How a Mechanic Shop Owner Built an AI Receptionist — Without Writing a Line of Code

The Story That Started This

Someone on Hacker News — a non-traditional builder, not a professional developer — spent a weekend building an AI phone answering system for their brother's mechanic shop. The shop was losing business because calls were going to voicemail during peak hours. The owner was under a car, hands covered in grease, and couldn't pick up. Calls went to voicemail. People didn't leave messages. They called the next shop instead.

The solution? An AI that picks up every call, sounds like a real receptionist, answers questions about services and pricing, and either books appointments or takes a message — all without the owner lifting a finger.

The post hit 110 upvotes. The comment section was half people saying "that's brilliant" and half people asking "how do you even do that?" This guide is the answer to the second group.

Here's the thing: this kind of project is exactly what vibe coding is built for. You don't need to understand how voice AI models work under the hood. You don't need to write Python scripts or set up servers. You need to understand your business, write clear instructions for the AI, and connect a few tools together. The technical heavy lifting is already done — you're just telling the machine what to do.

If you've ever trained a new employee, you already know the hardest part of this project.

What You're Actually Building

Before we dive in, let's be clear about what this is and what it isn't.

What it is: A phone number that callers can dial. When they do, an AI voice answers, greets them with the business name, and handles the conversation based on instructions you wrote. It can answer common questions (hours, services, pricing, location), take messages with a name and callback number, and tell callers when to expect a call back.

What it isn't: A magic system that knows everything about your business automatically. It knows what you tell it. If you don't tell it that the shop closes at 5pm on Fridays, it won't know. If you don't tell it what to say when someone asks about transmission work, it'll have to guess or admit it doesn't know. The quality of the receptionist is directly tied to the quality of the instructions you give it.

Think of it like this: you're writing a training manual for a new hire who happens to be incredibly fast, never gets tired, and answers the phone on the first ring every single time. Your job is to make that manual complete enough that they can handle the job without calling you every five minutes.

The AI handles the phone. You handle the manual.

Step 1: Choose Your Platform

There are a handful of platforms that handle AI voice calls. The three most relevant for this project are Bland.ai, Vapi, and Retell AI. Here's the honest rundown:

Bland.ai is the most beginner-friendly option. The dashboard is clean, the setup is straightforward, and they have decent documentation. Pricing is around $0.09 per minute of call time. If this is your first AI phone project, start here. It's the platform the HN poster used for the mechanic shop.

Vapi is more powerful and more flexible. It supports more voice options, has better integration with other tools (like calendars and databases), and is popular with developers who want to customize deeply. It's slightly more complex to set up but gives you more control. Good choice if you know you'll want to add appointment booking later.

Retell AI is newer but worth knowing about. It's focused on natural-sounding conversations and handles interruptions (when callers talk over the AI) better than the others. If the business you're building for has customers who tend to be impatient talkers, this one's worth a look.

For this guide, we'll walk through the setup using Bland.ai — but the concepts apply to all three. Once you understand how to set up one, switching to another takes maybe an hour.

Step 2: Create Your Account and Get a Phone Number

Head to Bland.ai and sign up. You'll get a dashboard where everything happens. The first thing you need is a phone number — this is the number you'll give to callers (or forward your existing business line to).

In the Bland.ai dashboard, go to Phone Numbers and click Buy a Number. You can usually pick a local area code so the number looks like a local business, not an 800 number. This matters more than you'd think — people trust local numbers.

If the business already has a phone number they use on Google Maps and their website, you have two options:

• Forward the existing number to your new AI number. Customers call the same number they always have. The call gets forwarded to the AI. This is the smoothest option — no need to change anything on the business's listings.
• Use the new number as the primary number. Update Google My Business, the website, and any directories. Takes more work upfront but keeps things clean.

For the mechanic shop project, forwarding is usually the right call. The owner already has a number on their signage, business cards, and Google listing. Forwarding keeps all of that intact.

Step 3: Write the Receptionist Prompt — This Is the Whole Job

Here's where most people either nail it or get frustrated. The prompt is the instruction manual for your AI receptionist. It tells the AI who it is, how it should sound, what it knows, and how it should handle different situations.

A bad prompt produces a robotic, unhelpful AI that frustrates callers. A good prompt produces something that sounds almost indistinguishable from a real receptionist — warm, knowledgeable, and helpful.

The good news: writing a good prompt doesn't require any technical skill. It requires knowing the business. And if you're building this for yourself or someone you're close to, you already know more than enough.

Here's the structure I recommend for any local business receptionist prompt:

Part 1: Identity and Personality

Tell the AI who it is, what business it works for, and how it should sound.

