This week's question: what work would you want AI doing first, and how would you know it was paying off?

Of course AI should be used to improve efficiency and margin expansion and make these agencies run better. That's the level zero understanding of how new technologies can be applied to improve the businesses.

So where would I start? The one initial piece of advice is just super high fidelity information capture. Just capture everything. Transcripts, every bit of information that flows in, out, or around the business just needs to be captured.

I have a central file system where every single thing that ever happens is there. Every piece of paperwork, every email, every transcript. That's how I'd know what's paying off. If you have all of that, then you can start to mine what's going on for a second order understanding of where to go next, and you can always come back to it. But if you don't record stuff, then you don't have a record of it, and that's lost.

This week's essay is about the other side of it: what still protects a business when anyone can copy the software.

Napper

By Napper Tandy · Why creative businesses are some of the best placed in the world to win the AI era.

Last month, a developer in Atlanta used an AI model to rebuild seven of Adobe's apps, Photoshop through to After Effects, and put them on GitHub for free. It caused a storm on X, and a moment of realisation: one person had built a credible alternative to apps a software company spent decades on.

So I want to argue something that might sound backwards. Creative businesses are some of the best placed in the world to win the AI era. But only if they act now.

The traditional business books will tell you that, for most of business history, protection has come from doing things that are hard to copy. That's something like:

Complexity. "We built something really difficult, so it'll be really difficult for anyone else to copy."

Cost. "We paid loads of engineers to build this, so it'll be very expensive for anyone else to replicate."

Specialist expertise. "We employ most of the very small number of people who know how to build this very specific dental software."

Time to build. "It took us five years to build this, so it'll take you five years just to catch up, and we'll be ahead of you by then."

Proprietary code. "We have millions of lines of code, and copying them would be exhausting."

Process know-how. "We've learned over many years how to build this in a very specific way, and you'll never do it as well as we do."

Switching costs. "Our customers have built everything around us, so leaving would be too painful."

Every one of these is a sentence a founder has said to an investor with a straight face.

All of these moats shared one thing. They were priced on the scarcity of skilled human effort. If building your business took a skilled team, £1 million and three years, then a competitor needed the same just to catch up.

All of this has collapsed. When intelligence becomes incredibly cheap and incredibly abundant, every moat priced in human effort gets repriced at once. The markets noticed ("the SaaSpocalypse"), but the whiplash for businesses is only just starting to be felt.

There are now open-source skills where you can literally put a website's URL into an AI model, and it will rebuild that website or product with remarkable fidelity in minutes or hours, simply by analysing the surface area that's publicly available online. These tools are becoming mainstream, and copying a sophisticated software product is getting closer to trivial.

You see this on the front line of product development today. Granola, the meeting notes app, which we're keen users of at Atlas, launched free dictation last week. A few months earlier, Wispr Flow, for whom speech-to-text was the entire business, had gone the other way and launched a meeting notetaker. It shows the velocity at which companies can accelerate into each other's product ecosystems when the only limitation is replicating code. Your entire business can become someone else's feature in a single release.

So what survives, in a world where the moats that gave businesses their historical protection no longer hold?

The first thing is that you want a business whose engine is bigger than its shell, where what you can see from the outside is only a partial view of the full extent and complexity of the business. Businesses that are just a website, or a single code-based software product, are very vulnerable to displacement by vastly cheaper replacements.

As I see it, six moats survive:

Brand. In a world where humans still make purchasing decisions based on taste, mood and emotion, brands with their own aesthetic and cultural footprint will only grow in their pull, because people make shibboleth decisions: identity decisions about which tribe they want to belong to. How you build a brand in the modern era deserves an essay of its own (next time), because what's possible now is an order of magnitude beyond what was possible before.

Network. Networks are very powerful moats. They move at human speed, and no model can hurry them. Social platforms, marketplaces, environments with buyers, sellers, communities and participants on two or ideally more sides. For example, Reddit has people reading posts, people writing them, and a large community of volunteer moderators. To build a plausible competitor to a platform like that, you need to attract the writers, the readers and the moderators all at once.

Relationships. With talent, buyers, partners, regulators, whoever matters in your world. The human layer is more consequential than ever in building competitive businesses.

Deep domain expertise. We believe strongly that a new generation of companies will be very deep, vertical-specific businesses. Industries have historically looked quite horizontal. An artist has needed a label, a publisher, a booking agent, a merch company and a platform to reach fans, each complex enough to be its own business. We think individual companies can increasingly look to own much more of the whole journey. The scarce resource is a deep understanding of the customer. Building the products is the easy part now. This domain expertise is the durable advantage.

Proprietary context and data. What does only this company know? What are the patterns, information and operating blueprints that only it has access to? Every company should have a compounding brain at its heart, built from its own data, IP and the human expertise only its team has. Every day it operates, that brain gets smarter, and the tools built on it pull further from anything a competitor can copy.

Rate of adaptation. This is the meta-moat, and in a way the most important. Static advantages decay. The ability to keep absorbing each new capability is the one that grows. So how do you build an organisation that sits at the leading edge of improvement in the models, and automatically benefits as the AI wave unfolds and these systems become more and more capable?

