Project ideas from Hacker News discussions.

eBay explicitly bans AI "buy for me" agents in user agreement update

πŸ“ Discussion Summary (Click to expand)

Based on the Hacker News discussion, here are the three most prevalent themes regarding eBay's updated user agreement prohibiting LLM agents from making purchases.

1. Financial and Operational Liability

The primary motivation for the ban is believed to be eBay's desire to shield itself from the financial and operational costs associated with AI-driven errors. Users suggest that LLM agents are more prone to "hallucinating" purchases or making mistakes that lead to chargebacks and support calls, costs eBay prefers not to absorb.

"LLM-initiated purchases probably rack up chargebacks, support calls, etc for mistakes the LLM makes. I'm not surprised they want to limit it." β€” advisedwang

"Ebay doesn't want to deal with charge backs for hallucinate purchases" β€” wobblyasp

"For me it seems a protective measure, the headaches of AI hallucinations leading to purchases you have to sort out (customer support, returns etc.)" β€” wobblyasp

2. Commercialization and Control Strategy

Many commentators argue the ban is a strategic move to maintain control over eBay's ecosystem and create new revenue streams. By restricting LLM access, eBay could force third parties to pay for licensed API access or ensure that any AI shopping capabilities are provided exclusively by eBay itself.

"More likely, they want to be the exclusive provider of LLMs that can purchase off of eBay, or at least charge for API access." β€” doctoboggan

"This; 'certified / authorized by eBay' and then agents have to pay access to the catalogue" β€” rvnx

"sniping bots keep people on ebay.com" β€” theamk

3. Skepticism Regarding Enforcement and Effectiveness

There is widespread doubt that the rule is actually enforceable or will significantly deter automation. Users argue that determined actors can easily mask their bot activity to appear human, rendering the policy largely symbolic or useful only for after-the-fact disputes rather than prevention.

"Impossible to enforce, they can read browser windows and pass captchas" β€” yieldcrv

"Probably less about direct enforcement, more about after the fact." β€” wobblyasp

"Sure, the cost of that goes way up though, especially if it has to emulate real world inputs like a mouse, type in a way that’s plausible, and browse a website in a way that’s not always the direct happy path." β€” drum55


πŸš€ Project Ideas

EBay SnipeBot Sentinel

Summary

  • A tool that helps everyday bidders beat the sniping bots by intelligently calculating the optimal "last millisecond" bid and its precise cent amount.
  • It levels the playing field for casual users who feel cheated by automated sniping tactics, allowing them to participate in auctions fairly without manual stress.
  • Core value proposition: Democratizes the sniping strategy, turning it from a bot-exclusive advantage into an accessible, user-controlled feature.

Details

Key Value
Target Audience Casual eBay bidders frustrated by last-second losses to snipe bots.
Core Feature Calculates the "cent-delta" (e.g., $X.01) to outbid the current leader's likely maximum based on auction dynamics.
Tech Stack Python (for logic), Headless Browser (Puppeteer/Playwright), Cloud Scheduler (AWS Lambda/Google Cloud Functions).
Difficulty Medium
Monetization Hobby (Open source, or "Freemium: Free for 5 alerts/month, Pro subscription for unlimited").

Notes

  • HN commenters frequently debate sniping: "Snipers essentially convert the ascending-bid proxy auction... into a Vickrey second-price sealed bid auction" (young_rutabaga). While many argue "just bid your max," others admit "I found it distasteful and irrational but it works so I put up with it" (Fwirt).
  • This tool validates user sentiment that sniping is a superior strategy and provides a free/cheap way for non-developers to utilize it, addressing the frustration of losing by pennies.

TruScout AI Agent

Summary

  • An "AI agent for hire" that searches eBay and other marketplaces using natural language but waits for human review before purchase, bridging the gap between automation and control.
  • It solves the fear of AI "hallucinating" purchases (e.g., buying a "box only" listing) while still automating the tedious search process for obscure items.
  • Core value proposition: Delivers the efficiency of AI search with the safety of a human-in-the-loop, specifically designed to comply with strict "no auto-buy" policies like eBay's.
Key Value
Target Audience Collectors and hobbyists looking for specific, hard-to-find items (e.g., vintage electronics, niche parts).
Core Feature Natural language search parsing (e.g., "1982 Stratocaster under $600 in good condition") with a "Cart for Review" interface.
Tech Stack LLM (OpenAI/Claude API), Vector DB (Pinecone/Milvus), Next.js/FastAPI.
Difficulty Medium
Monetization Revenue-ready: Subscription ($10/mo) or Pay-per-alert.

