AI Video Agents and Vibe-Coded Businesses: What Is Real This Week

Andrew Warner and Corey Ganim's weekly roundup has two stories running through it. Creative agents can now move a recording closer to a finished post, while small builders use AI to make narrower alternatives to software people already buy. Neither story means a one-prompt replacement for an editor, a profitable app or a proven market. The useful question is which repeated job becomes easier after you account for review, costs and distribution.

Watch the Episode

Source note, 30 October 2026: the video and supplied transcript describe what Andrew and Corey saw or tried. Product pages, repository READMEs and original posts linked below were checked for availability, pricing and scope. We did not reproduce the edits, build the featured apps or verify a buyer's outcome. TrustMRR figures are dated platform snapshots, not audited profit or a forecast. Zapier sponsored the episode.

The Tools by Job, Not by Hype

A recording workflow, an agent connector, a research voice model and a revenue marketplace should not be compared as if they solve the same problem.

Video Production Is Becoming a Directed Workflow

Riverside: four agents, one recording pipeline

Andrew is excited because his team already uses Riverside for interviews. The October 26 launch names four specialist agents: Producer, Editor, Clipper and Marketer. Riverside says they can prepare a session, draft the full cut, identify short clips and prepare distribution. Its product walkthrough says edits and posts require the user's approval. The agents are available across plans, but capabilities and AI-credit use vary. That reduces handoffs; it does not prove that the first cut matches a skilled editor's taste.

Test one completed recording. Measure minutes from raw file to an approved full cut and two usable clips; count human corrections. If those take as long as the old edit, the bottleneck has moved rather than disappeared.

Open the original Riverside post on X

Design skills, promo videos and visual context

The AI Design Skills repository contains a landing-page skill with intake, conversion copy and a visual system. It can make an agent's first draft more coherent; it cannot tell you whether the promise is true or whether visitors will convert. Corey also shows an open-sourced promo-video skill, but the public description does not identify its repository. Rather than attach an unrelated repo to that clip, use the 6:21 demonstration as a reference and inspect the original creator's source before installing anything it mentions.

Lasso solves a smaller but common problem: showing an agent exactly which screen region you mean. Its creator's release post says Lasso 1.3 can share and annotate screen context with Claude Code, Cursor and Claude Desktop on Mac. That is useful for UI review, but a screenshot can include customer data or private tabs. Keep a clean test desktop and inspect what is shared.

Open the Lasso demonstration on X

Suno Can Be Connected Today; Dessa Is a Preview

Suno's official MCP page documents a connector for Claude, ChatGPT, Claude Code, Cursor and other supported clients. It says songs save to your Suno library and use your Suno credits, so the connector is not an unlimited free music API. Andrew's family-name song is a personal example; the hosts' personalized sales-song idea is a hypothesis. Before using a prospect's details in generated audio, get permission to use that data and check for awkward or invented claims.

Block's Dessa is a separate story. The hosts react to a speech conversation sample, and Jack Dorsey's announcement describes a full-duplex speech model. On the checked public page, demos and a route for safety researchers were visible, but not a general download, model license or API. Do not turn “runs locally” in a demonstration into “you can run it now.” Voice realism also raises consent and impersonation questions that a polished sample cannot answer.

Open Jack's Dessa post on X

A Cheaper Build Is Only One Part of a Business

DataFast, TrustMRR and the copycat question

DataFast sells revenue-first analytics. Andrew points to a lower-priced tool inspired by the same buyer frustration and asks what happens when agents make software cheaper to create. The lesson is not "copy a product and charge less." A buyer still needs accurate attribution, integrations, privacy, maintenance and support. The original product's public TrustMRR listing was around $29,000 MRR when checked, but that is a changing revenue snapshot and says nothing about an imitator's retention.

TrustMRR is not just a leaderboard. Its FAQ explains read-only payment-provider verification, while its API terms restrict cloning and bulk republication. The platform's 200th-acquisition post is a milestone the hosts discuss, not proof that a new marketplace will acquire the same network effects. Marc Lou's own account says the marketplace emerged after users asked for it. That is the durable lesson: follow demand visible in real behavior.

Open TrustMRR's milestone post on X

Open Marc Lou's post on X

GodsView AI and Nomback: prove the narrow job

The hosts compare GodsView AI with the separate open-source God's Eye View. The repo offers a globe with public data layers; the hosted product sells its own experience. They are not the same project, and similarity alone does not establish code copying. The GodsView listing showed about $5,200 MRR on 30 October. That suggests customers pay for some combination of access and convenience, but the listing does not tell us margins, acquisition cost or how much the open-source project contributed to demand.

Nomback is more concrete: a branded direct-ordering app for a restaurant's regulars. The founder's original build discussion explains the starting point. The site shows Anthony's Pizzeria as a live example and lists $99 per month plus $399 setup for an app when checked. The site says pickup or delivery with the restaurant's own drivers, which is more precise than assuming every order is fulfilled through DoorDash. The test is whether enough regulars switch from a marketplace to direct ordering to outweigh app setup, payment, fulfillment and support costs. Andrew's agent-checkout idea is a future product direction, not a Nomback feature verified here.

Connecting and constraining agents

The episode is sponsored by Zapier. Its MCP product connects agents to business applications with managed permissions. The pricing page lists a free tier, but the amount of work and access depend on the plan. Start with read-only data and one narrowly scoped action; draft messages before sending. Andrew also points to a shared spam-call and personal-information-removal agent. Such tasks can reach sensitive records and third-party sites, so review every request and do not assume an agent can correctly remove all copies of personal information.

Open the original templates post on X

Corey closes with a short video recap of a long internal plan; Andrew shows visual status cards for agents and edits of family and fitness footage. These are useful workflow examples, not new products with documented release terms. A two-minute recap is a navigation aid, but it should link back to the full plan so decisions and omissions remain inspectable.

A First Business Pilot: One Buyer, One Costly Job

Our practical rule is to begin with the repeated job, not the exciting tool. If the buyer already has a workable solution, a cheaper clone must beat its total ownership cost and switching friction. The prompt below is an original interview-and-pilot brief for that decision; it is not from the video and has not been run against a real client.

Video Chapters

Video, design and software economics

Connected agents, audio and marketplaces

Restaurant app and personal workflows

Sources and Related Viewing

Checked 30 October 2026. The episode and transcript are the source for the hosts' demonstrations and opinions. The linked launch pages and repositories describe the products themselves. X posts are primary announcements or creator examples, not independent validation of performance. Prices, access and revenue can change after this date.

For a deeper look at the creator side, these two related demonstrations show different workflows rather than more headlines.