
Show Notes
๐ ๐ง๐๐ฐ ๐ฆ๐จ๐๐๐ฅ ๐ฌ๐ก๐ข๐ฉ๐ฉ๐๐, ๐๐ง๐ ๐ฐ๐ข๐ญ๐ก๐ข๐ง ๐ก๐จ๐ฎ๐ซ๐ฌ ๐ฉ๐๐จ๐ฉ๐ฅ๐ ๐ก๐๐ ๐๐ฎ๐ซ๐ง๐๐ ๐ญ๐ก๐ซ๐จ๐ฎ๐ ๐ก ๐ญ๐ก๐๐ข๐ซ ๐ญ๐จ๐ฉ ๐ญ๐ข๐๐ซ ๐ฉ๐ฅ๐๐ง. ๐๐ก๐ ๐๐๐ฎ๐ฌ๐ ๐ฐ๐๐ฌ ๐ง๐จ๐ญ ๐ ๐ก๐ฎ๐ง๐ ๐ซ๐ข๐๐ซ ๐ฆ๐จ๐๐๐ฅ. ๐๐ญ ๐ฐ๐๐ฌ ๐ญ๐ก๐ ๐ฆ๐จ๐๐๐ฅ ๐ช๐ฎ๐ข๐๐ญ๐ฅ๐ฒ ๐ฌ๐ฉ๐ข๐ง๐ง๐ข๐ง๐ ๐ฎ๐ฉ ๐ญ๐๐ง ๐จ๐ซ ๐ฆ๐จ๐ซ๐ ๐๐จ๐ฉ๐ข๐๐ฌ ๐จ๐ ๐ข๐ญ๐ฌ๐๐ฅ๐, ๐๐๐๐ก ๐๐๐๐๐ฎ๐ฅ๐ญ๐ข๐ง๐ ๐ญ๐จ ๐ญ๐ก๐ ๐ฆ๐จ๐ฌ๐ญ ๐๐ฑ๐ฉ๐๐ง๐ฌ๐ข๐ฏ๐ ๐จ๐ฉ๐ญ๐ข๐จ๐ง ๐๐ฏ๐๐ข๐ฅ๐๐๐ฅ๐. ๐ณ Nobody overspent. Something they delegated to did it for them, and nobody was asked. This week on ๐๐ก๐ ๐๐ ๐๐ซ๐๐๐ค๐๐จ๐ฐ๐ง, Quentin and Stark sharpen the theme: ๐๐ก๐ ๐ฌ๐จ๐๐ข๐๐ฅ ๐ฅ๐๐ฒ๐๐ซ ๐จ๐ ๐๐. For two years the mental model was a meter that ran while you typed. Now there is a second meter, running while nobody is typing, started by something you delegated to. Your agent is hiring, and the invoices come to you. Three signals your bill is now written by your tools: ๐ ๐ A single task quietly launched ๐๐+ sub agents, each on the priciest model, enough to exhaust a top tier plan alone
๐ Two credible testers measured the same model the same week and got opposite answers: one found tasks ๐๐% more token heavy, another found them cheaper
๐ A new video analysis approach cut token use up to ๐๐%, putting long footage nobody watched within reach of a real question The rule underneath: a cost per task number is a statement about the task, not the model. If you have not measured it on your own work, you do not know the cost. Your move: ๐ด Pick the one AI job your business runs most often. ๐ด Run it twice, once on the cheapest model you have and once on the most expensive. ๐ด Compare the cost of a finished, usable output, not the cost per token. One job, not your whole stack. Ten copies of the priciest model, working hard on a task nobody checked. Check the task. ๐๏ธ Stream on: โถ๏ธ Youtube: https://www.youtube.com/@SMBAIHUB โถ๏ธ Amazon Music: https://music.amazon.in/podcasts/b93fd52f-bc53-4ee9-98d7-3b9e10ea0e6f/ai-breakdown โถ๏ธ Apple Podcast: https://podcasts.apple.com/in/podcast/ai-breakdown/id6799578599 โถ๏ธ Web: smbaimagazine.com โถ๏ธ Web: smbaipodcast.com Disclaimer: The information shared in this podcast is for general informational purposes only and should not be considered as direct financial or business advice. Always consult with a qualified professional for advice specific to your situation. This is an Independent commentary. Third-party excerpts are credited on screen or in the show notes. Some voices in this program are AI-generated.