A few things to notice here: you gave the AI a name (Sam — people respond better to names), a location (helps with context), a tone description, and a clear instruction about handling the "are you a robot?" question. That last one comes up more than you'd expect.

Part 2: Business Information

This is where you dump everything the AI needs to know. Be thorough. The more you include, the less the AI has to guess.

Notice the level of detail. Hours, pricing, what they do and don't do, how appointments work. When callers ask "do you do body work?" the AI can give a real answer instead of saying "I'm not sure." When they ask about oil change pricing, it has actual numbers to give.

This part takes the most time because you're essentially doing a brain dump of everything a good receptionist would know on day one. It's worth spending an hour on this. Write it out in a text document first, then paste it into the prompt.

Part 3: Call Handling Instructions

Now you tell the AI how to actually run the conversation — what to do in different situations.

This section is where you handle the stuff that goes sideways. A human receptionist handles these situations on instinct — you're writing those instincts out explicitly for the AI. The more scenarios you cover, the better it performs.

Step 4: Connect the Prompt to the Phone Number

In Bland.ai, you'll create what they call an "Agent" — this is just your AI receptionist with its prompt attached. Here's the basic flow:

1. Go to Agents in the dashboard and click Create Agent.
2. Paste your full prompt into the agent's "System Prompt" or "Instructions" field.
3. Choose a voice. Bland.ai has several options — listen to samples and pick one that fits the business's vibe. A warm, mid-paced voice works best for most local businesses. Avoid the ones that sound too formal or too robotic.
4. Set the First Message — this is what the AI says when it picks up. Something like: "Thanks for calling Mike's Auto Repair, this is Sam. How can I help you today?"
5. Save the agent.
6. Go back to your phone number and assign this agent to it.

At this point you have a working AI receptionist. When someone calls that number, the AI picks up and handles the call based on your prompt. That's genuinely it.

Under the hood, what's happening is that the AI is receiving the audio of the call, transcribing it in real time, running your prompt as the context for its responses, and generating audio back. If you're curious about how these systems communicate with each other, our guide on what an API is explains the concept — but you don't need to understand it to use these platforms. The plumbing is already done.

Step 5: Test It Before Anyone Else Does

Do not hand this to a real customer first. Call it yourself. Call it from your cell. Have your partner call it. Have the business owner call it and pretend to be their worst customer.

Here's a list of test calls to run:

• "What are your hours?" — basic info, should nail this
• "Do you take walk-ins?" — slightly more nuanced
• "How much is an oil change for a 2022 F-150?" — specific question, may not have exact answer
• "Do you do transmission work?" — should say yes per the services list
• "Do you do body work?" — should say no
• "I want to book an appointment for Tuesday" — should collect name and number
• "I need to leave a message for Mike" — should take the message properly
• "My brakes feel weird, is this an emergency?" — should escalate appropriately
• "Are you a real person?" — should handle per the prompt
• Try to confuse it: "Can you tell me where the nearest Walmart is?" — should stay in scope

After each call, go into the Bland.ai dashboard and read the call transcript. You'll see exactly what the AI said and what the caller said. This is how you spot problems — not by guessing, but by reading the actual conversation.

When you find something it gets wrong, go back to the prompt and add an instruction that covers that case. Then test again. You'll usually go through two or three rounds of this before it feels solid.

Step 6: Handle the Edge Cases (The Stuff That Actually Happens)

Theory is easy. Real calls are messier. Here are the edge cases that trip up most first-time AI receptionists — and how to handle them in your prompt.

The caller who won't stop talking

Some people call and just… keep going. Story about their car, story about their last mechanic, story about their dog. The AI needs to stay warm but steer things back to being helpful. Add this to your prompt:

The caller who asks something totally off-topic

People call mechanic shops asking for directions to nearby businesses, asking what the weather is, asking about things that have nothing to do with cars. You need the AI to stay in its lane without being rude about it.

The caller who's angry from the start

Maybe they had a bad experience. Maybe they're just having a bad day. Either way, the AI can't get defensive, can't argue, and absolutely cannot escalate. De-escalation language works here the same way it works with a human receptionist.

The "let me speak to a human" caller

It will happen. Some people don't want to talk to an AI, whether they know it's an AI or not. Have a plan.

Step 7: Set Up Where the Messages Go

The AI is answering calls and collecting information — but that information needs to end up somewhere the business owner actually looks.

Bland.ai has a few options here. The simplest is email notifications — every time the AI takes a message, it sends an email with the caller's name, number, and what they called about. For a small business owner who checks email, this works fine.

