Last week we wrote about what happens to the calls you miss while you're on a job. The short version: a missed call is usually a lost job.
This post is the next part of that story: first a missed call, then an AI that answers it, and soon an assistant you can just talk to while you work. Here's how AI voice agents work, what's changing, and what a shop in Fort Myers or anywhere in Lee County can use now.
How most AI voice agents work today
Most AI phone assistants run in three steps, one after another:
- Speech to text. The caller's words get turned into a written transcript.
- The AI model. A text AI reads the transcript, decides what to say and can use tools, like checking your calendar.
- Text to speech. The written answer gets turned back into a voice.
OpenAI's developer guide calls this a "chained voice pipeline." The upside is control: you can read every step and check the answer against your rules. The downside is that each handoff adds delay, and the middle step only sees text.
Block's research team puts it bluntly on its Dessa page: current voice AI "turns your voice into text and throws away an incredible amount of information text alone can't carry." A frustrated sigh doesn't survive a transcript.
Speech to speech, and why "full duplex" matters
The newer approach skips the transcript. A speech-to-speech model takes audio in and sends audio out, so it can hear tone and answer faster. ElevenLabs launched two of these on October 20, 2026, saying each one "is a complete voice agent." In tests ElevenLabs reports from Artificial Analysis, its larger model started speaking in 0.57 seconds on average.
Full duplex goes one step further. It means the AI listens and talks at the same time, the way people do on the phone. You can cut in, and it can say "got it" while you're still talking. OpenAI describes its GPT-Live model as able to "listen and speak at the same time, a capability called full duplex," handing the heavier work to a separate system so "users can keep talking while backend work runs."
The demo that caught our eye is Block's Dessa. Block calls it "a true speech-to-speech model" trained "from scratch on nothing but human conversation," with "no text, no pretrained models." Block says it's full duplex, knows "when to cut in and when to backchannel," and runs "faster than real time." To be clear, Dessa is a research demo, not a product you can buy. Block is giving early access to a small group of AI safety researchers. It's worth watching because it shows where voice AI is going: calls that feel like talking to a person.
What a local shop can use now
Two things are practical today.
1. A missed-call receptionist. This is the front door. An AI assistant picks up when you're on a roof in Cape Coral or under a house in Lehigh Acres, says it's a virtual assistant, takes the caller's name, address and problem, and books the visit or promises a callback. Emergencies go straight to your cell, and callers can always ask for a person. One honest caution: even good voice AI can mishear a name or a phone number, so a solid setup reads details back and confirms by text.
2. An owner assistant you talk to. This is newer. Some consumer apps now let you call an AI assistant like you'd call a person. Sesame, for example, says its assistants can work in Gmail and Calendar with your permission, "have their own computer," and "let you know when they've finished." Its own guide also warns that the assistants "sometimes make things up," which is a fair warning for any AI tool today.
Where it's heading
Put those two pieces together and you get the idea we're building toward. Today the AI answers your customers. Next, you talk to it too: "What came in this morning?" "Move the Estero job to Thursday." "Text my two o'clock that I'm running late." It handles the background work and checks with you before anything important goes out.
That's what SideCrew calls the Full Crew. SideCrew is the crew behind your crew: we help shops get found, answer and book, and follow up and get reviews. Voice is the front door, not the whole house. We're also upfront that SideCrew is one operator plus AI tools, and you approve the scripts, prices and rules before anything goes live.
Native speech models with tools are still new, and we'd rather see them prove reliable on real phone lines than rush them into your business. Until then, the answering setup that works today is the place to start.
Questions owners ask
What is an AI voice agent? It's an AI that holds a spoken conversation, usually on the phone. It can answer questions, take details and book appointments, and it should say up front that it's a virtual assistant.
What's the difference between a voice pipeline and speech to speech? A pipeline turns speech into text, has a text AI reply, then turns that reply back into a voice. A speech-to-speech model works on the audio directly, so it can hear tone and usually answers faster.
What does full duplex mean? The AI listens and talks at the same time. You can interrupt it naturally, and it can react while you're still talking, like a person on the phone.
Can I buy Block's Dessa for my business? No. Block presents Dessa as research, with early access for a small group of AI safety researchers. It isn't a product for businesses.
Curious what your callers hear right now?
If you want to see how an AI answering setup would work on your line, read how we answer every call and book the job. Or start with our free Front Door Audit. It takes about 15 minutes, and you keep the report either way.
Sources
- Block, Dessa research page (read October 26, 2026): Dessa, the conversation model trained only on speech
- OpenAI developer docs, voice agent architectures, including GPT-Live full duplex and the chained voice pipeline (read October 26, 2026): OpenAI voice agents guide
- ElevenLabs, "Introducing Eleven Realtime," October 20, 2026, with speed figures from Artificial Analysis as reported by ElevenLabs: ElevenLabs Eleven Realtime announcement
- Sesame, "Getting started," October 5, 2026: Sesame getting started guide
Find out what’s costing you jobs. The free audit takes 15 minutes, and you keep the report either way.
Book a free audit