Antonym: The Data-Squirrel-Vibe-Work Edition

Outputs are cheap — invest in data and process.

Dear Reader

This week gave the closing presentation at The Robots Are Coming, a conference organised by Agency Hackers at the British Library. I’ve developed the following version of the talk for this newsletter.

Data-Process-Output

Revolution means when things turn upside down. Up becomes down. In a very important way, how we work and what we spend our effort on in terms of thinking, is changing before our eyes.

At Brilliant Noise, we’re in the business of helping organisations navigate the wild, often bewildering, world of generative AI. Our focus? “AI literacy” – a deep, practical understanding of how these systems function, where they stumble, and how to weave them effectively into everyday workflows. We empower individuals to harness AI, not as some distant, scary force, but as a “cognitive accelerator,” a “tool-making tool” that amplifies their own capabilities. It’s exciting: it’s a magical time to have a brain.

This presentation marks my third time addressing the Robots conference, and the sheer velocity of change in the AI landscape since the first time is astonishing. Feels like the ground’s constantly shifting beneath our feet, doesn’t it? In our AI literacy programmes, we use a slide that details the rapid succession of major AI releases. For example, in early March, just 16 days after the first workshop with a client, this was our update:

We saw the launch of Grok-3 (claimed to be 10x more powerful than Grok-2), Claude 3.7 (with its new data analysis chops), and a whole slew of updates to ChatGPT, including the UK release of Operator, Deep Research, and ChatGPT 4.5. And the beat goes on, with ChatGPT 5 looming on the horizon.

Those cryptic names and numbers on that slide hide the magnitude of these advancements. Claude 3.7, in particular, has supercharged AI applications, notably in coding – a development we’ll return to later.

Despite this deluge of new knowledge and capability, I find myself constantly echoing the words of William Goldman: “Nobody knows anything.” That quote? It’s become a mantra for me. There’s precious little certainty about how AI will ultimately play out. Just when we think we’ve got a handle on the current state of affairs, everything changes again.

This constant evolution is driven by the expanding “possibility space” that generative AI models and human innovation are creating. As depicted in another slide, we’re witnessing a surge in business innovations, small everyday innovations, and entirely new ideas. The technology is advancing at an incredible rate, and we’re continually discovering novel applications for even relatively older AI models.

It reminds me of the concept of “capability overhang”, which suggests that even if AI’s technological progress were to halt today, we’d still have five to ten years of significant business innovation to explore, such is the transformative power of this general-purpose technology. It's like we've been given the keys to a time-traveling DeLorean, but we're just using it to pop down to the shops for milk and arguing about whether going back in time to yesterday would help us get through our to-do list faster.

At Brilliant Noise, our experience of working with AI has been one of constant learning – a process of iterative refinement. The more we engage with these systems, the more we sharpen our understanding. I’ve come to view the development of AI literacy, and the progress of this technological revolution, as occurring in three phases.

In the first phase, we use AI to enhance existing processes – doing what we already do, but more efficiently. This is the realm of task-level innovation and productivity gains. We’re focused on optimising what we can do. Think of AI as a turbocharger for our current workflows.

The second phase emerges when AI-literate individuals collaborate, re-engineering workflows and discovering new ways of operating. This is where teams begin to think differently about processes, challenging long-held assumptions and finding smarter ways to get things done.

The third, and most transformative, phase is when we start to do entirely new things. Organisations begin to reimagine what’s possible – business models, structures, even paradigms shift. The internet revolution offers a useful analogy: it took years for new business models to emerge, and even longer for companies like Airbnb, Uber, and Amazon to disrupt the global economy. We are, I believe, at a similar inflection point with AI. We’re not just automating the old; we’re about to invent the new.

At Brilliant Noise, we’ve spent the past two years deeply immersed in working with AI, both internally and with our clients. We’ve seen firsthand how non-technical individuals can leverage large language models and related tools in their daily work. This experience has led me to a fundamental realisation, a kind of “lightbulb moment”: the traditional paradigm of work, where the majority of effort is concentrated on the final output, is no longer valid.

Historically, whether it was creating a presentation, a marketing campaign, or a piece of software, the focus was primarily on the final product. Research, checklists, and process management were often secondary considerations – almost an afterthought. However, the advent of AI has dramatically altered this dynamic, demanding a fundamental shift in our thinking.

As I emphasise, outputs are now cheap. The cost of content production is rapidly approaching zero, mirroring the trajectory of content distribution with the rise of the internet and social media. In this new reality, value has shifted decisively towards data and process. We put more effort into data and process because cheap outputs mean that investment in data and process yields better and better results. That’s where the real value is created – the sustainable competitive advantage.

This has profound implications. We’re no longer constrained to producing a single output, agonizing over every word and pixel. With robust data and processes in place, we can generate multiple variations, compare them, and iterate rapidly. This is vividly evident in software development, as I’ll discuss later, but it applies to virtually any form of knowledge work. We become less attached to any single output, more willing to “kill our darlings” and experiment, because we can easily create alternatives. It’s a bit like having a superpower that lets you try out multiple versions of reality before settling on the best one.

This realisation leads to another crucial insight: everything is data. We’re surrounded by it, swimming in it, and by focusing on how we organise it and use it in a process, we can do so much more. In essence, cheap outputs and ubiquitous data necessitate a strong emphasis on process to ensure quality, reliability, and, dare I say, sanity.

However, much of our data needs cleaning. While there are technical aspects to this, the issue is often more about organisation and mindful data creation. I recall a conversation with a data analyst on my team who was preparing client spreadsheets for an AI-driven planning process. I was surprised to learn that “data cleaning” often involved reorganising spreadsheets that looked like Word documents, with titles and subcategories that made sense to humans but were gibberish to machines. It was like trying to teach a computer to read poetry.

