Gemini 4 Argon: AI Marketing, 3D Printing and SaaS

Greg Eisenberg argues that AI can produce distinctive, interactive websites, but the quality depends heavily on the creative references it receives. He says Argon rebuilt his holding company’s site as a Mondrian-inspired experience with clickable buildings, sound and games. By comparison, his team’s original site, built in 2021, took 40,000 hours and probably cost $200,000. Eisenberg says the business has generated more than $40 million in revenue, with a large part attributable to its website.

Eisenberg considers Argon the strongest frontier model for writing and says it captured his voice using a “skill package” of rules, vocabulary, examples and templates drawn from his writing and speaking. He rated an AI-written newsletter 9.2 out of 10 and recommends using AI to produce more marketing content. His 95/5 rule is that AI handles the plumbing while people provide the taste: the model can create and adapt material, but the human still needs to supply creative judgment and direction.

He describes “vibe manufacturing” as using AI to design personalized physical products and generate STL files for 3D printing. He expects customization to become more common, while noting that businesses can test demand before investing in printers by printing products in-house or using a print-on-demand API. One business opportunity is a Shopify product customizer that lets shoppers choose options, preview products in 3D and buy them. Eisenberg points to niche ideas such as snowboard-themed products and espresso portafilter stations. Espresso enthusiasts may spend $2,500 on machines, he says, and could pay $49 for a custom station. The customizer itself could also be sold to other Shopify businesses, with illustrative margins of 40% to 70%.

Eisenberg rejects the idea that SaaS is dead, arguing that people will keep paying for software that makes their lives easier. He recommends looking for recurring, preferably boring problems, then studying complaints on G2, Capterra and Reddit to identify a focused product for a specific niche. Legacy software can present an opening when it no longer fits users’ needs and a more suitable replacement can be built.

A second speaker cautions that vibe-coded projects often fall short on deployment or security, but says Anti Gravity performed well in their testing. They used it to build a commercial-cleaning app that turns site-visit notes into price proposals for review. Its visual design still needed better references and iteration. The speaker recommends agreeing on scope in plan mode, using subagents for parallel research and checks, testing user journeys in browser mode, and converting corrections into reusable rules with learn mode.

Eisenberg says Anti-gravity is not his preferred harness and that Google still has catching up to do, though he expects the product to improve. He says Argon can also be used with Cursor and other harnesses. Another speaker expects frontier models to become better at complex tasks and is watching for future Anti Gravity updates and Gemini 4.1.