Gemini 4 Argon for Business: Greg Isenberg's Three Builds

In Make cool things with AI that people buy, Greg Isenberg shows what a week of Gemini 4 Argon produced for him: a playful rebuild of his holding-company website, a personalized physical-product customizer, and a quoting app for commercial cleaners. The useful lesson is not that a single prompt creates a business. It is that sharper references, a reachable niche, and repeated review can turn a model demo into a testable offer.

Watch Greg Build With Argon

Credit and disclosure: Greg Isenberg created the demos and explains them in the original episode. He says Google sponsored the video and supplied early Argon access. This article links his timestamps and checks product availability against Google's current release notes.

Build 1: An Interactive Marketing Site

At 03:09, Greg compares the older Late Checkout holding-company site with an Argon rebuild. He gives the model a Mondrian-inspired direction, then rejects an early version as crowded and game-like. The later demo has navigable buildings, motion, sound, a world passport, and small interactions. The important step is his editorial judgment: the first output captured the concept, but not the desired taste.

Greg describes the original site as a major investment and credits it with helping the business. Those are his estimates, not a controlled before-and-after test of conversion. Nor does an engaging new demo prove that it is accessible, fast, maintainable, or better for customers. A useful first pilot would build one interactive section from licensed brand assets and compare completion rate, time on task, performance, and customer feedback with the current page.

At 07:49, Greg also shows Justine Moore's interactive Apple-computer museum as another creative example. The clip is an inspiration reference, not evidence that every generated historical image, specification, or asset license was independently verified.

A Reusable Writing Voice, With Review

Greg's second marketing test starts at 09:20. He asks Argon to analyze his posts, newsletters, and videos, then make a reusable voice skill: writing principles, sentence patterns, a lexicon, examples, and templates. A post about marketing engineers echoes his earlier ideas and sign-off. Greg scores the draft 9.2/10; that is his subjective rating of one output, not a model benchmark or permission to publish without editing.

The transferable workflow is to provide your own or licensed samples, ask for an explicit style guide, hold out a few examples for a blind review, and check every factual claim before publishing. If the source material contains customer messages or private drafts, do not feed them into a tool without the right permissions. Good writing also depends on a real point of view; a voice file alone cannot supply one.

Build 2: A Customizer That Exports an STL

At 13:03, Greg demonstrates a Vibe Manufacturing Studio prototype: a 3D preview, customization controls, estimates for print time and material, a suggested price, and an STL export. His example is a personalized espresso portafilter station. He proposes selling the customizer to a niche merchant or letting the merchant sell the items. The Reddit price and margin examples in the video are ideas to test, not verified demand or unit economics.

His proposed order flow is 17:41: choose size and color, pay through Shopify, validate the generated file, then print and ship in-house or through a partner. Slant 3D documents a print-on-demand API. Shopify documents product-media support for 3D models, but an embedded made-to-order widget, order metadata, and STL fulfillment still need an actual app integration. An exported STL is not automatically printable: check dimensions, wall thickness, fit, material, safety, printer tolerances, failures, shipping, and returns on real samples before accepting orders.

Build 3: A Free Quoting Tool for Commercial Cleaners

Greg's software angle begins at 19:01: find a recurring, specific problem through buyer interviews and complaints on G2, Capterra, Reddit, or Idea Browser. Give that buyer a focused free tool first, then learn whether a deeper paid workflow is worth building. A free tool is a distribution experiment, not a guaranteed funnel.

His example at 21:59 asks Antigravity for a web app that converts site-visit notes into a reviewed commercial-cleaning proposal. The demo includes margin calculations and a quoting flow. Greg asks the agent to inspect the project, explain its plan, use the existing stack where possible, avoid deployment or sending proposals, and list unfinished work honestly. He likes the engineering on the first pass but says the design needs references and more rounds.

Before a cleaning company relies on it, test labor rates, travel, supplies, taxes, exclusions, minimum margin, unusual jobs, and edits against known quotes. Keep a human approval step before anything goes to a customer. The video does not show independent security testing, deployed production use, or a paying customer.

Six Techniques Greg Uses

Greg's 25:12 recap is about the agent workflow around the model. Google documents Antigravity's subagents and /goal runs; command availability can vary by client and release.

1. /plan: agree on scope and acceptance criteria before changing files.
2. /subagents: separate independent research, implementation, and checks; review the combined result.
3. /goal: pin a quality target and references across a longer run. Higher effort can use more resources.
4. /browser: inspect the actual user journey, not just generated code or a screenshot.
5. /learn: turn corrections into reusable project rules or skills, after checking they generalize.
6. Effort setting: Greg starts with medium and prefers high for important work. Test the extra cost against the quality gain for your own tasks.

Argon Access and Pricing: Read the Fine Print

Google's October 13 release says Argon is available to Pro and Ultra subscribers in the Gemini app and AI Mode, to Ultra subscribers in Antigravity, and through the paid Gemini API for developers. Google's plan announcement introduced a $100/month Ultra tier and a separate $200/month top Ultra tier. Greg reports using the $100 tier; pricing and quotas can depend on region and plan details. His statement that Pro is not the plan for Argon in Antigravity matches the release notes, but Pro users can use Argon in the Gemini app.

The often-repeated one-million-token number is Google's unified input/output model limit through the API, not a monthly token balance included in Ultra. Google's introductory Argon API price is $2 per million input tokens and $10 per million output tokens; its release notes say that changes after November 30, 2026. Subscription access, API billing, limits, and model selection are separate decisions. Check the current plan and a small task before paying for a full workflow.

Video Chapters

Business Idea Prompt: Find One Buyer First

This prompt uses Greg's three categories but makes the next move a small, reviewed customer test. It does not assume a prototype is a product or that a model's output proves market demand.

Sources and Limits

Creator account: Greg Isenberg's sponsored episode and the supplied transcript are the evidence for what he built, preferred, and estimated. Product facts: Google's Argon release, subscription announcement, and Antigravity update. Production boundaries: Shopify's 3D media documentation and Slant 3D's API page. The sponsor link in the video description was truncated, so this post uses the direct Google release page. No independent sales, print-quality, security, or conversion results were supplied for Greg's three demos. Evidence checked October 16, 2026.