AI-Built Apps and Agent Workflows Show Revenue Gains as Founders Cut Costs

Key Takeaways

1. A non-programmer built a fitness app with Claude Code that now earns about $14K in 28 days, while another founder used Lovable and Claude Code to reach over $130K in Stripe revenue in four months.
2. Noah Kagan's Claude analysis of 1,298 support tickets found 30% were partners replying to the company's own automated emails, and fixes cut daily partner tickets from 60 to 15.
3. Interior AI added 3D style previews and rendering, and its founder is weighing a pivot toward AI-reconstructed walkthrough videos.
4. A technical investigation claims weight-layout analysis shows six Chinese open-source models initialized on NVIDIA Blackwell systems, a finding Patrick McKenzie called interesting with policy implications if true.
5. Acquire.com listings and commentary highlight how ownership structure and deal preparation can outweigh headline exit prices.

1. AI-Assisted Products and Agent Workflows

1. Max, a 22-year-old Michigan graduate with no prior coding experience, built a fitness tracking app using a $15-per-month Claude Code membership; it integrates training, nutrition, sleep and supplement tracking with an AI coach on each page, charges a flat $24.99 annual fee, and has generated about $14K in the last 28 days with over 1,000 active subscribers, 20,000+ iOS downloads and 164 App Store ratings. His stack includes Supabase at $20 per month, Vercel, Expo and RevenueCat, and he runs about $250 per day on Meta ads after testing selfie videos, saying Meta currently outperforms TikTok on intent and willingness to pay; about one-third of revenue comes from X 1 2. Separately, Bogdan, who has a computer science degree but said he could not program, used Lovable for a first version and Claude Code to rebuild GetEbook AI, reaching over $130K in Stripe revenue in four months with 1,700+ paying customers and about 5,000 registrations, priced from $29 per month plus one-time credit packs and tutorials, with growth almost entirely from Meta ads 3.
2. Noah Kagan had Claude autonomously analyze his company's partner support inbox over a weekend, covering 1,298 tickets in 33 days; 30% were partners replying to the company's own automated emails, 117 were replies to the same rejection email, and 16% were not partners at all but spam, newsletters or users who could not find a buyer. He also found the "Go to support" link pointed to the wrong page and the FAQ was outdated and unsorted, then stopped nudging completed items, showed submission status, told rejected partners they could reapply after 30 days and moved noise out of the inbox; after one day of fixes, partner tickets fell from 60 to 15 per day 1. Kagan also said a free Grok Bot search found about $59.91 claimable for him and his spouse in a US official unclaimed property database, and shared a prompt limited to official free sources that lists amount, holder, last known address, property ID and official claim link before assisting with submission 2.
3. Tony Dinh said he expects infrastructure costs to fall by at least half but that it remains to be verified whether these tokens and the harness actually deliver results 1. He also promoted ShowTake, a ScreenStudio for coding agents that generates polished demo videos for iOS and macOS apps, arguing that building an app takes one prompt while filming a demo should not take four hours 2.

2. 3D, Rendering and Product Direction

1. Interior AI added a hover-based "🎨 Style" menu for quickly previewing styles in 3D, with a "Render" button to produce photorealistic results; its founder said the next step is trying 3D walkthroughs and integrating video models such as Seedance for photorealistic interior renovation tours 1. He said several planned features are now live: users can walk around a 3D room, upload wallpaper textures onto any wall or object, set object materials and press "✨ Render" to convert the scene into a photo 2. After trying ThreeJS shaders, which he called cool but slow, he switched on advice to WebGPURenderer r187 with SSGI and SSR and said the results were clearly better 3 4. Combining lighting probes suggested by @threejs and @mrdoob, he described an AI reconstruction of a walkthrough video generated on a VPS in under five minutes and called the result good; he plans a demo video for the landing page and may pivot the whole product in that direction to reignite growth, arguing it is far more powerful than basic image-to-image style changes, and asked for feature suggestions 5.

3. AI Infrastructure Claims and Deal Economics

1. Patrick McKenzie called an independent technical investigation into Chinese AI labs' chip sourcing "interesting" and said it has clear policy implications if true; the cited thread claims analysis of initial layout features tied to GPUs and random seeds in model weights shows six Chinese open-source models, including DeepSeek, Moonshot, MiniMax and Ant Group, were initialized on NVIDIA Blackwell systems, with four matching GB200/GB300, that some initial weights were reproduced on real hardware, and estimates Kimi K3 and DeepSeek-V4-Pro each used 2,000 to 3,000 Blackwell GPUs at a single-cluster purchase cost of roughly $150M to $200M. This is disputed/unverified 1.
2. Acquire.com listed several businesses for sale, including a quantitative finance publication and proprietary market-data subscription at $91K TTM revenue asking $199K; a body-type styling platform with $38K TTM profit on $57.2K revenue, 5,600+ paying customers and 100% organic traffic asking $187.8K; an AI LinkedIn content agent for teams and agencies at $80K ARR, 3,100% growth and $68.6K TTM revenue asking $250K; and a blue-light glasses e-commerce brand at $83.5K TTM revenue asking $50K 1 2 3 4. The account also argued that a founder selling a bootstrapped, 100%-owned company for $3M could take home more than a founder selling for $30M after raising $20M and holding 15%, since a 1x non-participating liquidation preference would return $20M to investors first and leave $10M for common holders; it called this a simplified example dependent on ownership, preference terms and the full cap table, and said founders underestimate the importance of ownership percentage 5. It added that exit price is formed long before a business is listed, with founder dependence, revenue quality, bookkeeping, growth and diligence readiness all reflected in a deal 6.

4. Founder Operations and Hiring Signals

1. Codie Sanchez said she has never met a business owner who regretted firing a poor performer too early, only those who regretted waiting too long 1. She described a buyer acquiring a $700K-revenue exterior maintenance company where about half of revenue came from Christmas lighting, bringing in roughly $280K in six to eight weeks, with the rest of the year spent on patio repair, house washing and asphalt maintenance; the 27-year-old company spent about $3,000 a year on marketing and relied on repeat customers, and the buyer valued the structure of summer irrigation and winter lighting each supporting half the year 2. She also said the best candidates make you feel interviewed: @BobbyBizScout, discussing the @bizscout_ president role, asked back about target company size, the worst partner and why, what happens if funding runs out and when the company last fired someone for performance, and she hired him on the spot 3.
2. Flippa relayed Betsy Kauffman's experience that a team was told about a deal after signing a letter of intent, but the deal ultimately fell through, leading her to argue that only leaders critical to the transition should be informed 1.