Tony Yu

tom found a free AI that edits podcasts better than a $75 editor, so two weeks later he and tony are running growth for the company that made it. here’s everything from episode 23 (sponsored by Ahrefs mcp) the tab origin story is the whole pod in miniature. tom’s tech guy in bangladesh told him about tab, tom said “I don’t need another agent, I have instinct,” tried it anyway, and the video editing was insane. frame-by-frame copy of any style you feed it, free. he posted about it, saw a mutual friend stafford post about it too, DMed him asking if he knew the founder, and the answer was “bro, that’s my company.” now tom and tony handle digital growth. this episode was edited by tab: it logged into riverside, pulled the files, cut it, made the thumbnail, chopped ten shorts, and posted. three days of editor back-and-forth, gone. why it’s free and why it works: the industry is seeing agent recommendations convert at 25%+. a website click converts at 1-5% on a good day. own the conversation inside the agent and even a 1.5-2.5% clip on every transaction is massive, which is why the VCs are funding free harnesses. tony’s take: it’s not zero-sum, there’ll be multiple winners, but nobody wants to be the AOL of agents. only 3% of people pay for AI, so the normie market is the whole game, and the normie won’t download an app. a phone number in iMessage is the lowest friction possible. the hook writer fight was the funniest segment. tom wants to hire a creative strategist on upwork at $100 a script. jacky physically reacted like he saw a ghost. tom’s defense: brands have five creative strategists at $20K a month each, it’s the hottest role in DTC, and jacky lives in the stone age of direct response. tony took tom’s side: writing skill has gone the other way and a real storyteller is worth hiring. jacky’s tier list for 2026: artists and marketers with taste at the top, autistic indie hackers at the bottom, and “I’m somewhere near the bottom, McDonald’s is calling.” the real workflow they built for tab is the template for every brand. a tool pulls the top-performing hooks of the day by GMV and watch time, tab breaks each ad into its exact structure, repurposes it into their pillars, writes the brief, then they approve in a tinder-style swipe interface. approved briefs upload to tribe, creators film, and there’s an auto-upload to meta. tony’s line: “we don’t do anything anymore.” he’s also dropping tab into every company group chat to produce a 1-2 minute video report of what everyone did that day at 4:30pm. the data industry got a second look. tom talked to a data cleaning company and the value chain is clearer: stripping PII is table stakes, the money is in hiring lawyers and engineers to comb the data so every thread makes contextual sense. then a 5-10% human audit, and if the error rate is too high the whole database gets recleaned, like QC on physical goods. the group’s speculation on broker margins: whatever you get paid, times four. tom’s running an experiment with a friend’s construction company to find out the real unit economics. restaurants are worthless because everything’s verbal. law firms are gold. clio’s data might be the most valuable asset in vancouver. rank surge hit $10K MRR with the 7-day trial funnel. $99/$197/$297 self-checkout by keyword count, auto-enroll into $497/$697/$997 monthly, zero salespeople. trial-to-retainer conversion is 67%. tom says the ads still suck and cost per sale hit $500 yesterday, but it converts anyway. “this idea is easily a mil a month and it makes me very uncomfortable.” tony’s response: let me start a growth team for you. bonus: the PHP vulnerability is one of the biggest of all time, 70% of the internet runs on the affected version, railway got hit, and the patch landed today. the reason attackers move so fast now is jailbroken deepseek: an opus-level model that never refuses. jacky’s defense plan is to run his own red team with it. the business opportunity: AI security audits all day. and the holy trinity of consumer AI according to twitter is banker, yapper, and autist. tony volunteered as banker. “this is us.” the new channel is youtube.com/@NGMIthepodcast. NGMI with @itstonyyu and @tomwang24. tom filmed at 11pm, tony from a hotel lobby in france, jacky in asia. watch/listen ↓

Here's the new Claude experience on the iPhone Duo! It's going to be amazing for multitasking, Claude Code, and creating new things. Looking forward to see folk start using it tomorrow!

The best AI assistant nobody knows about yet: tab.bot Enjoy the secret while it lasts. It'll be everywhere in your feed soon. cc @itstonyyu

We’re releasing a broad range of new mathematical results produced by an internal frontier model. We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results. github.com/openai/math

Claude Dashboards and Claude Motion are in beta today. Ask Claude to turn your data into live dashboards and your ideas into animated explainers.

Ben Affleck (Hollywood star & Artists Equity CEO) talks about how he fine-tunes open video models by unfreezing weights and trained only the last cinematic layer so a film crew can hit real production standards. for context, Ben Affleck founded InterPositive in 2022, a 16-person AI shop for film post and Netflix bought it in March 2026 for $587 mn in cash. He needed that model because public video models were trained on his peers' films, and he did not think that was a real business. So InterPositive raised money, shot its own dataset for 8 months on a controlled stage, and used it only as late-stage training. Each new film then trains a private model on its own dailies, so the production keeps the footage and the learning. That is the product Netflix paid $587 million for. ---- From "Bloomberg Live" YouTube channel, (link in comment)

Introducing Griffin, the first model to pass the video Turing test. 48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video. It’s the first Human Interaction Model (HIM).