Look back at that list. The creative industries have always run on exactly the moats that survive: taste, relationships, rosters, reputation and trust. The threat from AI sits in the operating layer underneath them.

So creative companies are better placed than almost any sector to win in this new world. But they have to move now to capitalise on that advantage.

I think this is the first technological development in history where the gap between the frontier and what most people understand is widening. If you put AI systems in place this spring, you're already a generation behind what's possible today. Static implementations depreciate faster than any software in history.

There were long diffusion lags with electricity and the internet, but those were largely a game of business and society catching up to what was possible. This velocity, with the technology moving on a cycle of weeks, creates an operating environment no business has ever had to run in.

Code and human effort are now the cheap part. The opportunity is a layered human and technical problem. How does a company actually work? Who owns change? How does a culture adopt it?

Implementing new tools and systems in a creative company is not the same as doing it in a bank or an insurance business. There's a specific set of opportunities and problems, but also a specific cultural environment, and a particular kind of person who populates these companies. Understanding how to harness that dynamism, and these technologies, inside that culture is a unique problem, and the one we've made our specialism.

What Atlas ultimately offers is a dynamic system: bespoke to each company, and improving in line with frontier model intelligence.

Our goal is to help an organisation that's threatened by the exponential curve of technology become one that rides it. That means making our clients' own people the owners of the change, and the heroes inside their organisation.

Atlas's own modules and technologies can be copied too. What can't be copied is the domain depth, the relationships, the knowledge of how to work with the people, and the taste. We're betting on the same moats we're asking our clients to bet on, and we think our understanding of what the organisation of the future needs to look like is what's going to get us where we need to go.

There's an alchemy available here. If creative companies take advantage of it, and reposition themselves to ride this exponential curve, they can be some of the most important businesses in the future economy. We want to be at the absolute centre of making that happen, and of making sure these people are the protagonists of this change rather than its casualties.

Anyone can copy your website now. Nobody can copy your people.

Each one read in full by one of us. The initials say who picked it.

Anthropic launches reports that show companies what their AI spend is buying

Anthropic has launched reports for its enterprise customers that show managers a team's work over a date range: what it produced, what it cost and where it broke down. Pooled group budgets come with them, and the data isn't used for training by default. Jellyfish, which measures engineering teams, told TechCrunch that spend on Anthropic-based tools is up tenfold since January, though still about 2% of engineering cost. Its chief scientist, Nicholas Arcolano: "that 10x growth is growth that most CFOs didn't plan for." I invest in Anthropic through Principia.

Our takeaways.

Read on TechCrunch →

WPP's decline narrows, and Cindy Rose talks about AI agents and margins

WPP's third-quarter revenue less pass-through costs fell 2.1%, a narrower drop than the second quarter's 2.8%, and Cindy Rose says the group is on track to return to organic growth by 2027. Adweek's interview with her is headlined on why she's not claiming victory yet, and how AI agents could boost margins.

Our take. There is an opportunity for services businesses to convert themselves towards an economic character that looks more like a software business.

Read on Adweek →

Spotify plans an early 2027 launch for its AI covers and remixes tool

Spotify plans to launch its AI covers and remixes tool in the first quarter of 2027 and is weighing $4 to $5 a month on top of a subscription, not yet final, the FT reports. Artists choose whether to take part even when their label has licensed the catalogue, and royalties from fan remixes go to the artists behind the original songs.

Our take. We view artists and creatives as the founders and CEOs of their own business. It's about decision-making. Ownership. Being the prime mover of what you do, rather than being figured as a product.

Read on MBW →

Universal: ‘The volume is growing faster than the safeguards’

On Universal's third-quarter call, Sir Lucian Grainge said about 90,000 fully AI-generated tracks are now uploaded every day, citing Deezer's figures. He wants unethically made AI tracks paid differently from music people actually want, and says UMG's streaming agreements have to change.

Our take. Proliferating content across channels with AI makes sense. But if you were just creating net new content with AI and just posting that more frequently, I think there would be high potential of compromising the product.

Read on Music Ally →

Meta's new business terms could let it train AI on business accounts' content

From 20 November, Meta's new terms for commercial accounts on Facebook, Instagram, Messenger, Threads and WhatsApp appear to let it use their content in its business products, AI training included. The Copyright Alliance's Keith Kupferschmid calls them "unconscionable". If your business runs any of those accounts, read the terms before the 20th.

Our take. The ownership piece needs to be backed up by enforcement to mean anything.

Read on Digital Music News →

As AI spend starts showing up in agency margins and CFO reports, what work in your business would you want it doing first, and how would you know it was paying off? Reply to this email. A few lines is plenty.

First published on atlasstudios.ai. Field Notes is written by Napper Tandy, Kieran Mallon and Abdo El-Mobayad at Atlas Studios, applied AI for the creative industries. Forwarded this? Sign up at fieldnotes.atlasstudios.ai/subscribe. Napper invests at the frontier of AI through Principia (Anthropic, xAI, Replit). We say so whenever we cover them, and never write about valuations, rounds or share prices.