Notes

  • HN users expressed interest in AI monitoring but not auto-buying: "I don't really see any value in having the AI do the purchase itself though" (__jonas). Users are paranoid about errors: "What if the electronic parrot buys from an obvious counterfeit vendor?" (Yizahi).
  • This tool directly addresses the "fear of autonomous spending" while solving the "search fatigue" mentioned in the thread.

LLM Arbscaler

Summary

  • A tool for power sellers and dropshippers to automate arbitrage by scanning unstructured listings for pricing errors or undervalued items using LLM vision and text analysis.
  • It solves the difficulty of spotting "fat finger" pricing or mis-categorized items at scale, which human review cannot catch quickly enough.
  • Core value proposition: Turns the "unstructured chaos" of eBay listings into a structured opportunity feed for high-volume resellers.
Key Value
Target Audience Power sellers, dropshippers, and arbitrage bots (B2B).
Core Feature LLM-powered semantic analysis of images and text to flag items priced significantly below market value or "swap meet" deals.
Tech Stack Multimodal LLM (GPT-4o/Gemini Vision), Proxy Rotators, Database (PostgreSQL).
Difficulty High
Monetization Revenue-ready: Tiered API access (Pay-per-scan) or SaaS subscription.

Notes

  • HN users acknowledged the scale of automation: "Agents are being used to automate things like... purchase-refund arbitrage schemes at an increasingly large scale" (Nkharrl).
  • One user noted, "I run up the prices in less competitive auctions just for fun occasionally" (blitzar). This tool helps professional arbitrageurs beat the "fun" runners by detecting deals instantly.

Vault: API-First Agentic Commerce Gateway

Summary

  • A browser extension or local middleware that acts as a "human-in-the-middle" proxy, allowing LLM agents to interact with sites like eBay without violating ToS (visually) while logging the action for user approval.
  • It solves the "impossible to enforce" detection problem mentioned in the thread by making the agent look like a real user (via real browser inputs) but ensuring the user retains final click authority.
  • Core value proposition: Enables agentic workflows on locked-down platforms by technically complying with "no automated orders" rules via user-verified input simulation.
Key Value
Target Audience Developers building personal AI assistants and power users.
Core Feature A local server that intercepts AI agent commands, renders the page in a real browser, and waits for a human "click" or confirmation keypress.
Tech Stack Electron (for the wrapper), WebSocket (for communication), Local Browser Automation.
Difficulty High
Monetization Hobby (Open Source) or "One-time purchase license".

Notes

  • HN user "drum55" noted eBay is "hyper aggressive about fingerprinting," while "Nextgrid" countered that "Multimodal models with vision input can operate a real computer... where the computer itself returns a real, plausible browser fingerprint."
  • This project bridges that gap, using a real browser fingerprint (as per Nextgrid) but injecting the "human approval" step to avoid the "charge backs" and "support calls" (advisedwang) that eBay fears.

Second Price Auction Simulator

Summary

  • A browser extension that visually enforces "proxy bidding" behavior on users, acting as a psychological shield against the emotional urge to bid incrementally or panic-snip.
  • It solves the "nibbling" behavior and emotional frustration of auctions by simulating a sealed-bid environment (like Google Ads or Treasury auctions) directly on the eBay UI.
  • Core value proposition: Helps users stick to their "hard limit" by removing the real-time bidding interface and only showing "You are winning" or "You have been outbid" statuses without the price anxiety.
Key Value
Target Audience Rational bidders frustrated by emotional bidding wars and "nibblers."
Core Feature UI overlay that hides the "Current Bid" and "Place Bid" buttons, only allowing a single "Max Bid" input and sending the bid 5 seconds before the end.
Tech Stack Browser Extension (Manifest V3), JavaScript.
Difficulty Low
Monetization Hobby (Free).

Notes

  • The HN thread contains extensive debate on price ceilings: "Sniping removes the 'contemplation window' to reconsider your bid" (dingaling). Another user noted, "People's competitive behavior... take over. People's railing against sniping also demonstrates this" (nutjob2).
  • This tool addresses the "Sorites paradox" (direwolf20) and "irrational" behavior (krackers) by enforcing a strict, pre-defined max bid, effectively automating the "ethical" bidding strategy.

Read Later