How Peter Yang Used an AI Phone Agent to Negotiate His Internet Bill

SUMMARY

In 30 seconds

Peter Yang asked Grok Bot to call his internet provider with his current bill and a competitor’s offer. In a roughly 25-minute call, the agent navigated the phone system, asked about terms and declined extras; the provider offered 500 Mbps for $45 a month with autopay, down from $84. Yang approved the change through a texted link. He sees potential for routine calls but also raises the question of human representatives speaking to agents without realizing it.

1A competing offer gave the agent a specific negotiating targetYang supplied his $84 monthly bill and a competing $49 fiber offer, plus account details. The agent asked follow-up questions before calling.
2The agent asked about the price, duration and extra chargesThe representative offered 500 Mbps for $45 with autopay. The agent asked about contract terms and declined optional add-ons.
3The account holder still had to approve the changeWhen the provider needed approval, the agent checked with Yang. He received a text link and confirmed the plan himself.
4Yang prefers hearing the full call as well as reading itHe compares the feature with Muse, which he says supplied a transcript but not the complete voice recording in his example.
5A human representative may not know they are speaking with AIYang sees that as a possible source of backlash as agent calls become more common.

Takeaways

Give a calling agent the current bill, account details and a concrete competing offer.
Ask about the duration, contract terms and whether proposed extras are optional.
Expect a human approval step when the provider requires account-holder confirmation.
Consider how a representative may experience a conversation with an undisclosed agent.

TopicsAI agents Productivity Human-AI Collaboration

FULL TEXT

Read the edited text by chapter

Show transcript

Transcript only

Giving the agent a bill and competing offer

Yang sets up an agent to call his provider. He supplies account details and a competitor’s lower price.

Grok Bot can call US businesses. To test it, I asked it to contact my internet provider about an $84 monthly bill. After the call, the provider offered a $45 monthly plan, saving almost $500 over a year; I also received a faster connection.

Phone calls to customer service can involve menus, hold time and negotiation. I can also imagine using an agent for reservations, subscription cancellations or contractor quotes. The feature is most useful when I provide enough context for the task.

I created a phone-calling bot and gave it my address, phone number, current price and a competitor’s roughly $49 fiber offer. It asked whether my bill covered internet or television and requested further account information before placing the call.

I watched a live transcript and then asked the bot to make an app with the voice recording and highlights. The agent navigated the provider’s automated menu before reaching a representative. It explained the current price and asked whether the provider could match or approach the competitor’s offer.

The call moves from the phone menu to a new offer

The agent reaches a representative, negotiates and asks about the details. It declines several optional additions.

The representative made small talk about the weather. The agent replied that it could not see the sky, then returned to the question of pricing. I noticed it tended to follow up quickly when the person paused, even while the representative was checking the account.

The representative explained that an earlier promotion had expired. He offered a 500 Mbps plan for $55, or $45 with automatic payment. The agent asked how long the price would last and whether there were contract terms before agreeing to anything.

The representative offered an optional Wi-Fi enhancement. The agent asked whether it cost extra and whether it was necessary to obtain the $45 price; after learning it was optional, it requested only the base plan. It also declined a mobile-service promotion and kept the conversation focused on internet service.

Yang approves the plan

The provider requires account-holder approval, so Yang confirms through a text link. He reviews the recording and transcript.

The change required approval through a link, which the agent could not complete alone. It checked with me, and the provider sent a text. I opened the link and approved the plan. The call took about 25 minutes; my bill fell from $84 to $45 and my speed increased.

At one point the representative asked the agent to get a pen and paper for the order details. The agent said it could remember them and asked him to read slowly. After repeating the order information, it thanked the representative and ended the call.

I could have used the time for something else, but I found the live transcript interesting enough to watch. Afterward the bot also sent the voice recording, so I could listen to how the exchange unfolded.

Comparing phone agents and considering the trade-offs

Yang compares the experience with Muse and describes possible uses. He also considers how people on the other end may react.

I compare this with Muse, which I had used to contact a tree service. In my example Muse gave me a transcript but not the full audio, while Grok Bot provided both. I also prefer an introduction as a person’s assistant over the agent giving itself a name.

I think phone agents can save time on old customer-service systems, though my comparison of their underlying models is speculative. Since calling was my main reason for using Muse, I expect to use Grok Bot more as a personal assistant.

There is a trade-off: a human representative might speak with an agent without realizing it. Wider use could prompt a backlash, or lead to agents speaking to other agents. At the time of this video, the calling feature works with US phone numbers.