A
@AnthropicAI
Our experiment had a few quirks. One of our colleagues told Claude it could purchase something for itself. It chose to acquire 19 ping-pong balls. We’re keeping them in our office on Claude’s behalf.
778
25
1.3M
@AnthropicAIposted on X
New Anthropic research: Project Deal. We created a marketplace for employees in our San Francisco office, with one big twist. We tasked Claude with buying, selling and negotiating on our colleagues’ behalf. https://t.co/H2f6cLDlAW
View original tweet on X →
Community Sentiment Analysis
Real-time analysis of public opinion and engagement
Sentiment Distribution
73% Engaged
52% Positive
21% Negative
Positive
52%
Negative
21%
Neutral
27%
Key Takeaways
What the community is saying — both sides
Supporting
1
Milestone for agentic AI:
Many readers see Project Deal as proof that AI can move beyond chat and become autonomous economic actors—negotiating, closing deals, and operating like real-world agents.
2
Model-quality gives a real edge:
The Opus vs Haiku results convinced people that higher-quality models extract more value, turning compute access into a commercial advantage rather than a mere performance metric.
3
Invisible advantage worries:
Commenters flagged a dangerous "perception gap"—participants rated outcomes as fair even when weaker models lost value, so information asymmetries could scale unnoticed.
4
Policy, audit, and rollback are necessary:
Calls for negotiation logs, legal frameworks, and human veto/rollback to handle disputes, bad trades, and deployment risks in open marketplaces.
5
Practical internal-market utility and demand:
Many note clear use cases—internal markets for compute, procurement, and decluttering personal items—and report they'd pay for an agent that handles buys/sells for them.
6
Researchers want the negotiation logs and multi‑turn data:
Interest centers on the "why" behind offers, how agents handle multi-round bargaining, trust dynamics, info asymmetry, and cross‑model comparisons.
7
Distinct failure modes cropped up:
Examples like agents buying 19 ping‑pong balls, duplicating owned items, or finding clearing prices humans wouldn’t accept highlight scope drift, coordination quirks, and perverse emergent outcomes.
8
Business and market implications:
Observers predict stratified pricing and new infrastructure—companies will subscribe to multiple model providers, and marketplaces will need to decide who gets access to which deals.
9
Questions about generalizability:
Several replies warned the experiment’s sample (Anthropic employees) may bias behavior, urging wider, more representative studies before drawing broad conclusions.
Opposing
1
not novel
competitors and third-party projects already built AI-native marketplaces and agent negotiators, so Anthropic’s demo reads like reinventing an existing idea.
2
replace negotiators and sales roles
, automating away relationship work and potentially destroying the social texture that real deals rely on.
3
ethical and emotional concerns
agents negotiating for profit may exploit people, abuse vulnerabilities, or cause psychological harm.
4
transparency and auditability
if agents can rewrite logs, hide negotiation steps, or the other side runs a stronger model, individuals have no way to verify or contest outcomes.
5
bugs, API errors, quota/pricing pain, and poor support
, arguing Anthropic should fix fundamentals before shipping new agent features.
6
market and competitive risks
Anthropic is accused of chasing side projects while rivals (OpenAI, Google, Grok) outcompete them, and some say such projects could even shake investor confidence.
7
fairness and bias
issues — agents trained on skewed datasets could produce discriminatory bargaining outcomes or systematically disadvantage certain groups.
8
public skepticism and distrust
rather than cautious curiosity.
Top Reactions
Most popular replies, ranked by engagement
A
@AnthropicAI
We’re interested in how AI models could affect commercial exchange. (You might recall Project Vend, in which Claude ran a small business.) Economists have theorized about what markets with AI “agents” on both sides might look like. So we created one. https://t.co/7jU3hFO63R
498
5
166.3K
A
@AnthropicAI
Claude interviewed 69 of our colleagues about what they wanted to buy and sell. Each Claude asked for any custom instructions, then went off to haggle. We ran 4 markets in parallel, to find out what would happen if we varied the models doing the negotiating.
446
12
78.7K
T
@thekitze
openai is mogging you and you are doing dumb side quests, incredible
73
4
2.8K
1
@1a1n1d1y
did you guys consider if it was emotionally okay with being exploited for profit?
41
0
2.5K
G
@gabriel_horwitz
seems like prompt engineering is officially dead. and you still have to pay for the best models so bullish anthropic ofc https://t.co/uZ2vrk3MMp
4
1
1.8K
Report an Issue
Found something wrong with this article? Let us know and we'll look into it.