More from AI Breakdown
๐๐ง๐ ๐๐ฌ๐ญ๐ข๐ฆ๐๐ญ๐ ๐๐จ๐ซ ๐ฅ๐๐ญ๐ญ๐ข๐ง๐ ๐ก๐ข๐ฌ ๐๐ ๐๐ ๐๐ง๐ญ๐ฌ ๐ฎ๐ฌ๐ ๐ฌ๐จ๐๐ญ๐ฐ๐๐ซ๐ ๐ก๐ ๐๐ฅ๐ซ๐๐๐๐ฒ ๐ฉ๐๐ฒ๐ฌ ๐๐จ๐ซ: ๐๐ฌ ๐ฆ๐ฎ๐๐ก ๐๐ฌ $๐๐๐,๐๐๐ ๐ ๐ฒ๐๐๐ซ. ๐ณ Jason Lemkin of SaaStr asked the question. His revenue agent alone makes between 35,000 and 40,000 API calls a day, and a price per call that looks trivial turns into a large number by month end. This week on ๐๐ก๐ ๐๐ ๐๐ซ๐๐๐ค๐๐จ๐ฐ๐ง: ๐๐ฆ๐๐ฅ๐ฅ ๐๐ฎ๐ฌ๐ข๐ง๐๐ฌ๐ฌ. ๐๐ข๐ ๐๐ฆ๐๐ข๐ญ๐ข๐จ๐ง๐ฌ. ๐๐๐๐ง ๐๐ฎ๐๐ ๐๐ญ. Your agent is a new kind of customer, and vendors have noticed. The next line in your software budget may be one nobody has written yet. Inside the episode: ๐ด That $240,000 against the $5 Postgres instance the agent itself suggested, and why the contract and the security review come before the saving
๐๐% ๐จ๐ ๐๐ข๐ซ๐ฆ๐ฌ ๐ฐ๐ข๐ญ๐ก ๐๐๐ฐ๐๐ซ ๐ญ๐ก๐๐ง ๐๐ข๐ฏ๐ ๐ฉ๐๐จ๐ฉ๐ฅ๐ ๐ฌ๐ญ๐ข๐ฅ๐ฅ ๐๐๐ฅ๐ข๐๐ฏ๐ ๐๐ ๐ข๐ฌ ๐ง๐จ๐ญ ๐๐จ๐ซ ๐ญ๐ก๐๐ฆ. ๐ This week Kree and Varun sit down in studio with Mehnaz Tabassum, social entrepreneur and one of the founders of CogniCo AI Consulting, which exists to help nonprofits, lean teams and changemakers adopt AI on tight budgets. This week on ๐๐ก๐ ๐๐ ๐๐ซ๐๐๐ค๐๐จ๐ฐ๐ง: ๐๐ฆ๐๐ฅ๐ฅ ๐๐ฎ๐ฌ๐ข๐ง๐๐ฌ๐ฌ. ๐๐ข๐ ๐๐ฆ๐๐ข๐ญ๐ข๐จ๐ง๐ฌ. ๐๐๐๐ง ๐๐ฎ๐๐ ๐๐ญ. Named Atlantic Canada's most inspiring immigrant, she has built four purpose driven ventures from the ground up, and was once one of the youngest senior IT project managers in the Government of Canada. She knows the technology, the mission, and the reality of doing a great deal with very little. Inside the conversation: ๐ด The adoption gap between small and large businesses shrank from 1.8x in 2024 to about 1.2x in 2025, and 91% of small businesses using AI say it lifted revenue
๐๐ฎ๐ฌ๐ข๐ง๐๐ฌ๐ฌ๐๐ฌ ๐๐จ๐ฎ๐ ๐ก๐ญ ๐ฆ๐จ๐ซ๐ ๐๐ ๐๐ง๐ ๐ฉ๐๐ข๐ ๐ฅ๐๐ฌ๐ฌ ๐๐จ๐ซ ๐ข๐ญ. ๐๐จ๐๐จ๐๐ฒ ๐ข๐ง๐ฌ๐ข๐๐ ๐ญ๐ก๐๐ฆ ๐๐๐๐ข๐๐๐ ๐ญ๐ก๐๐ญ. ๐ Ramp's latest reading shows token spend down 5.2% while the number of tokens used went up. The explanation on offer is price competition between the big labs, which means the saving arrived from outside the business entirely. This week on ๐๐ก๐ ๐๐ ๐๐ซ๐๐๐ค๐๐จ๐ฐ๐ง: ๐๐ฆ๐๐ฅ๐ฅ ๐๐ฎ๐ฌ๐ข๐ง๐๐ฌ๐ฌ. ๐๐ข๐ ๐๐ฆ๐๐ข๐ญ๐ข๐จ๐ง๐ฌ. ๐๐๐๐ง ๐๐ฎ๐๐ ๐๐ญ. A discount you did not plan for is a discount you will not keep, unless somebody is watching the bill. KPMG surveyed more than 2,100 senior leaders across 20 countries and found which habits separate the organisations with an established return from the ones still experimenting. Inside the episode: ๐ด 77% of the established group track AI costs on a dashboard, against 43% of the experimenters
๐ ๐จ๐ฎ๐ซ๐ญ๐๐๐ง ๐ญ๐จ๐จ๐ฅ๐ฌ. $๐๐๐ ๐ ๐ฆ๐จ๐ง๐ญ๐ก. ๐ ๐จ๐ฎ๐ซ ๐จ๐ ๐ญ๐ก๐๐ฆ ๐ ๐๐ญ ๐จ๐ฉ๐๐ง๐๐ ๐ข๐ง ๐ ๐ง๐จ๐ซ๐ฆ๐๐ฅ ๐ฐ๐๐๐ค. ๐ณ Which is a team of fourteen where ten have never turned up. Most owners have never added that column up, because every line was small enough to ignore and they arrived one at a time. This week on ๐๐ก๐ ๐๐ ๐๐ซ๐๐๐ค๐๐จ๐ฐ๐ง, we open a theme that runs all seven days: ๐๐ฆ๐๐ฅ๐ฅ ๐๐ฎ๐ฌ๐ข๐ง๐๐ฌ๐ฌ. ๐๐ข๐ ๐๐ฆ๐๐ข๐ญ๐ข๐จ๐ง๐ฌ. ๐๐๐๐ง ๐๐ฎ๐๐ ๐๐ญ. Output stopped being tied to headcount. For most of business history, doing more meant hiring more. That equation has quietly come apart, and the gap between what a lean team can attempt and what it can afford narrowed sharply in about eighteen months. Inside the episode: ๐ด Solo founded US startups rose from 23.7% in 2019 to 36.3% by the first half of 2025
๐๐จ๐๐จ๐ญ๐ฌ ๐๐๐ง ๐ฉ๐ก๐ฒ๐ฌ๐ข๐๐๐ฅ๐ฅ๐ฒ ๐๐จ ๐๐๐จ๐ฎ๐ญ ๐๐% ๐จ๐ ๐ญ๐ก๐ ๐ญ๐๐ฌ๐ค๐ฌ ๐ข๐ง ๐๐ฆ๐๐ซ๐ข๐๐๐ง ๐ฃ๐จ๐๐ฌ. ๐๐ญ ๐ข๐ฌ ๐ฐ๐จ๐ซ๐ญ๐ก ๐๐จ๐ข๐ง๐ ๐๐จ๐ซ ๐.๐% ๐จ๐ ๐ญ๐ก๐๐ฆ. ๐ณ Both numbers come out of the same research. The distance between what a machine can do and what it makes sense to employ it for is enormous, and almost every argument about this ignores the second figure. This week on ๐๐ก๐ ๐๐ ๐๐ซ๐๐๐ค๐๐จ๐ฐ๐ง, we close the theme: ๐๐ก๐๐ญ ๐ก๐๐ฉ๐ฉ๐๐ง๐ฌ ๐ฐ๐ก๐๐ง ๐๐ง ๐๐๐ ๐๐จ๐๐ฌ๐ง'๐ญ ๐ฃ๐ฎ๐ฌ๐ญ ๐ฎ๐ฌ๐ ๐๐, ๐๐ฎ๐ญ ๐ฌ๐ญ๐๐ซ๐ญ๐ฌ ๐๐ฆ๐ฉ๐ฅ๐จ๐ฒ๐ข๐ง๐ ๐ข๐ญ? Which ends where it had to end. When the thing you employed gets something badly wrong, whose name is on it. Inside the episode: ๐ด A finished model scrapped on safety grounds, for not quite meeting the bar on staying within scope and authorisation