The Fri-Up August: Sunshine, design aesthetics and AI's expensive reality check

We’re writing this from under the sunny Ibiza skies because we love swimming and we love birthdays. So shout out to Lauren who turns 30 (again) in approximately 3 days. If you didn’t know, we’re basically European so August is our favourite month, because everything quietens down, except apparently when you’re running a small business.

If the below newsletter isn't enough reading for you, we recently put out a report on how to buy data for AI. Because before you invest in models, you need to understand the data that powers them.

We’ve also been popping off on socials this month, so don’t forget to follow Emma on LinkedIn and Lauren on Instagram (or TikTok!) You might get to see some behind the scenes footage of Ibiza (teaser above).

Now, we’re kicking off with something a little different this month, and that is an interview with our pal and designer Simon Blanckensee. He’s a pro at creating visual identities for technology companies so we spoke to him about what makes a good brand, how to become more aesthetically minded and also how to be a good client.

Read on below:

What's your favourite brand identity and why? My favourite identity at the moment is Sigma by Stockholm Design Lab. It's a beautifully considered identity which to me screams classical beauty. It carries SDLs usual exacting technicality but this time it's buttoned up in a romantic melancholy which feels so lovely for the brand and space.

How do you develop your own aesthetic without just following trends? I think personal aesthetics are inevitably pushed and pulled a little by trends but I / designers have spent so long over the years looking at visual media that eventually something just clicks when you look at something, you can’t control what that is and it isn’t really shifted by trends it feels like an innate scratching of an itch that you didn't know you ever had.

What's the biggest misconception people have about hiring a designer? Often companies come into a rebrand or design engagement thinking the designer will naturally have all the answers. Design engagements always require far more importance from the client to reach something successful than they expect. It's so important the client understands that they are responsible for helping the designer understand their world and context, the agency can then work within that world to solve the problem at hand.

What's the most common mistake startups make with their visual identity? Usually adding diversity when they don’t need to, startups need to align their visual identity around a single cohesive thought and repeat that thought again and again where possible through all their brand assets. Oftentimes you see startups bring in all kinds of bells and whistles which only serve to muddy the water for their audience.

Do you think branding will become more or less important as AI makes building products easier? Not just in the hope for self-preservation; I think more. The speed at which companies and individuals can create products and copy each other is so breakneck right now, without an incisive understanding about what your company stands for and against you get completely lost in the noise. I still think the companies that people most want to support have the most resonant brands: companies like Patagonia, Lego, Bang & Olufsen, all clearly stand for things beyond their bottom line which give them staying power and an identity which is difficult to replicate. You’re also now seeing companies like Granola starting to aggressively counter position against the same-ness of their industry in a bid to cut through.

Quick-fire:

Best font? Calibri

Most overrated design trend? Gradients

Most underrated design trend? Gradients

Favourite London discovery? The Hampstead Heath ponds in the evening

And now, as always, some fun stories that we’ve been debating recently.

Canva Afford It?

The AI arms race is showing no signs of slowing, with the world's largest technology companies expected to spend close to US$900 billion on AI infrastructure this year. There's just one problem. Revenues aren't keeping pace.

A new discipline known as "tokenomics" is emerging as organisations scramble to understand what they're actually spending on AI, whether it's delivering productivity gains, or stopping engineers accidentally burning through six-figure compute budgets.

And, we even have a real life example for you. Canva has delayed parts of its AI rollout after discovering frontier models were far more expensive to run than expected. Rather than absorb the cost, it paused deployment, rebuilt its architecture around cheaper in-house models and has slashed compute costs by almost 90%. With 75 million monthly AI users on a freemium model, Canva simply couldn't afford millions of free users chewing through pricey OpenAI tokens. Its fix is to build cheaper in-house models (leaning on its Leonardo.AI acquisition), which have already cut cost-per-task by nearly 90%.

Interestingly, companies including DoorDash, Siemens and Airbnb are increasingly using lower-cost Chinese AI models for routine tasks while reserving premium US models for more complex work. As AI becomes more expensive to run, performance alone is no longer the deciding factor. Cost, flexibility and avoiding vendor lock-in are becoming just as important. So clearly, the AI race is quickly becoming an economics race.

Meta Culpa

The AFR has done an incredible podcast on Andrew Forrest's war with Meta and how it could reshape platform accountability. What started with a phone call from Andrew Forrest's father has become one of the biggest legal challenges Meta has ever faced. After scammers used AI-generated versions of Forrest's image in hundreds of thousands of fraudulent investment advertisements, the Australian billionaire has spent more than A$60 million taking Meta to court.

