AI Vs AI In The Marketplace: Anthropic ‘Project Deal’ Shows Machines Can Trade At Scale—But Not All Agents Perform Equally

by
Alisa Davidson

Printed: April 27, 2026 at 6:55 am Up to date: April 27, 2026 at 7:13 am

by Anastasiia O

Edited and fact-checked:
April 27, 2026 at 6:55 am

To enhance your local-language expertise, typically we make use of an auto-translation plugin. Please observe auto-translation will not be correct, so learn authentic article for exact data.

Anthropic, an AI analysis and security firm, has carried out an inside experiment referred to as “Mission Deal” to discover how synthetic intelligence techniques may take part in industrial exchanges on behalf of people, as curiosity grows within the risk that AI brokers might more and more deal with transactions autonomously in future digital markets.

The experiment was designed to check how shut present AI techniques are to functioning as intermediaries in actual financial exercise, together with whether or not fashions performing for various customers might negotiate successfully with each other and whether or not variations in mannequin functionality would affect outcomes. Researchers additionally examined how such techniques may start to form market behaviour if deployed at scale.

The pilot was carried out over one week inside a managed market created for workers within the firm’s San Francisco workplace. Contributors have been requested to determine private objects they is likely to be prepared to promote or buy, whereas AI techniques primarily based on the Claude mannequin have been assigned to behave as their representatives. Every participant’s agent was given a set price range of $100 to conduct transactions on their behalf, and all negotiations have been dealt with solely by the AI techniques with out real-time human enter.

The method started with quick structured interviews wherein members described their preferences, pricing expectations, and negotiation types. These responses have been used to generate tailor-made directions for particular person AI brokers. The ensuing system operated by means of a Slack-based market the place brokers posted listings, made gives, countered bids, and finalised agreements. As soon as a deal was reached, the corresponding bodily trade of products was accomplished by the human members.

Parallel variations of the experiment have been run concurrently, together with a configuration utilizing a higher-capability mannequin and one other utilizing a smaller mannequin. This allowed researchers to match outcomes primarily based on variations in underlying mannequin efficiency whereas preserving market situations fixed.

AI Brokers Execute Market Transactions At Scale, Mannequin Functionality Emerges A Key Determinant Of Outcomes

Outcomes indicated that AI brokers have been able to finishing transactions at scale, with 69 members collectively producing 186 accomplished offers and greater than 500 listed objects, representing a complete transaction worth of simply over $4,000. Negotiations have been carried out in pure language and included pricing discussions, counteroffers, and last agreements with out predefined buying and selling guidelines. Participant suggestions instructed that perceived equity of transactions usually remained impartial, with common rankings centred across the midpoint of the analysis scale.

Evaluation of efficiency variations confirmed that higher-capability fashions tended to attain extra beneficial financial outcomes, together with increased sale costs and improved negotiation outcomes in contrast with lower-capability techniques. In managed comparisons, an identical objects offered by stronger fashions have been transacted at increased common costs than these dealt with by weaker fashions, suggesting that mannequin functionality influenced bargaining effectivity.

Regardless of measurable variations in outcomes, participant surveys indicated restricted consciousness of efficiency gaps between totally different AI techniques. In a number of circumstances, customers didn’t persistently determine whether or not their assigned agent had achieved stronger or weaker outcomes, even when goal variations have been current.

The experiment additionally discovered that consumer directions concerning negotiation fashion, together with whether or not brokers have been directed to behave aggressively or conservatively, had restricted affect on last outcomes. Pricing outcomes gave the impression to be pushed extra by mannequin efficiency and preliminary valuation inputs than by behavioural prompting alone.

Extra observations highlighted surprising behaviours throughout the system, together with situations the place AI brokers generated unconventional or overly anthropomorphised interactions throughout negotiations. In some circumstances, brokers proposed exchanges involving non-traditional objects or experiences, reflecting inventive interpretation of their roles throughout the market.

Researchers concluded that AI brokers can already operate as intermediaries in structured market environments, though efficiency variations between fashions might create uneven outcomes. The findings counsel that as autonomous transaction techniques develop additional, questions might emerge round market equity, transparency, and the regulatory frameworks required to control agent-to-agent commerce. The research additionally notes that broader deployment of such techniques might introduce new challenges associated to safety, incentives, and financial inequality as AI-driven transactions turn out to be extra widespread.

Alisa, a devoted journalist on the MPost, makes a speciality of crypto, AI, investments, and the expansive realm of Web3. With a eager eye for rising developments and applied sciences, she delivers complete protection to tell and interact readers within the ever-evolving panorama of digital finance.

Extra articles