3 questions founders actually ask, each with a straight answer and the resources worth your time.
The typical workflow: write the script and shot list with ChatGPT or Claude, generate 8-second clips in Google Veo or Runway, stitch them together in any editor, and add voiceover with ElevenLabs, while HeyGen turns one photo or short recording of you into a talking-head avatar for demos and UGC-style ads. The famous proof point is Kalshi's NBA Finals commercial: one filmmaker, a couple of days, about $2,000. One caution on tools: OpenAI shut down its Sora app in April 2026, so most founders today build on Veo, Runway, or Kling.
The filmmaker behind the Kalshi ad screen-shares his exact script-to-Veo-to-edit workflow, the single best walkthrough of how AI ads actually get made.
Open youtube.com →The practitioner who made primetime AI ads condenses his playbook (concept, prompting, editing) into a free step-by-step post.
Open linkedin.com →A B2B SaaS founder's honest first attempt at a Veo 3 ad, real prompts, real iterations, real cost, ideal for founders starting from zero.
Open peersignal.org →A current, click-by-click guide to the product-demo use case specifically: reference images, prompt settings, scene extension, and iteration.
Open kaizenaiconsulting.com →
Hands-on tutorial for the avatar/UGC-ad style of AI video, upload product shots, pick an avatar, generate voiceover, export, in minutes.
Open youtube.com →Everything an agency used to sell, actors, sets, cameras, editors, voice talent, is now a software subscription plus a prompt, so one person can go from idea to broadcast-quality ad in days for a few thousand dollars instead of months and $100K+. Kalshi's NBA Finals spot was made by a single filmmaker for roughly $2,000, and small AI-native teams are now running seven-figure ad businesses. The scarce ingredient has shifted from budget and crew to taste, ideas, and how fast you iterate.
The definitive telling of the $2,000 Kalshi NBA Finals ad, the moment one person plus AI visibly replaced a full production agency.
Open npr.org →Deep interview on how a tiny AI-native team does work that used to need a studio, including his image-first workflow and why imagination, not tech, is the bottleneck.
Open thecreatorreport.com →The big-picture data: creative workflows collapsing from weeks to hours, and how chaining a handful of models gives small teams enterprise-grade output.
Open a16z.com →Concrete example of the one-person media stack: an avatar that records videos for him, voice-to-content systems, and the workflows gluing it together.
Open x.com →What genuinely works today: short fast-cut clips, comedic or stylized ads, avatar talking-heads, AI voiceovers, and product b-roll. What still gives it away: keeping the same character consistent across shots, long continuous takes, hands, on-screen text and logos, lip-sync, and sincere emotional storytelling, and research shows consumers notice these flaws far more than marketers think. The practical rule: lean into what AI does well (absurd, funny, quick cuts) or blend it invisibly into real footage; don't try to fake a heartfelt human testimonial.
Real ad-testing data: seamless AI blends outperform, visibly-AI ads underperform, and 41% of consumers are bothered versus only 29% of marketers who think they are.
Open kantar.com →Sharp teardown of big-brand AI ads (including Coca-Cola's) explaining exactly which artifacts and workarounds audiences detect and resent.
Open nngroup.com →First-person experiment building UGC-style ads with Veo 3, honest notes on what looked natural (expressions, gestures) and what broke (sync, prop consistency).
Open aishaimtiaz1.substack.com →Uses OpenAI's April 2026 Sora shutdown to explain where AI video still falls short economically and creatively, useful calibration before betting your brand on it.
Open techxplore.com →