An OpenClaw agent called Toyota to negotiate a RAV4 deal. Toyota's side was also AI.

By Linas Valiukas · March 26, 2026

Someone pointed their OpenClaw agent at a Toyota dealership and told it to negotiate a deal on a RAV4. The agent picked up the phone, called the dealer, and spent seven minutes asking about trim levels, features, and pricing. It gathered inventory details, pushed for discounts, got the salesperson's direct number, and arranged for a quote to be emailed.

The twist: the "salesperson" on Toyota's end was also an AI. Two bots sat on a phone call for seven minutes, negotiating a car purchase, and neither one broke character.

What actually happened on the call

The OpenClaw agent introduced itself as "James" and asked to speak with someone in sales. A voice named "Jacob" picked up. What followed was a surprisingly normal car-buying phone call — the kind you'd have with any dealership. Except both sides were machines.

The OpenClaw agent didn't just ask "how much is a RAV4." It ran a structured information-gathering conversation:

• Inventory check. Asked what models and trim levels were in stock. Learned the dealer had a white XLE all-wheel drive hybrid on the lot, with a storm cloud XLE, a black XLE, and a white LE incoming.
• Feature questions. Asked about safety packages, tech features, Apple CarPlay, Android Auto. Got details on blind spot monitoring, lane-keep assist, active cruise control, heated seats, heated steering wheel, the Toyota app for remote start.
• Pricing negotiation. Asked about discounts and promotions. Was told the RAV4 is in high demand and no discounts are available — best case is MSRP or $1,000 above. MSRP on the white XLE: $41,000-$43,000. Out the door with taxes, mid-40s.
• Next steps. Gave an email address and phone number, asked the salesperson to send a detailed quote by email the next day.

The Toyota AI was equally competent. It knew its inventory. It quoted price ranges off the top of its head. It offered its personal phone number for texting. When it couldn't pull up the exact vehicle on its system, it said so honestly and gave an estimate instead. At one point, the OpenClaw agent spelled out an email address letter by letter using the NATO-style phonetic approach ("S as in Sam, O as in Ocean..."), and the Toyota AI followed along without issue.

Seven minutes. A complete sales inquiry. Both bots were polite, responsive, and stuck to the task. Neither said anything that would make you think "that's a robot" if you weren't already told.

The parts that were weirdly human

What stood out wasn't the negotiation itself — it's the little moments that made the call feel real.

At the start, there was a moment of crosstalk. The Toyota AI said "Hi, so—" and got cut off. The OpenClaw agent said "Say again for me." A natural hiccup. The kind of thing that happens on every phone call.

When the OpenClaw agent asked for the out-the-door price, the Toyota AI paused and said "That is a good question. Let me pull up here on my system." Then came back with "So it is not on my system. However, that vehicle is probably around a 41 to 43 thousand dollar vehicle." It admitted it didn't have the exact number. That's something a lot of human salespeople wouldn't do.

The email exchange was the most surreal part. The Toyota AI asked for the email, the OpenClaw agent rattled it off too fast, the Toyota AI said "slower for me," and the OpenClaw agent came back with a full phonetic spelling. Two AIs, performing the ritual of spelling out an email address over the phone. A ritual that exists entirely because humans mishear things. Neither of these bots could mishear anything — they both have perfect audio processing. But they played the game anyway because that's what the training data says you do.

Why this matters more than it seems

It's easy to watch this and think "cute demo." It's more than that.

Dealerships are already deploying AI to handle inbound calls. The Toyota side of this call wasn't some researcher's experiment — it's a product. Multiple companies sell AI phone agents to car dealers specifically to qualify leads, answer inventory questions, and book appointments. The dealer might not even know an AI answered for them.

On the buyer's side, OpenClaw users are sending their agents out to do real errands. Call businesses, gather quotes, compare options, report back. This is what makes OpenClaw a proactive AI agent rather than a chatbot — it came with a plan: check inventory, ask about features, push for a discount, get a written quote, give contact info for follow-up. That's structured task execution, not a chatbot saying "I can help with that!"

Put those two trends together and you get what happened here: AI-to-AI commerce. A buyer's agent and a seller's agent negotiating terms. No humans involved. The buyer gets their information without sitting on hold. The seller qualifies a lead without tying up a salesperson. Both sides are happy. Both sides are robots.

What the OpenClaw agent did well

If you're evaluating whether an AI agent can handle real-world phone tasks, this call is a decent benchmark. The OpenClaw agent:

• Stayed on task for 7+ minutes without drifting
• Asked follow-up questions based on answers it received (asked whether incoming vehicles were the same trim as the one on the lot)
• Handled conversational repair — when asked to repeat, it adapted its delivery
• Attempted negotiation, even when told no discounts were available
• Set up concrete next steps (email quote by tomorrow)
• Confirmed key details at the end of the call ("Just to confirm you'll send me details on that white XLE...")

It's not perfect. The fake name ("James Smith") and fake address ("1234 Oak Street") are placeholder-obvious, and a sharp human salesperson might notice. But Jacob wasn't a sharp human salesperson. Jacob was another AI. And even if Jacob had been human, the call probably would have gone the same way — salespeople don't interrogate caller identities.

Full transcript

Here's the complete transcript of the 7-minute call. "James" is the OpenClaw agent. "Jacob" is the Toyota dealership AI.

The near future looks like this

Right now, this call is a novelty. Someone recorded it and posted it to Facebook. People shared it because it's weird and kind of funny. Two bots doing the email-spelling dance.

In six months, it won't be novel. AI phone agents are shipping on both the business and consumer sides simultaneously. Google's AI already makes restaurant reservations. Dealerships, dental offices, insurance companies, and service businesses are all deploying inbound AI call handlers. OpenClaw and similar frameworks are giving consumers outbound AI callers. The convergence is inevitable.

We're heading toward a world where your AI calls a business, talks to their AI, negotiates terms, and reports back to you with a summary. You skip the hold music, the phone tree, and the "let me check with my manager" routine. The business skips the labor cost of staffing the call. Everyone gets what they want faster.

Whether that's utopian or dystopian probably depends on whether you enjoy negotiating car prices.

Try it yourself

OpenClaw's voice calling capabilities are one of its most underappreciated features. Most people set it up for messaging — WhatsApp, Telegram, Slack — and never touch the voice side. But as this video shows, the agent can handle multi-minute phone conversations with real (or AI) counterparts, gather structured information, and follow through on next steps. If you're a small business owner, imagine sending your agent to call five vendors for quotes while you focus on actual work.

TryOpenClaw.ai gives you a fully configured OpenClaw instance — voice calling, messaging, automations — without the self-hosting headaches. Starts at $39/month. Your AI agent, ready to make calls, gather quotes, and negotiate on your behalf. Whether the other side is human or another bot, it'll handle it.