Antonym: The AI Native Company Edition
Also: A million poo-coins a week, world leader bots at the CIA the reason to use reasoning AI apps.
Dear Reader
This newsletter is coming later in the day than usual, as I'm writing on Chicago time. As a fan of the Nolan Batman movies, it’s a thrill to visit what feels like Gotham City, and I find myself editing Antonym in a hotel lobby which really does feel like Wayne Manor.
I’m here to run some AI acceleration workshops sessions for a leadership team. The number of workshops, coaching sessions, and innovation programmes we’re running is increasing—and it’s a joy. The more you work with others in their discovery of tools and ways to use them, the more you learn. It’s a virtuous cycle.
The AI-Native Business
It’s been two years since ChatGPT made reasonably priced AI available to all, and the first AI-native businesses are emerging—companies growing and developing their business models, culture, and systems unencumbered by assumptions and habits formed when SaaS and search engines were the cutting edge of enterprise technology.
A recent Wall Street Journal article by Steven Rosenbush explores some of these early AI-native companies and asks: does this give us a glimpse of the future of work?
The article highlights several examples, including two from the marketing services sector and a platform for creating code, StackBlitz (which Brilliant Noise was an early customer of):
Twice the revenue with half as many people (Supernatural): Supernatural’s CEO reports that their company has achieved twice the revenue of their previous agency, Heat, while employing significantly fewer staff—demonstrating efficiency gains from AI-driven processes.
Accelerated market entry (Supergood): The CMO of U.S. Bank states that Supergood, a creative agency, was able to bring a branding campaign to market in just three and a half months—half the typical time—due to their AI-powered workflows.
$20 million in ARR in eight weeks (StackBlitz): StackBlitz hit $20 million in annual recurring revenue just eight weeks after launching its product, Bolt.new. StackBlitz is an online tool that allows users to build and test websites or apps directly in their browsers, facilitating real-time collaboration without needing to install software.
At Brilliant Noise, our experience has been similar. One silver lining of pivoting our business shortly after the launch of ChatGPT is that we’ve been compelled to rethink many of our systems and approaches to work. We’re not AI-native, but we are maybe AI born again…
As we mentioned two Antonyms back (The Last 5% Edition), the speed at which data analysis can now be conducted allows us to focus more resources on creative work. But once you succeed in rethinking processes for AI, your mind shifts to larger transformations—operating models, business models, and markets.
A reminder of the phases of technology revolutions:
First we do things we do better.
Then we do them in new ways.
Then we do new things.
Sponsored by Brilliant Noise
Antonym is sponsored by Brilliant Noise, home of our sister newsletter, BN Edition. We specialise in short, useful briefings to help you keep up with AI and business developments. Recent topics include:
DeepSeek's impact on financial markets—what does it mean for your business?
The Kim Il Jong Simulator
The CIA has created chatbots based on data they have about different world leaders, according to reporting by Julian Barnes in the New York Times.
One thing AI is particularly good at is simulation and mimicry. The first time I nearly fell off my chair was when we translated an automotive client’s product page into the style of Jeremy Clarkson. And that was with the ancient GPT-3!
It’s not real, of course. But the benefits are substantial:
Surprise: Seeing familiar ideas expressed in an unexpected way can challenge your assumptions.
Simulation as rehearsal: Useful for preparing for difficult conversations or anticipating responses.
Perspective-taking: Insights emerge when you inhabit someone else’s point of view.
Deflection: AI-generated opinions can provide a buffer when exploring contentious or unpopular perspectives.
You can also put simulated personas to work, modelling an approach or tone of voice based on real-world data, books, and public statements.
Give Me a Reason(ing)
Reasoning models are popping up everywhere. Many people are unsure how to use them, just as they were when ChatGPT first launched. Ethan Mollick likens this moment to the early days of ChatGPT, when people were still figuring out prompting techniques.
If you’re just starting with AI, I’d recommend sticking with the excellent GPT-4o default model for now. Reasoning models are something to explore later—except perhaps for research-focused versions, which seem highly useful right from the start.
DeepSeek has disrupted the field, prompting big tech companies to accelerate their reasoning AI efforts. These models work similarly to ChatGPT but add a step where they evaluate the best way to answer a question, critique their own approach, and refine it before responding.
For example, OpenAI’s premium Deep Research model offers up to 100 research projects per month. It functions like a highly efficient research assistant, delivering tailored insights. While hallucinations still occur (especially with earlier models), Deep Research significantly improves accuracy, offering embedded sources and even footnoted formatting.
Two companion tools I’ve built make reasoning models even more useful:
Prompt Optimiser: A custom GPT that refines prompts for reasoning models using OpenAI’s guidelines.
Fact-Checking Bot: A tool that verifies claims against online sources and annotates existing copy.
(If you would find either of these useful, leave a comment or DM me and I will send you a link - they are free if you have a ChatGPT account.)
Coincidentally, after creating my fact-checking bot, Ethan Mollick posted about a study showing that using three independent fact-checking bots can reduce hallucinations by 96%. (Study link.)
One Million Poo-Coins
If OpenAI struggles with naming its products, it’s nothing compared to the world of “memecoins.”
Corruption enthusiasts were amused by a new twist: a political leader issuing coins in their own name (and their spouse’s), creating an untraceable way to transfer wealth to the most powerful figures in the world. In the past, U.S. politicians had to place assets in blind trusts or face scrutiny over property investments. Now, with the chaos of the current moment, they can mint their own currency—designed for speculation rather than utility.
Six of the coins deposited into Trump’s wallet use the same name and symbol as the official Trump coin, while eight feature the “fight” slogan used on the coin’s supporting image. Other coins include “Trump King”, “Trump Kicks Biden”, “OFFICIAL HITLER” and “POO COIN”.
Aside from the crazy car crash spectacle of all of this, one stat in the article stood out:
Brian Armstrong, chief executive of Coinbase, recently said the company needs to rethink its token listing process “given there are [about] 1mn tokens a week being created now, and growing.”
“Evaluating each one by one is no longer feasible,” he added.
One million A WEEK?
There's an eerie parallel with the news there. "Flooding the zone" means that fact-checking and critical thinking are becoming near impossible tasks. Then again, surely there will be an AI-powered process someone could design that could help with this.
That’s all this week…
Thank you for reading. Thank you for taking a look and please do give our like button 🤍 a click if you found it useful!
Best,
Antony