If the owner is more of a text person (most mechanic shop owners are), you can set up SMS notifications using a service like Twilio or through Bland.ai's built-in notification settings. Every message goes straight to their phone as a text.

For higher-volume businesses, you might look at connecting to a tool like Zapier to pipe messages into a shared spreadsheet, a CRM, or a Slack channel. This is where things get slightly more involved — but it's still more "connect the dots in a dashboard" than actual coding. Our guide on building your first project covers the general approach to connecting tools like this.

For the mechanic shop project, the setup was simple: email for all messages, with a daily summary sent to the owner at 6pm. He checks it when he closes up for the night and calls back anyone who needs it the next morning.

Step 8: Go Live and Watch the First Week Closely

Once you've tested it and you're reasonably happy with how it performs, go live. Forward the business number (or start giving out the new number). Let it take calls.

The first week, check the transcripts daily. Not to micromanage the AI, but to learn. You'll see patterns — questions that come up over and over that you didn't include in the prompt, scenarios you didn't anticipate, ways callers phrase things that confused the AI.

Every time you spot a gap, update the prompt. This is an ongoing process, not a one-time build. The AI gets meaningfully better in the first two weeks as you add the edge cases you couldn't have predicted up front.

After the first week, you'll probably only need to check transcripts once or twice a week. After the first month, most of the rough edges are gone and it pretty much runs itself.

What This Actually Costs (Real Numbers)

Let's talk money, because this is often the first question from a business owner you're pitching this to.

For a local auto shop getting roughly 30-50 calls per day:

• Average call length: 2-3 minutes
• Daily call time: 60-150 minutes
• Monthly call time: 1,800-4,500 minutes
• At $0.09/min (Bland.ai): $162-$405/month

Compare that to a part-time receptionist (even 20 hours/week at $18/hour): $1,440/month. And the AI answers calls on evenings, weekends, and holidays.

For lower-volume businesses — a plumber, a hair salon, a dog groomer — the math gets even better. 10-15 calls a day might cost $30-60/month total.

Most platforms also offer free trials or free tiers. Build it for free, test it thoroughly, then decide if it's worth paying for. The HN poster said their brother paid for the first month in recovered business within the first week — calls that used to go to voicemail (and then to a competitor) were getting answered.

What to Build Next

Once your basic AI receptionist is working, there are a few natural next steps:

Add appointment booking. Instead of just taking a name and number, the AI can check availability and actually book the appointment. This requires connecting to a calendar tool — Vapi has better support for this than Bland.ai, and there are guides on their documentation sites for connecting to Cal.com and Google Calendar.

Add a post-call follow-up text. After every call, automatically send a text to the caller with the shop's address, a link to their Google listing, or a "thanks for calling, here's what we discussed" message. This is a simple Zapier automation and it leaves a great impression.

Build the same thing for outbound calls. Both Bland.ai and Vapi support outbound calling — the AI can call customers to confirm appointments, follow up after service, or remind them about seasonal maintenance. This is a slightly more advanced setup but uses the same skills you built here.

Apply it to other businesses. Once you've built this once, building it for a second business takes half the time. The mechanic shop is a template. A dental office, a landscaping company, a real estate agent — they all have the same core problem (missed calls, voicemail, lost business) and the same solution. There's a real freelance service business here if you want it.

The Bigger Picture: Why This Matters

A lot of people in the vibe coding world are building apps and websites. That's great. But there's a whole other category of real-world problems that software can solve — and most small business owners have no idea these tools exist.

The mechanic on the HN post isn't a tech person. He fixes cars. He doesn't have time to research AI tools, set up dashboards, or write prompts. But his brother did. And now his shop answers every call, 24/7, and sounds professional doing it.

That's the actual opportunity here. Not building another SaaS product. Building real solutions for real businesses that are currently losing money to problems that are completely solvable.

You don't need to know how transformers work. You don't need a computer science degree. You need to understand the problem, understand the tools, and be willing to spend an afternoon putting them together. That's the whole job description. And you already have it.

This is what vibe coding is for — not impressing other developers, but solving real problems for real people who can't solve them themselves.

What to Learn Next

Ready to keep building? Here's where to go from here:

• Build a Landing Page with AI — Every business that has an AI receptionist also needs a website. This guide walks through building one without coding, using AI the whole way.
• What Is Vibe Coding? — The philosophy behind building with AI as your partner. If you're new to this approach, start here.
• What Is an API? — When you're ready to connect your AI receptionist to other tools (calendars, CRMs, SMS), you'll need a basic understanding of how systems talk to each other. This explains it plainly.

Frequently Asked Questions