true today: it does a morning’s worth of research, content, and lead work before we start, every day, and it takes spoken tasks and finishes them.

Hermes is part of a habit at Dabster Labs: we run our own business on the automations we sell, and we try them on ourselves first. It sits next to how we security test our own CRM and the digest that watches our servers every morning. If we would not run it, we would not sell it. You can see the rest of these own-build stories on our case studies page.

If you have a small team and too much to do, this is the pattern worth borrowing: build the cheap, useful version first, put it to work, and pay for polish only where it earns its keep. If you want an agent like this pointed at your own systems, tell us what you would hand off and we will show you the cheapest version that does the job.

Related service: AI Chatbots