This highlights a systemic problem: a widespread lack of training in everyday digital tools. Take spreadsheets. You know, spreadsheets – where we put all the important numbers and calculations. Studies show that less than 50% of people who work with spreadsheets like Excel and Google Sheets have had any training in how to use them. A much smaller fraction has any training in how the most recent versions work. In many cases, data cleaning is about rectifying the mess caused by this under-investment in basic digital literacy. We’re essentially paying the price for years of digital neglect.

We assume we are hiring and working with people who know what they are doing, but mostly they are just muddling through, getting by, using workarounds and things they have learned by osmosis.

Another critical element – the unsung hero of effective AI use – is the brief. Whether it’s a product requirements document (PRD), a creative brief, or a statement of work (SOW), clear and precise instructions are essential. While generative AI can easily produce plausible-looking outputs, creating something truly robust and precise requires meticulous attention to the input. Dr Ryan, my business partner, a true AI whisperer, reckons it’s often better to spend a day crafting a detailed brief for AI than a day on trial and error wrangling ChatGPT or Claude.

It’s easy to get generative AI to produce something that looks okay, that looks plausible. What’s difficult is getting it to produce something – a document that’s robust and precise – unless you’ve spent a lot of time thinking about the instructions that you’re giving it.” As any project manager will tell you this has always been true, and as everyone else will tell you, we’ve largely ignored it and dived in to doing “the work”.

[As we discussed in last week’s Antonym] the impact of AI is particularly has recently been spectacularly shown in software development, where a new phenomenon known as “vibe coding” has emerged. Coined by Andrej Karpathy, vibe coding describes a style of programming where developers fully embrace the flow and exponential capabilities of AI, almost “forgetting that the code even exists.” It’s a kind of symbiotic dance between human and machine.

On stage at the conference I somewhat spontaneously replicated the founder of Lovable’s demo creating a copy of Airbnb with a single prompt: “Clone Airbnb.” Here’s the original (press play and it starts at the demo).

While the result wasn’t a fully functional service, it was a compelling proof of concept. This demonstrates how these tools are expanding the limits of what’s feasible, moving expertise further down the development lifecycle, from proof of concept to MVP, alpha, and beta. As the founder of Lovable suggests, we may be approaching a future where we simply ask for the software we need, and, like magic, it appears. (Here’s the NOT-AirBNB clone I literally generated on stage while delivering about five minutes of excited blathering (I did change the logo just now).)

I also shared data from Y Combinator and Sergey Brin’s comments about the relatively low adoption of AI-generated code at Google. There’s a clear push to accelerate the use of AI in coding to achieve even greater levels of artificial general intelligence (AGI). The message is clear: the future of software development is inextricably linked to AI.

Paradoxically, while vibe-work – fast and intuitive like vibe-coding - is fun and creative, a structured approach to process is more important than ever. (I say again, the most brilliant and disciplined process people I’ve worked with are creative directors). In preparing for this very presentation, I practiced what I preached by using a simple Google Doc with tabs to manage the various aspects of the project. This approach, which we’ve adopted at Brilliant Noise, has proven to be effective for both machines and humans. It’s a bit like having a well-organised toolbox versus a chaotic jumble of tools.

For machines, a well-ordered and structured document is easily readable. Connecting it to a tool like Notebook LM, for instance, means that I can talk to the whole of the project easily simply by updating the document every time I’ve added something. Connecting a document like this to NotebookLM (for instance) means that I can talk to the whole of the project as I am working on it easily simply by updating the document every time I’ve added something. It works well with the AIs that are my co-thinkers because I’ve ordered the data clearly, making it easily readable (or addressable as they say in the tech trade).

But this method also has significant benefits for the human involved. In my case, preparing this presentation was fitted in and around the usual demands of running a company, delivering workshops and coaching sessions for clients, and the rest of life. Having such a clear process document was calming – one feels held by the process, and it’s easy to return and get back into the groove. That’s not to say that notes and brainstorms and scattered post-it notes around one’s desk don’t play a role, but working in this way was very satisfying. It brings a sense of order to the inherent chaos of the creative process.
Finally, I want to touch on the concept of “data squirrelling” and its power, as exemplified by Readwise Chat. For years, I’ve been collecting and organising my notes and highlights, primarily within Readwise. This has been invaluable for reviewing long-forgotten nuggets of insight and interesting quotes, and for searching across my knowledge base. However, the integration of a generative AI chat window has exponentially increased the value of this curated data. It’s like turning a library into a dynamic conversation.

As I explain, “Before now, that has mainly been a way of reviewing long-forgotten nuggets of insight and interesting quotes as well as providing a search engine across all of them. Incredibly valuable. But since they added in generative AI chat window, the value of those notes has squared (if not cubed) – it has become exponentially more valuable.”

This demonstrates a crucial principle: managing data effectively, even at a personal level, allows machines to extract, discover, and generate value more easily. The better the structure, the greater the return. It’s a powerful reminder that in the age of AI, our ability to organise and curate information is more important than ever.

In the AI age, the future doesn’t belong to those who can write the best outputs. It belongs to those who ask the best questions.

That’s all for this week…

Thank you to everyone who gave me feedback at the conference, it’s one of the strongest reactions I’ve had to a talk and is encouraging me to go further in the thinking and experimentation. And an even bigger thank you to my brilliant colleagues at Brilliant Noise, who inspired so much of this talk during our weekly AI show-and-tells.

If you liked this post leave a comment or a like – they are very much appreciated!

Antony