But this case isn't really about Andrew Forrest. His legal team argues that Meta's advertising algorithms do far more than simply host content, they actively optimise, modify and distribute advertisements, making the company partly responsible for the scams they amplify. If the court agrees, it could fundamentally change how social media platforms are held accountable for the content their algorithms promote, opening the door to class actions, tougher regulation and a significant rethink of the legal protections that have underpinned the platform economy for decades.

SEO? Meet GEO

In perhaps the most 2026 headline yet, Time magazine has started selling advertisements designed for AI agents rather than humans. Brands are paying to influence how large language models interpret and retrieve information, effectively treating AI models as a new audience. SEO is quietly evolving into GEO (Generative Engine Optimisation), and apparently even the robots have become advertising targets.

We had to learn how to GEO our very own website recently, so if you have any questions give us a call. It's a strange feeling optimising content for language models rather than people, but that's increasingly where discovery is happening.

999 Problems

The Police National Legal Database, used by all 43 forces across England and Wales for over three decades, was hit by a cyberattack that exposed names, organisations, and work email addresses of officers, staff, and criminal justice professionals, plus members of the public who'd used its "Ask the Police" service.

An extortion group called ExfilSquad claims it grabbed approx 1.9GB and 135,000 records, has posted a sample, and is demanding ransom. No passwords or victim/witness data were reportedly taken, but it's especially unnerving for staff working organised-crime cases. It lands days after similar breaches at the Department for Education and UK Government Investments, a run that has security experts warning the public sector's legacy systems are years behind the threat. Organisations spend a lot of time worrying about AI risk. Legacy systems continue to do a remarkable job of reminding us not to forget the basics.

Fixture impossible

Adding a 19th team to the AFL isn't just creating another fixture, it's creating another layer of complexity for one of the world's most sophisticated scheduling problems. Behind every season sit more than 400 weighted rules covering travel, rest periods, shared stadiums, television audiences, marquee games and competitive balance, with billions of possible fixture combinations to analyse.

The AFL's challenge isn't finding a schedule, it's finding the best one. Anyone who's ever tried to build an explainable system based on hundreds of competing rules will recognise the problem. It turns out sport and data governance have more in common than you'd think. Both are really exercises in balancing hundreds of competing rules without breaking the whole system.

Risky Business

Businesses have spent the past two years racing to adopt AI but Insurers are only just catching up. A new report warns that more than 90% of AI-related risks currently sit hidden inside conventional insurance policies that were never designed to cover autonomous AI systems. As organisations move from chatbots to AI agents that can take actions, potential liabilities are expanding from simple mistakes to fraud, cyberattacks, professional negligence and even wrongful death.

The message is clear: adopting AI is one thing, understanding who carries the risk when it goes wrong is another. AI governance isn't just becoming a technology issue, it's becoming an insurance one.

RIP Consultants?

As people who spend a lot of time helping organisations make sense of their data, we clicked on this one with mild panic. The Financial Times asked whether AI will make strategy consultants obsolete. Luckily, the answer was more nuanced than we expected.

AI is already brilliant at research, benchmarking and producing slide decks, turning much of the analytical work that consultants once charged for into a commodity. The hard part however is aligning people with different incentives and turning recommendations into action.

Luckily for us, real transformation is still stubbornly human. As we've learnt, there's no AI that can sit down with Shane from Finance and persuade him why data matters. As one former consultant put it, "this is a contact sport." Can confirm.

Minute Details

We're slightly devastated by this one because we're big Granola fans. The AI note-taking app is facing a privacy lawsuit alleging it allows conversations to be recorded and transcribed without all participants' consent, with questions also being raised about how those transcripts are used to train AI models. Similar legal action is already underway against Otter.ai.

Whatever the outcome, the case is a reminder that the fastest-growing AI products are also bumping into some of the oldest questions in data governance: transparency, consent and what happens to your data once you've handed it over. We just hope the lawyers don't ruin it for everyone.

Watercooler Chat: A section of the things we like that keep us sane while running a small business…

Schiaparelli at the V&A: So good, Emma's been twice. It’s astounding how modern clothes from the 1930s can be.

The AFR Podcasts: Emma’s obsessed, and now Lauren is too.

Bar Termini: If you like your negronis filled to the brim, go.

Mexican fleabane: It makes a great garden border and it turns out you can’t kill it, we know, we’ve tried.