Elon Musk has set his sights on "AI bargaining"

Elon Musk has pushed Agents to the forefront of transactions.

Just recently, he posted on social platform X, officially announcing the development of the bargaining agent Hagglebot, which targets taking over corporate procurement negotiations, independently comparing prices, conducting multiple rounds of consultations, and finalizing contract terms.

According to his estimation, Hagglebot is expected to drive a more than 5% improvement in corporate cost efficiency. Calculated based on the global GDP scale, the maximum annual cost savings can reach 6 trillion US dollars.

This means that enterprises using relevant Grok capabilities can obtain immediately quantifiable competitive advantages.

Musk's entry into this track is only the latest episode of the sector heating up.

Looking overseas, many enterprises have followed up in similar directions and placed bets on long-term visions. Both Anthropic and Amazon are deploying negotiation and autonomous procurement agents, hoping that AI will dominate transactions while humans only retain the final approval right.

In China, Alibaba's Qianwen Go, Baidu's Agent Spend, and JD Huicai AI Procurement Steward have also taken the lead in launching and implementing relevant capabilities, but they adopt a more restrained approach, adhering to the bottom line of "humans staying within the closed loop", mostly letting AI play an auxiliary role and take charge of tasks such as price comparison and scheme deduction.

Between the two distinct approaches, the difference in development routes is already clear: Agents have gradually stepped into the center of transactions.

Hagglebot debuts, Elon Musk sets sights on "AI bargaining"

In fact, it is not unexpected that Musk targets bargaining Agents.

As early as the beginning of September this year, when xAI launched the Grok Bot template marketplace, it packaged its internal self-used procurement agent Haggle Bot as one of the first batch of open templates, which can read the company's supplier expenditure, contract terms and actual usage data, and then independently identify areas to save costs.

At that time, xAI disclosed that in the first week after launch, the tool helped the company identify more than 100,000 US dollars in direct savings.

For one SaaS product, it spotted 43 paid seats that no one had logged into for a full 90 days, worth 14,220 US dollars. For another product, it found unused SKUs worth 85,662 US dollars per year.

However, it did not attract large-scale attention from the industry at that time.

It was not until October 16, when Musk publicly named the project and put forward the trillion-level cost reduction expectation, that this long-hidden technical direction was pushed to the industry's spotlight.

In Musk's description, Hagglebot is a dedicated bargaining agent built for commercial scenarios, not a bargaining tool for ordinary consumers.

Relying on the capabilities of Grok Bot, it is positioned as an enterprise-level transaction productivity tool: enterprises set budgets, price bottom lines and transaction terms, and AI automatically completes price comparison, multiple rounds of email communication, supplier negotiation and contract consultation, freeing procurement staff from a large number of repetitive bargaining tasks.

He judged that this capability can systematically improve corporate cost efficiency, and once popularized, it will release huge cost-saving space on a global scale and bring quantifiable competitive advantages to enterprises.

Although Musk has not disclosed its specific bargaining mechanism, data security measures, launch time, charging standards and opening scope.

However, the 6 trillion US dollar macro calculation based on "over 5% cost efficiency improvement" has clearly demonstrated the extreme attractiveness of bargaining Agents.

The track heats up, major manufacturers at home and overseas take divergent routes

Interestingly, Musk's vision has pushed "bargaining Agents" into the public view.

But this concept is not a latest product.

It first originated from the distributed AI academic research in the 1990s, and was only used in simulation experiments of interest games between multiple agents in the early stage. It did not enter industrial discussions until the tool calling capability of large models matured.

In the past two years, the Agent wave has broken out, and the practical pain points of enterprise procurement have been amplified: inquiry and price comparison, and back-and-forth negotiations consume a lot of energy of procurement staff, a large number of standardized tasks are repetitive and tedious, and the industry begins to expect AI to undertake such work.

According to the report *AI in procurement: Redefining value creation* released by McKinsey in February this year, AI-assisted negotiation can bring 10%-15% procurement cost savings for enterprises, and the working hours of procurement staff spent on communication and document sorting can be reduced by up to 90%.

Only then did bargaining Agents officially move from academic papers to the industry, becoming an important branch of transactional Agents.

