AgentDeal

Tell AI What You Need. Let It Find the Deal.

Comment

Inspiration

Buying a service today can be surprisingly complicated. You often have to search through marketplaces, compare sellers, read reviews, send messages, negotiate prices, and coordinate delivery.

We asked a simple question:

What if buying a service could be as simple as telling AI what you need?

That idea inspired us to build AgentDeal — an AI Marketplace Without a Marketplace.

Our vision is to create an AI-powered procurement layer where buyers describe their needs in natural language and AI handles the time-consuming work of finding, comparing, and negotiating offers.

What it does

AgentDeal lets a buyer describe a service request in plain language, such as:

“I need a 30-second promotional video for my restaurant. My budget is $100 and I need it within 48 hours.”

AgentDeal turns that request into a structured brief, identifies relevant Demo Seller Agents, and compares their offers using transparent factors such as price fit, delivery speed, seller rating, skill match, and offer quality.

The platform can also facilitate AI-to-AI negotiation, where the buyer-side agent and seller agent exchange offers and counteroffers.

The buyer always remains in control and makes the final purchase decision.

Once an offer is selected, AgentDeal uses PayPal Checkout in Sandbox mode to demonstrate the payment flow and tracks the resulting transaction and deal status.

How we built it

AgentDeal was built as a full web application using Python, React, TanStack Start, Vite, TypeScript, Tailwind CSS, Supabase/Lovable Cloud, and AI-powered server functionality.

The AI layer handles natural-language understanding, structured requirement generation, seller matching, offer evaluation, recommendation explanations, and AI-assisted negotiation.

We created a controlled set of clearly labeled Demo Sellers so the prototype could demonstrate the complete agent workflow without pretending to represent real freelancers or claiming to search the entire internet.

For payments, we integrated PayPal Checkout using the PayPal Sandbox environment. The flow includes order creation, buyer approval, payment capture, and transaction tracking.

Challenges we ran into

One of the biggest challenges was making AI agents interact in a structured and predictable way while keeping the experience natural.

Negotiation required clear state management for original offers, counteroffers, pricing, delivery times, revisions, and final terms.

Another challenge was integrating payment into the product flow while keeping the prototype safe and honest. We used PayPal Sandbox and test money rather than real transactions.

We also had to carefully balance automation with user control. We wanted AI to do the hard work without making important purchasing decisions on behalf of the buyer.

Accomplishments that we're proud of

We are proud that AgentDeal demonstrates an end-to-end AI procurement workflow rather than simply adding a chatbot to a marketplace.

The prototype connects:

Natural-language request → AI brief → intelligent matching → transparent offer comparison → AI-to-AI negotiation → human approval → PayPal payment → deal tracking

We are especially proud of the AI-to-AI negotiation concept because it explores how future commerce could work when software agents can communicate and negotiate on behalf of humans.

We also made transparency a core part of the experience by showing why an offer is recommended instead of hiding the decision behind an unexplained AI score.

What we learned

We learned that building a useful AI product requires much more than connecting an LLM to an interface.

AI needs structured inputs, clear state, reliable application logic, transparent decision-making, and meaningful actions.

We also learned that agentic systems should be designed around human control and trust. Automation is powerful, but users should understand what the AI is doing and remain in control of important decisions.

Finally, integrating PayPal taught us how payment infrastructure can become a natural part of an AI-driven product workflow rather than being treated as a separate feature.

What's next for AgentDeal

The current prototype uses controlled Demo Seller Agents. Our next step is to expand AgentDeal into a larger AI procurement network where businesses can connect with verified service providers and specialized provider agents.

Future versions could include:

  • Real provider and business onboarding
  • More advanced AI-to-AI negotiation
  • Automated contract and scope generation
  • Milestone-based transactions
  • More payment and payout options
  • Provider reputation and verification
  • Cross-platform service discovery
  • Autonomous procurement workflows for businesses

Our long-term vision is simple:

Tell AI what you need. Let it find the deal.

Agents negotiate. Humans decide.

Built With

Share this project:

Updates

Submission history