Codex, Claude Code, Antigravity
I shared my take on the AI super-app wars—Codex, Claude, Antigravity—then had Claude explain my own words back to me three ways: using one AI to decode my thoughts about which AI to use.
In response to releases during Google’s big event this week, I wrote this TEXT:
<TEXT>
I switched from Antigravity to Codex six weeks ago as my primary AI app. The next year is going to be a slugfest, a battle royale among the user-level AI super apps: Codex, Claude Cowork/Code, and--when they work the bugs out--Antigravity 2. Today I’m probably 60% Codex (GPT-5.5), 30% Claude Cowork/Code (Opus 4.6), and 10% Antigravity (Gemini Pro 3.5). Codex and Claude Code actually work well together as a left-brain, right-brain team when you set up a messaging system between them. I have zero brand loyalty and am model agnostic, ready to switch to the strongest model any week--don’t fall in love with your chatbot. These tools are similar enough that switching between them is like moving from a Ford to a Subaru to a Toyota--once you can drive one AI app, you can drive any AI app, and switching to the strongest tool becomes easier and easier.
</TEXT>
What does that mean?
This is one of the best prompts I’ve ever written, intended to answer the question: What does this mean? This prompt helps me clarify my own thinking, to express myself better to others.
<PROMPT>
Explain the TEXT as if to a fifth grader, a tenth grader, and a curious adult otherwise unfamiliar with these matters, each in about the length of the original, maintaining point of view.
</PROMPT>
Claude Opus 4.6 unpacks what I was trying to say these three ways:
Fifth Grader
Imagine you have three really smart robot helpers on your computer that can do homework, write stories, and answer questions. The writer tried all three and right now likes one called Codex the best, uses another called Claude sometimes, and barely uses the third one called Antigravity. But here’s the thing: he doesn’t have a favorite team. He’s not like “Codex forever!” the way you might be loyal to PlayStation or Xbox. If one of the other robots gets smarter next month, he’ll just switch. He also found out that two of the robots can actually work together, kind of like how one friend might be great at math and another is great at reading, and they help each other. He says learning to use these tools is a lot like learning to drive a car — once you can drive one, you can pretty much drive any of them, because they all work in similar ways.
Tenth Grader
The writer recently switched his main AI assistant from a Google-based app called Antigravity to an OpenAI-based one called Codex, and he’s predicting that the next year will be an intense competition among three major AI platforms: Codex, Claude, and the next version of Antigravity. He breaks down his current usage as roughly 60/30/10 across those three. One interesting technique he’s found is running Codex and Claude side by side with a way for them to pass information to each other, so they complement each other’s strengths. His bigger philosophy is to stay completely platform-neutral — no emotional attachment to any brand. He compares it to cars: a Ford, a Subaru, and a Toyota all have steering wheels, gas pedals, and brakes, so once you know how to drive, switching is no big deal. He’s arguing that AI apps are converging the same way, and clinging to one brand will only hold you back.
Curious Adult
The author is describing the current landscape of what he calls “AI super apps” — full-featured AI assistants that work at the user level to handle complex tasks. He’s currently splitting his workflow across three: Codex (powered by OpenAI’s GPT-5.5), Claude Cowork/Code (powered by Anthropic’s Opus 4.6), and Antigravity (powered by Google’s Gemini Pro 3.5). He’s found that Codex and Claude can actually be paired together through a messaging bridge, playing complementary roles — one more analytical, the other more nuanced — like two specialists collaborating. His core argument is strategic: these platforms are converging in capability and interface design fast enough that brand loyalty is a liability. The switching costs between them are dropping rapidly, much like how driving a sedan is driving a sedan regardless of manufacturer. His advice is pragmatic — stay flexible, evaluate constantly, and follow the strongest model wherever it lands.
So, who “wrote” that final passage? What is the most honest way to attribute/credit/disclose that Curious Adult text? The ideas in the TEXT are all mine, the PROMPT is all mine, and surely the output surely doesn’t belong to the model. How would you attribute this collaborative text?
— Steve Little, with Claude Opus 4.6 used as an explanatory drafting aid
or
— Written by Steve Little using Claude Opus 4.6
or something else? Lemme know what you think in the comments.