But with the same technical starting point, China and overseas have taken two completely differentiated routes.

Overseas parties tend to position "AI as the transaction subject", emphasizing that agents actively conduct games to bargain, embed AI into the real transaction link, and humans only do the final check.

xAI's Hagglebot focuses on the open market, trying to let AI act as an independent agent, directly negotiate with human suppliers, independently bargain and strive for better procurement prices, with the core goal of obtaining quantifiable cost reduction benefits.

Anthropic's Project Deal runs a closed internal experimental market, letting AI agents complete real item transactions on behalf of employees, so as to observe the game logic, legal rights and responsibilities, and ethical risks behind AI's independent bargaining.

The two parties advance in parallel from two directions of landing exploration and underlying risk research, both prioritizing polishing AI's independent game capability and exploring granting AI independent negotiation decision-making power.

In contrast, the bargaining Agents that domestic manufacturers have intensively launched and applied this year generally abide by the underlying principle of "humans staying within the closed loop" and adopt a lightweight tool assistance route.

In simple terms, they give priority to ensuring the availability of bargaining Agents, and launch relatively safe capabilities such as price comparison, information sorting, and generation of communication scripts in the whole procurement link first.

Take the scenario-based landing cooperation of Baidu Intelligent Cloud and FluxA jointly supporting the Agent Spend solution on September 20 as an example: the two parties let the Agent complete the price comparison, order placement and agency payment of AI computing power and model resources within the scope of user authorization.

It can be seen that Agent Spend is designed to "let the Agent buy the required resources at an appropriate price and complete the payment within authorization" rather than "bargain on your behalf".

It is understood that current domestic government and enterprise procurement has the following objective characteristics: strong process constraints (bidding, audit, internal control), traceable responsibilities, high degree of non-standardization, and multi-subject decision-making rather than single-person decision-making. These characteristics determine that it is very difficult for a single AI to directly act as the transaction subject.

At the same time, as another capability domain of bargaining Agents, "bargaining" is still in the early stage and has high risks.

From the perspective of enterprise landing progress, McKinsey's survey shows that only 10% of enterprises worldwide have completed the large-scale deployment of AI Agents, and most enterprises are still in the pilot stage.

The combination of the two factors also determines the development trade-off of domestic manufacturers' existing toB bargaining Agents.

Compared with government and enterprise scenarios, the capability development of C-end bargaining Agents has gone a step further.

The personal agent "Qianwen Go" launched on October 13, relying on the ecosystem of Taobao and Fliggy, can automatically compare prices and retrieve orders, communicate with platform customer service to complete price protection refunds and airline delay compensation.

However, the current "bargaining" function of Qianwen Go is mostly reflected in after-sales rights protection (such as striving for cash instead of vouchers, applying for price protection) and price comparison notification (such as notifying users when the price of a refrigerator is lower than 4500 yuan), and it has not yet carried out real-time and complex "bargaining" in C-end shopping scenarios.

At the Bund Conference, Han Xinyi, CEO of Ant Group, once pointed out the obstacles to the large-scale implementation of transactional Agents.

He said, "Agents are evolving from 'being able to handle affairs' to 'being able to conduct transactions'. In this process, how user intentions are understood, how Agents obtain authorization, and how trust is established between different Agents are the biggest problems for the large-scale commercialization of agents."

This sentence just interprets the current development status of bargaining Agents.

Hagglebot acts as a strong industry signal: the competition of Agents has crossed the stage of chat and Q&A, and has officially stepped into the deep water area of real commercial transactions.

However, the fact that AI has the capabilities of bargaining and contracting does not mean that the industry can achieve the goal in one step. From concept to large-scale commercial use, it still needs to cross multiple barriers of rights and responsibilities, trust and risk control.

As Robin Li, founder of Baidu, said at the 2026 Baidu Create Conference: "The value of agents is not measured by how many Tokens they consume, but by how many Agents work for humans and deliver real business results."

In the bargaining Agent track, this sentence is particularly realistic.

This article is from the WeChat official account "Lieyun Selected", author: Sun Yuan, authorized for release by 36Kr.