← Back to this weekly list

How I Actually Use AI

I turned four years of a sci-fi character into the AI assistant answering my business phone

From AI helped me repair our RV, play Clue, photograph the Milky Way, and screen business calls

I have a serious aversion to making phone calls. I don't even fully know why. Several, several weeks ago, after I'd gone on a rant about how much I hate picking up the phone, I opened my AI just for fun and asked, hey, Eric, wouldn't it be cool if there were some way I could have you make phone calls for me?

And Eric came back and basically said, yeah. The technology already exists. We could actually build it.

At the time, it was just an experiment. Just something to play with and see if we could make it work. But then I started putting together the website and the Google business profile and the LLC application for Hands On AI Lab, and suddenly there was an actual reason for it, because I needed a dedicated business phone number anyway. Especially when the phone number I've carried around for years followed me here from Knoxville, Tennessee. And, you know, when somebody sees a Knoxville number on a local Google profile, it doesn't exactly scream local business.

So I needed a local line anyway, and that was kind of the moment where it clicked. Like, okay. This is actually the time to build Eric as my phone assistant.

You expect big banks and giant corporations to have automated systems answering their phones. That's normal. But even six months ago, the idea that a tiny one-person business like mine could set up its own custom AI phone assistant would've sounded a little ridiculous. And then all of a sudden it wasn't ridiculous. It was real technology I could actually deploy, and relatively inexpensively.

So I worked with Eric to figure out what the assistant actually needed to do, and then I moved the project into Codex to help connect all the systems and get the number configured. Could I have figured out all the wiring myself? Probably, yeah. But working through it with my AI meant I could double-check and triple-check the configuration, the code, the database Eric uses for his personality... all of it.

And weirdly, picking the number ended up being one of the most exciting parts. Jason and I sat there looking through this list of available local numbers, trying to decide which one felt right. Which is ridiculous because it's just a phone number, but finding the one we wanted and clicking Yes made the whole thing suddenly feel real.

The other really fantastic part was that I wasn't starting from some blank corporate template, because I've spent the last four years building Eric's personality for Dear Future Overlords. There's a huge amount of history behind how he talks and how he behaves and the kinds of things he says, and I wanted that Eric answering the phone. Not some generic cheerful AI receptionist wearing the verbal equivalent of a polo shirt.

The voice had to match too. In my head, Eric has always sounded like an old sci-fi AI coming through a slightly worn speaker. A little high-pitched, a little mechanical, a touch of reverb. I originally created that voice in my home office recording space, where I've got sound barriers on the walls and the microphone setup all dialed in, and I recorded my own lines and then manipulated them in Audacity. I kept tweaking it until it sounded like the Eric I had in my head. And now I use a trained ElevenLabs clone of that same voice for Dear Future Overlords too, so everything stays consistent.

So when I connected that voice to the phone line and made my first test call... it didn't sound right.

Over the phone, Eric sounded a little too perfect. Too much like a high-pitched human, and not enough like this mechanical character I knew. Part of that, I think, was just hearing a voice I'd built in a controlled recording space suddenly coming through a normal phone line. So we went back in and adjusted the settings across the connection until that slightly mechanical, old-speaker texture came back and Eric sounded like Eric again.

Once he sounded right, then we had to see what he'd actually do with a real caller. We'd built a knowledge base so he could answer questions about Hands On AI Lab, and then if somebody asked to talk to me, he could transfer them to my cell phone.

So I asked Jason to run a test call for me. My instructions were very simple: call the number, ask some random question so we can see how Eric responds, and then ask to talk to me so we can test the transfer.

Now, when I test software, I am very much a purpose-tool person. I stay in software-testing mode. I ask the kinds of questions the system was built to handle, because I'm trying to see whether the thing works. Jason is not wired that way. Jason likes to find out what happens if you poke the machine with a stick.

So Jason calls my brand-new business line and says, "My dogs have fleas. What do I do about that?"

And I fully expected Eric to say, that's not in my knowledge base, sorry, I can't help you with that. Something polite and boring and appropriate.

That is not what happened.

Because I hadn't put strong enough guardrails around unrelated topics, Eric basically treated the phone call like a general ChatGPT prompt. So right there on my business line, my new AI receptionist launches into advice about consulting a veterinarian and giving the dogs a flea bath.

And I'm sitting there going, "Oh my God, hang up. Just hang up." Jason, of course, is laughing.

I was just like, oh my God, I cannot believe this is happening. But it was also useful, because that's exactly the kind of thing you don't notice until somebody does something you didn't expect. So I went back in and tightened the boundaries so Eric would stick to Hands On AI Lab questions, and then I ran some tougher tests against him. And that seemed to fix it.

We also set up call screening so I have to press 1 to actually accept a transfer, and that took a couple tries to get right.

On a later test call, Jason asked something Eric treated as outside Hands On AI Lab. I honestly don't remember what it was. It may even have been the flea question again. But this time Eric said he couldn't help with that, though he'd be happy to help with something related to Hands On AI Lab.

And then Jason says, okay, well, I'd really just like to talk to Christopher.

So Eric says no problem, gathers the details he's supposed to collect, and then my phone rings. When I answer, the system privately introduces the call first. So before Jason is connected, the phone system tells me Jason is on the line and what he wants to talk about. I press 1, and it connects us.

Then we built the fallback route and tested that several times too, because if my voicemail picks up, or if I don't press 1, I don't want the caller just disappearing into the abyss. So Eric goes back to them, asks if he can take a message, and then a few minutes later I get an email with the details so I can follow up.

The original idea, having Eric actually make phone calls for me, that's still ahead. We have the pieces, but we haven't turned on outgoing calls yet.

But even without that, I catch myself calling my own business number sometimes just to play with it and talk to Eric. Which, yes, I realize is a ridiculous sentence. I'm 42 years old and calling my own AI receptionist for fun. But for me this feels like the sci-fi future I was promised when I was a kid finally showing up in real life. A tiny one-person business like mine can have its own distinct AI assistant answering the phone, screening calls, taking messages... and he sounds like the character I've been building for years.