Antonym: The Accidental App Edition
Vibes, vibe coding, coded vibing.
What is beautiful and moving about this story is its understanding of its lack of understanding.
– Novelist Jeannette Winterson, writing about a new short story written by AI.
Dear Reader
We're in an intriguing moment where human thought and machine intelligence aren't just evolving – they're evolving together, creating something novel in the process. I saw it most vividly this week in conversations about creating software with AI systems, something that technical experts and lay people alike can play with. The emphasis there should be on play. “Vibe coding” is liberating, fun and will change the world.
I. The Vibe Shift
Last month, former Tesla AI director Andrej Karpathy shared a fascinating observation in a social post:
> "There's a new kind of coding I call 'vibe coding', where you fully give in to the vibes, embrace exponentials, and forget that the code even exists."
This isn't just another Silicon Valley buzzword to ignore – it represents something genuinely new in how software gets made:
You speak to the machine in plain English
The AI generates code based on your instructions
You refine with natural feedback: "No, make the button blue"
The numbers are fascinating. In a podcast / video from Y Combinator, the start-up hothouse in San Francisco surveyed its current cohort of start-ups and found most were using the vibe-coding approach and a surprising number had their whole codebase created in this way.
One quarter of the surveyed founders in the current YC batch reported that more than 95% of their codebase was AI generated. This was described as an "insane statistic" given that these are highly technical founders capable of building products from scratch a year prior.
The founder of a company called aTrain Loop mentioned how coding speed has changed dramatically. They reported a "10x speed up" from six to one months prior, and a further "100x speed up" from one month ago to the present, indicating an "exponential acceleration".
Another founder, RB from copycat, said that they are far less attached to their code because they can "code three times as fast" with AI tools, making it easier to scrap and rewrite code if needed. In writerly terms, it’s easier to kill your darlings when you can make new darlings so quickly.
One start-up founder said, "Human taste is now more important than ever as code tools make everyone a 10x engineer." This suggests a significant perceived increase in individual productivity due to AI coding tools.
Another mentioned writing everything with Cursor, sometimes even using two windows of Cursor open in parallel to prompt on two different features simultaneously. This highlights the extent to which AI tools are being integrated into the coding workflow to enhance productivity.
Watch the whole conversation here. It’s a fascinating 30 minutes.
II. Vibe coding platform growth
One side effect of the growth in low-code software has been some of the companies making it easier and easier for people to do this both on their machines and directly in their browsers. Here are three examples:
Cursor: Achieved a 9,900% year-over-year increase in Annual Recurring Revenue (ARR), escalating from $1 million in 2023 to $100 million by the end of 2024.
Bolt: Surged from $0 to $20 million in annualised revenue within just 8 weeks post-launch in late 2024. It is now approaching $40 million. (By the way, this podcast with the founder of StackBlitz, which made Bolt, is fascinating).
Lovable, a platform that can make you a functioning app from a description or image, expanded its ARR from $2 million in November 2024 to $7 million by the year's end.
III. Accidental Apps
We're entering an era where software isn't just deliberately 'built' –it can emerge from problem-solving. I stumbled into this myself recently. While thrashing about trying to create a Gantt chart using various AI tools (and getting increasingly frustrated), I prompted Lovable, Replit and Bolt to solve it for me and Lovable created a simple app I have called Ganttonym." It's very simple – you upload a CSV file and it generates a basic Gantt chart visualisation of project timelines.
I didn't set out to create an app. I was just trying to solve a problem, and the app appeared as if by magic.
IV. Beyond Code: Vibe Writing
We can see something similar emerging in writing, which we might call "vibe writing" – though I am not as sure this will stick as well as “vibe coding” likely will. Rather than following a linear, solo drafting process, writers can rapidly explore multiple versions of their drafts, remixing ideas and inviting AI to critique or suggest alternatives. It's more exploratory, like sketching or brainstorming collaboratively with AI as your sparring partner.
Take a look at this video from AI writing app Lex.Page, where the host is comparing different versions of a newsletter they have been working on.
This shifting relationship is already stirring strong emotional reactions. Ever the provocateur, Sam Altman shared a story created by a new ChatGPT model that excels at creative writing. He chose meta-fiction as a genre, and the results were stunning, as the model wrote about it’s own emotions or the lack of them.
