Andrew Warner and Corey Ganim's The Next New Thing episode moves quickly: a reverse-engineering repo, a plugin marketplace, an agent idea directory, a new voice model, a video workflow, and a pixel-ad site. The useful question is not which demo looks most impressive. It is which one has a buyer, a distribution route, and a first test you are actually allowed to run.
There are five business directions here, with Grok Bot's connectors and Grok Voice grouped into one platform opportunity. REA and Hypit are real open-source projects; Grok Bot has a reviewed plugin submission path; the use-case directory is live; and the pixel-ad site has a public counter. None of that proves a repeatable business. In particular, the episode's "$37,000 in five days" line is not an independently verified profit figure.
Watch the Episode
Credit: Andrew Warner, Corey Ganim, and The Next New Thing for the episode and examples. Zapier sponsored the video; see Zapier. This article adds direct project links and separates their business speculation from verified product capabilities.
1. REA: Rebuild an Authorized Workflow
At 00:18, the hosts introduce REA (Reverse Engineer Anything). Its MCP tooling can help an agent inspect application behavior and, with supported analysis tools, compiled binaries. The official guides and showcases explain the setup. The episode's supporting Eric Tech demonstration shows a Wispr Flow-style reconstruction; it is a demo, not proof that every feature, integration, or edge case was reproduced.
Business test: ask a buyer about one expensive workflow trapped in software they own or are authorized to inspect. Deliver a small replacement or migration prototype and compare task completion, data portability, and maintenance cost. REA's MIT license covers REA, not the target application's source code, brand, assets, or license terms. "Copy any software for free" is a headline, not a legal or technical guarantee.
2. Grok Bot: Useful Plugins, Then Voice
At 01:54, Warner and Ganim discuss the Grok Bot marketplace. The official submission form accepts an MCP server URL or public GitHub repo, asks for testing instructions, and says submissions will be reviewed. Grok Bot and Cursor are listed; Grok itself is marked "coming soon." Corey's AI-assessment connector is described as an idea he plans to submit, not an approved marketplace listing.
Business test: build one narrow, permission-scoped connector for a workflow a real customer already uses, then test it privately before relying on marketplace discovery. For example, an assessment assistant could turn an authorized interview transcript into a reviewer draft, but it should not ingest client data or send recommendations without consent and review. A Zapier trigger may be enough for a small first pilot.
At 06:32, they pivot to voice receptionists. xAI announced Grok Voice 3.0 for its Voice Agent API at $0.08 per minute as of October 2026. That is a model/API price, not a complete phone service: telephony, testing, supervision, escalation, and privacy controls add work and cost. Start with after-hours FAQs or message capture, with a clear route to a person.
3. Agent Use Cases: Sell Curation, Not a List
Agent Use Cases, created by Alex Finn, collects public examples and prompts by tool and topic. The site listed 1,342 use cases when checked on October 16, updated October 15. In the episode's 04:20 segment, the hosts discuss it as both an idea source and a possible directory business. The site's examples are discovery leads, not validated customer outcomes; the hosts' revenue estimate for the directory is also not independently checked here.
Business test: curate ten high-quality workflows for one industry, verify the original sources, test two with willing users, and publish what failed as well as what worked. Charge for implementation, training, or a maintained workflow library only after the audience demonstrates a recurring need. A generic scraped catalog is easy to reproduce; useful judgment and maintenance are harder.
4. Hypit: Reuse a Format, Create Original Media
At 07:48, the hosts show Hypit. Its open-source repository describes a workflow for analyzing a reference video's shots, pacing, captions, and effects, then assembling a new version. It can also start from a template or a fresh idea. The tool is free to use; coding-agent and optional generation-model costs are separate.
Business test: make three original short videos for a client's real offer, using a proven structural pattern but licensed footage, approved music, and consented voices or faces. Measure qualified enquiries or watch-through, not only views. A familiar format can improve production speed; it does not guarantee virality and does not grant rights to reuse another creator's media or likeness.
5. Pixel Ads: Distribution Is the Product
At 09:50, the episode covers 1 Million Pixels (note the three l's in its domain), a simple site selling limited ad space. The appeal is the product's instantly understandable constraint. The hard part is attracting buyers and traffic, preserving ad quality, and knowing whether a burst of curiosity becomes repeat business.
The episode says "$37,000 in five days." The site's own public stats endpoint reported $30,795.94 in cumulative revenue when checked on October 16, 2026. That counter is first-party and may change; it does not expose an audited five-day window, refunds, acquisition cost, or profit. The transcript also moves from this project to another leaderboard example, so those stories should not be merged into one proven playbook.
Business test: choose a niche audience you can already reach and pre-sell a small, clearly disclosed placement with an expiry date, traffic report, and refund rule. The offer is access to attention, not a grid of pixels. If you cannot reach advertisers or demonstrate useful exposure, building the page is not the bottleneck.
Which One Is Worth Testing?
These examples point to different buyers. REA suits an authorized migration or internal-tool customer. Plugins and voice fit teams with a repetitive workflow and someone accountable for permissions. A directory needs a trusted niche audience. Hypit needs a real creative brief and rights-cleared assets. Pixel ads need distribution before code. Choose based on a customer conversation and a measurable first result, not the size of the model or the most viral screenshot.
Jump to the Episode
| Time | Segment |
|---|---|
| 00:00 | Five launches and business ideas |
| 00:18 | REA explained |
| 00:52 | Rebuilding expensive apps |
| 01:54 | Grok Bot connector marketplace |
| 02:51 | Product as an agent plugin |
| 03:39 | AI assessment connector idea |
| 04:20 | Agent Use Cases |
| 05:03 | Directory business model |
| 06:32 | Grok Voice 3.0 |
| 07:48 | Hypit video formats |
| 08:45 | Short-form distribution |
| 09:50 | 1 Million Pixels |
| 10:24 | $37,000 claim |
| 11:04 | Simple ideas and traction |
| 11:45 | Business or short-term spike? |
| 12:34 | Closing opportunities |
Business Idea: Pick One, Prove One
This copyable prompt turns the episode's inspiration into a seven-day customer test. It works in a general AI assistant and does not require any of the featured products.
Test one AI business opportunity
A buyer, a first deliverable, and a stop rule.
Source Map and Limits
Episode and demo: Andrew Warner and Corey Ganim's discussion and Eric Tech's REA demo. First-party product sources: REA GitHub, xAI plugin submissions, Grok Voice 3.0, Agent Use Cases, Hypit GitHub and its official X profile, and 1 Million Pixels' public counter. Product descriptions and prices can change. The linked X posts in the video description were truncated, so no unverified post URLs or embeds are presented as primary evidence.