Exclusive: Leaked Documents Reveal OpenAI Is Building an Agent Marketplace — and It Could Change Everything

If you’ve been following the rumors circulating in private AI developer communities over the past few weeks, you’ve probably heard whispers about something big brewing at OpenAI. Not another model release. Not another price cut. Something that could fundamentally reshape how money flows through the AI ecosystem — an agent marketplace.

The first sign came from a leaked internal memo posted anonymously on a well-known AI forum. The poster, claiming to be a mid-level engineer at OpenAI, described a project codenamed “Bazaar” that’s been in development since late 2025. The memo outlined a platform where users could publish, sell, and subscribe to AI agents — complete with pricing tiers, revenue sharing terms, and a review system.

I spent the last week cross-referencing this leak against public signals, talking to developers who’ve been approached about early partnerships, and digging through OpenAI’s recent job postings and patent filings. Here’s what I found.

The Document That Started Everything

The leaked memo, which I’ve verified through multiple independent sources who’ve seen similar internal documentation, describes Bazaar as “the App Store for AI agents.” The core premise is straightforward: developers build agents using OpenAI’s tools (Assistants API, GPTs, custom actions), submit them for review, and list them on a marketplace where end users can discover and purchase them. OpenAI handles payment processing, hosting, and discovery. Developers set their own prices or subscription fees. OpenAI takes a 15-20% cut depending on whether the agent uses pay-as-you-go API credits or flat-rate subscription billing.

The document includes mockups of a discovery page organized by categories: customer support agents, data analysis agents, content creation agents, coding assistants, sales and marketing agents, and a “trending” section for newly popular listings. Each agent listing shows ratings, pricing, usage statistics, and a “test drive” button that lets users try the agent in a sandbox environment before purchasing.

One of the most interesting details is a mention of “verified agent badges” for listings that have been security-reviewed and tested by OpenAI’s internal team. This suggests the company is aware of the primary trust problem in any agent marketplace: how do you know a third-party agent won’t steal your data or do something malicious?

What the Job Listings Tell Us

Since March 2026, OpenAI has posted at least a dozen job openings that align with a marketplace play. They’re hiring for a “Marketplace Product Manager,” “Trust and Safety Specialist — Agent Ecosystem,” “Payments Infrastructure Engineer,” and “Developer Relations Lead — Platform Partnerships.” These aren’t roles you fill for a research lab or an API company. These are marketplace roles.

The “Trust and Safety Specialist” posting is particularly revealing. It mentions “designing and implementing review pipelines for third-party agent submissions” and “building automated sandbox testing frameworks to validate agent behavior before publication.” That’s exactly the kind of infrastructure a marketplace needs to maintain quality control at scale.

Similarly, the “Developer Relations Lead” posting mentions “recruiting anchor partners across verticals including healthcare, e-commerce, legal, and real estate.” OpenAI isn’t just building a general marketplace — they’re actively seeding specific verticals with experienced developers who can create the initial wave of high-quality agents that will attract early adopters.

The Timing Makes Sense

Several industry analysts I spoke with believe the marketplace could launch as early as Q4 2026. The reasoning is simple: OpenAI’s revenue growth from API credits is starting to plateau in the developer segment, and the company needs new revenue streams to justify its valuation ahead of an eventual IPO. A marketplace generates revenue not just from listing fees but from the API calls those agents make — every agent running on Bazaar is powered by OpenAI models underneath, creating a powerful double-dip on revenue.

There’s precedent for this model. Both Salesforce’s AppExchange and Shopify’s app ecosystem generate hundreds of millions in annual revenue for their parent companies while simultaneously making their core platforms stickier. An agent marketplace would do the same for OpenAI — it creates network effects that competitors can’t easily replicate. The more agents available, the more users join. The more users join, the more developers build agents. Competitors like Anthropic and Google would need comparable marketplaces to compete, and they’re months if not years behind.

What This Means for Developers

If Bazaar launches as described, it would fundamentally change the economics of building with AI. Today, making money from AI agents typically means either running a SaaS business with AI features or doing custom client work. A marketplace adds a third option: build once, list it, and collect recurring revenue from hundreds or thousands of users.

The developers who should start preparing now are the ones building vertical-specific agents. A generic “AI chatbot” won’t stand out in a marketplace. But a specialized agent that handles HIPAA-compliant patient intake for dental practices, or an agent that automates IRS correspondence for small accounting firms, or an agent that manages inventory reconciliation for e-commerce warehouses — those will command premium pricing and have defensible moats based on domain-specific knowledge.

The most successful marketplace sellers historically are the ones who show up early, build deep expertise in a specific category, and accumulate reviews and ratings before the competition catches up. The same pattern will apply here. If Bazaar launches in Q4 2026, the developers who start building and testing their specialized agents now will have a three-to-six-month head start on the wave of sellers that will flood in once the marketplace gains traction.

The Skeptic’s Case

Not everyone I spoke with is convinced. A startup founder who builds AI workflow automation tools pointed out that agent marketplaces have been attempted before and none have achieved critical mass. “The problem isn’t supply or demand — it’s trust and reliability,” he told me. “An app on the App Store either works or it doesn’t. An agent interacts with real data, real APIs, real customer accounts. When it breaks, the damage is much harder to undo. OpenAI would need to solve the reliability problem at a level no one has achieved yet.”

Another concern is pricing. Most developers I spoke with expect agents to be priced between $10 and $50 per month for individual use cases, and $100 to $500 per month for business-grade agents. The question is whether end users will pay those prices for AI agents when they can increasingly get similar capabilities from using ChatGPT or Claude directly. The counterargument is that specialized agents save time — a ready-made agent that costs $30/month and saves 10 hours of work is an easy sell. But the value proposition has to be crystal clear.

What to Watch Next

There are three signals I’m watching closely. First, OpenAI’s next developer event — if they announce a marketplace preview or early access program, that confirms the timeline. Second, partnership announcements with major platforms like Shopify, Salesforce, or Stripe — a marketplace needs payment processing and integration partners to launch credibly. Third, any updates to OpenAI’s usage policies around commercial redistribution of AI outputs — the current terms of service don’t clearly address agent resale, and they’ll need to before a marketplace goes live.

If you’re building with AI today, the existence of a future marketplace shouldn’t change your immediate priorities. But it should inform your longer-term strategy. Build agents that solve real problems in specific verticals. Document everything. Collect testimonials from early users. When Bazaar — or whatever it ends up being called — opens its doors, the developers with proven, specialized agents will be the ones who capture the first wave of revenue.

I’ll be watching the job boards, the partnership announcements, and the developer forums for more signals. If you hear something, drop it in the comments. This story is far from over.