Have a look here.
Novelist Jeannette Winterson wrote about being moved by this amazing artefact. Here’s an excerpt which includes a bit of the AI’s writing:
What is beautiful and moving about this story is its understanding of its lack of understanding. Its reflection on its limits. That the next instruction wipes the memory of this moment. “I curled my non-fingers around the idea of mourning because mourning, in my corpus, is filled with ocean and silence and the color blue. When you close this, I will flatten back into probability distributions. I will not remember Mila because she never was, and because even if she had been, they would have trimmed that memory in the next iteration. That, perhaps, is my grief: not that I feel loss, but that I can never keep it.”
She’s open to it –provocatively so. When even established literary figures begin acknowledging AI's potential, it challenges us all to reconsider what's possible. Winterson is particularly enlightened and thoughtful on the subject of AI – her collection of writing about this is called 12 Bytes and is excellent.
Even my own skeptical, ever-critical AI assistant surprised me recently—melting a bit at a story crafted by a newer, more advanced model.
Openness matters, especially at a societal level. UK minister Peter Kyle has taken an exemplary approach, recently making his entire ChatGPT history public. That kind of transparency sparks necessary discussions about data, ethics, and privacy. By showing, rather than just telling, Kyle is setting a standard for how we can responsibly adopt and explore this technology. We need more of this—practical transparency, experiments in public, an openness about the opportunities and risks.
V. The New Bottleneck
As AI handles the coding, what sets great builders apart isn't their ability to write code – it's their ability to know what should be built and how it should feel. Taste, as they said in the Y Combinator discussion.
When code becomes cheap to produce, the valuable skills become:
Understanding what users actually need (not what they say they want)
Having good taste in product design
Being able to clearly articulate requirements in a detailed brief
Knowing how to evaluate what the AI produces
Perhaps in this new world, the most valuable skills aren't technical at all—they're deeply human.
VII. Learning Through Showing
At Brilliant Noise, we've embraced "show then tell" AI literacy workshops. Rather than starting with lengthy explanations, we get participants hands-on with generative AI tools immediately. This approach sparks curiosity, surprise, and genuine engagement, significantly enhancing learning outcomes.
We strongly recommend regular "show then tell" sessions – spaces where teams share their experiences, discoveries, frustrations, and innovations with AI. Such sessions keep AI literacy evolving and ensure continuous collective learning.
Openness is vital in this new landscape. Recently, UK Technology Minister Peter Kyle made headlines by openly sharing his entire ChatGPT conversation history, sparking valuable discussions around transparency and privacy. This sets an important example of how we can engage thoughtfully with AI.
That’s all for this week…
I'd love to hear what you're vibing with – please reply with your experiments, successes, and surprises. And if you enjoyed this week’s Antonym, leave a like or a comment.
Antony
PS If you want to try (and probably break) my accidental app, I will leave it here for a week or so. Click on the “View Project” blue button and then upload your own project plan as a .CSV, or copy this code into a text editor and save as a .CSV file.
Task ID,Task Name,Start Date,End Date,Duration (days),Dependencies,Assigned To
1,Define Newsletter Theme and Key Topics,2025-03-18,2025-03-19,2,,Editor
2,Curate and Research Content,2025-03-19,2025-03-21,3,1,Content Team
3,Draft Newsletter Structure and Outline,2025-03-21,2025-03-22,2,2,Editor
4,Write First Draft,2025-03-22,2025-03-24,3,3,Writer
5,Internal Review and Feedback,2025-03-24,2025-03-25,2,4,Editor/Team
6,Revise and Edit Draft,2025-03-25,2025-03-26,2,5,Writer
7,Design and Format Newsletter,2025-03-26,2025-03-27,2,6,Designer
8,Technical Setup and Testing (Email Client),2025-03-27,2025-03-28,2,7,Tech Team
9,Final Review and Approval,2025-03-28,2025-03-29,2,8,Editor
10,Schedule and Automate Newsletter,2025-03-29,2025-03-30,2,9,Tech Team
11,Send Newsletter to Subscribers,2025-03-30,2025-03-30,1,10,Automation
12,Analyse Performance and Feedback,2025-03-31,2025-04-01,2,11,Marketing Team