I fucking hate car dealerships.
Not the people who work there—many of them are just trying to make a living. I hate the system. The mandatory 20 hours of your life spent in showrooms. The “how much would you like your payment to be?” The endless trips to “talk to my manager.” The bait-and-switch on financing. The discovery that the price you negotiated doesn’t include the $2,000 in mandatory dealer add-ons they conveniently forgot to mention.
This week, I watched someone do something that made me realize we’re living through a fundamental shift in how these transactions work. They automated the entire negotiation. Not by hiring a broker or using a car-buying service, but by deploying an AI agent that spent three days playing dealerships against each other. The dealerships treated the AI as a normal, price-sensitive buyer.
The Asymmetry Problem
Car buying hasn’t fundamentally changed in decades. We’ve digitally transformed everything from banking to groceries, but when you want to buy a car, you’re still trapped in a negotiation system designed in the 1950s.
It’s brutal. You buy a car every five to seven years. The salesperson across from you closes 100+ deals annually. They’ve been trained on every psychological tactic. They know exactly what you paid for your trade-in because they have access to auction data you don’t. They know their actual margin. They know which objections are real and which are theater. You know… that you need a car.
Services like CarEdge try to level this playing field. For $40-50, they’ll deploy an AI agent to contact dealers and negotiate on your behalf. Since launching in mid-2024, they’ve helped thousands of customers save an average of $1,500 and five hours of dealership time. Other companies like Cargantic offer similar services with human consultants or AI tools.
These services work, but they have limitations. Dealers know you’re using them and adjust tactics accordingly. Most are US-only, leaving Canadian buyers (and others) out. And there’s an inherent cost barrier—not everyone wants to pay $50 for a service when they’re already stressed about the purchase price.
The information asymmetry persists because dealerships profit from customer fatigue, confusion, and emotional decision-making. They’re designed to extract maximum value from your exhaustion.
Enter the Personal AI Agent
Then MoltBot happened.
MoltBot (recently renamed from Clawdbot after a trademark dispute with Anthropic) is an open-source, self-hosted AI agent platform that’s gone properly viral in the past month—hitting 60,000+ GitHub stars almost overnight. It’s a persistent agent that lives in your messaging apps, can control your browser and computer, and executes multi-step tasks autonomously.
One user, Aaron Stuyvenberg, documented using MoltBot to buy a car. He instructed it to research fair prices on Reddit, contact 8-10 dealerships in his area, collect quotes, play them against each other, and negotiate down to the best price.
MoltBot filled out contact forms on dealership websites and waited for responses. When dealers replied via email, it sent them competing quotes from other dealerships. When they tried to call or text, Aaron and the bot politely redirected them back to email (AI agents don’t do phone calls well yet, but voice is coming and it’s going to be huge).
Three days later, MoltBot and Stuyvenberg had reached an agreement with a dealership. He saved over $4,200 and never set foot in a showroom until he was ready to sign paperwork.
This is fundamentally different from ChatGPT writing you an email template. MoltBot is actually a real agent. It maintains state across days or weeks, integrates with email, SMS, web search, and your browser. It runs 24/7, doesn’t get tired, doesn’t get emotionally manipulated, and doesn’t fall for artificial urgency.
The Stealth Advantage
Here’s the key insight most people are missing: an AI agent is not the same as a broker service, from the dealership’s perspective.
Dealerships have playbooks for dealing with brokers, fleet buyers, and car-buying services. When someone contacts them through CarEdge or similar services, they know immediately. They might still give you a good price, but they’re operating from their “informed buyer” playbook rather than their standard consumer playbook.
A self-deployed AI agent appears as an individual customer. It triggers the dealership’s normal sales process and they deploy their standard tactics, except none of those tactics work on machines.
The scarcity play (“this price is only good until end of business today”) doesn’t create FOMO in software. The relationship-building doesn’t establish trust that can be leveraged later. The confusion about what fees are “mandatory” doesn’t work when the agent has been explicitly instructed to reject certain charges.
The AI agent reverses the information asymmetry. It sees all 10 dealers’ offers in real-time, whereas each dealer sees only their conversation thread. They have no idea you’re simultaneously negotiating with their competitor down the street (or if they do know multiple dealerships are being played against each other, they won’t expect the sheer scale at which AI can do this). By the time they figure it out, the agent has already extracted their best pricing.
Risks and Security
If you’re thinking about doing this, be very cautious. MoltBot recently had some serious security incidents—researchers demonstrated prompt injection attacks via email that could trick the agent into forwarding your private emails to attackers. The creator himself describes running it on your primary laptop as “spicy.”
You almost certainly need to create a burner identity for your AI agent. Use a separate email address and get a Google Voice number or similar for SMS. Consider a dedicated VPS or Mac Mini or other small computer specifically to run MoltBot. Whatever you choose, do not run this on production systems. Give it limited access to only what it needs.
What This Means for Every Industry
As someone who advises enterprises on AI strategy, I see this car-buying example as a microcosm of something much larger.
AI agents excel at structured negotiations where rules are knowable but complex, where one side negotiates professionally and the other does so infrequently, and where information asymmetry and fatigue create advantage. Car buying fits this perfectly. So do insurance renewals and long-term service contracts.
Some companies will productize this capability on behalf of customers. Others will find themselves negotiating against it. The advantage shows up quickly for whoever internalizes the pattern first.
CarEdge is an early example of what this looks like when it becomes infrastructure rather than a one-off hack: persistent agents doing the repetitive negotiation work at scale, at a cost that undercuts traditional intermediaries.
For business leaders, the question is uncomfortable but unavoidable: Where in your industry do customers accept friction-based pricing? Where does negotiation complexity create margin? Where does revenue depend on customers not having perfect information?
Your customers are already building AI agents for exactly those situations. The car dealership is just an early signal.