👋 Greetings {{first_name|earthlings}},

Emma here, co-founder of Spark AI. Some of the most interesting AI work right now is coming out of the smallest agencies. Jason Bradwell is a brilliant example – he's rebuilt B2B Better around AI, and he's a walking case study in the core skills we believe every agency needs now. This edition covers both – the skills, and how Jason has put them to work with AI. It draws on our latest Spark Session, where I sat down with him to talk it all through and see a live demo of the AI operating system he has built for his agency. And as always, there's a practical exercise included below to try with your team – this time, how to think through an AI process redesign.

Historically, the large agencies and networks have embraced new technology first. Specialist skills and capital were the price of entry, and that made it hard for smaller agencies to adopt and compete. AI has changed the terms. Those barriers no longer stand – subscription costs sit comfortably within reach of a small team, and the core skills to take advantage of it aren't specialist. They are:

None of these needs a technical background – we've written the full framework up on the Spark blog, with a question to test yourself and others against each D. With everyone having mostly equal access to the technology, the biggest blocker is not having the leadership direction to apply these skills to AI.

Jason has all four in spades. Here's what they look like in practice.

Jason Bradwell is the founder of B2B Better, an agency based in Bournemouth helping leaders turn their point of view into business advantage - through podcasts. Seven months ago, he realised the opportunity AI could bring to his business. Rather than using AI on his existing workflows, he built B2B Hub – an internal operating system where the agency's workflows now live, redesigned around what AI is capable of.

That decision – redesign, rather than bolt-on – is where the real value from AI sits. Layering AI onto an old process speeds up a step. Redesigning the process around AI changes what's possible altogether. One agency leader at Cannes last month put numbers on it: adding AI to an existing process gets you roughly 1.2x, but redesigning the process with AI in mind can get you 6x. Quite a difference. And it matters for two reasons.

The first is collapsing steps. Humans tend to work through a process one stage at a time – brief, write, check, build. AI can run several of those stages at once, so the process doesn't need to be broken down that way to begin with. Jason's demo of almost instant landing pages shows this in practice: one prompt covers copy, imagery and build, referencing his brand guidelines and tone of voice. A page that would normally take four separate stages comes back as a working first draft.

The second is capability that didn't exist before. Not new in the sense of magic – more that something which would have taken so long by hand it was never realistically attempted becomes possible within a normal working day. Jason built his ‘Counsel’ for this: his team can put a real decision to eight different perspectives – a risk analyst, a customer advocate and a sceptic playing devil's advocate among them – and get a considered recommendation back in under five minutes.

As the models constantly update, both of these will compound over time. Having these processes in place sets you up to harness those future developments, not just todays.

Beyond this, he's also built his own software tools, replacing subscriptions with versions built for B2B’s own ways of working. That not only cuts considerable cost, it means the features are shaped around what the agency actually wants.

And being a small team works for him. Decisions move fast, without several layers of approval to get through first. It also changes the risk of trying something new – client relationships at that size tend to be closer, so an experiment that doesn't land is a conversation, not a crisis.

Watch the full interview and demo:

The threat isn't lost on the people running larger operations. PwC's 2026 Digital Trends in Operations survey found that 93% of the 767 operations leaders polled agree that affordable cloud and AI tools help smaller firms reach parity with digital leaders. Coming from the people who traditionally hold the advantage, that's interesting.

Recognising it and acting on it are two different things though. Larger businesses may have the resources to set up innovation teams, but bound by layers of approval, they find it much harder to turn the ship. 

The first step for larger agencies is making sure AI leadership is clear and visible. Senior people need to be seen engaging with AI themselves, not just funding it, so they set an example for how the rest of the organisation should a act. It's part of how we coach agencies to build an AI forward culture in our AI for Leaders programme - part of AI Accelerator.

Last week Jules and I were in Norfolk (mostly for my dad’s 70th) but we also fitted in talking at a brilliant event in Norwich organised by Brandland. Jules gave the keynote on how AI is changing brands, consumers and agencies all at once, and we finished with a panel alongside Luke Taylor, co-founder of UnitedUs.

As an approach, UnitedUs are deliberate about where technology is allowed – and it stays out of the ideation phase entirely. That's a brilliant example of the D for Delegation done clearly: knowing exactly what to hand over and what to protect.

Luke shared that UnitedUs's strongest AI output comes from their most experienced designers – including a former lifelong print designer who doesn't code, and is now building interactive assets with AI. Experienced creatives bring real-world references AI doesn't have, and the discernment to know what's good when they see it. The 4Ds in action.

That story raised the obvious question from the room - where does this leave junior talent, who haven't had the years to build those references? Luke's answer was one you don’t hear often (although I know one of our clients is huge on this too). UnitedUs do two things: They regularly take their most junior designers' computers away and send them out of the studio – to draw, to try photography, to build the real-world reference points AI can't provide. And every other week they run speaking and presentation coaching, with juniors pitching internally in a safe environment until it feels normal – because in a world where anyone can generate the document, articulating the idea is what carries it. Rather than shielding their young team from AI, they're deliberately building the human skills (in line with our 4Ds) around it.

I loved this brilliantly simple example from Jules's keynote too. One of our client’s head of client services was drowning in deck reviews – every deck needed her eyes before it went out. So they built a presentation checker, loaded with good and bad examples of every slide type, which gives junior team members their first round of feedback instantly. Amends are down 80%. What they've really done is bottle her judgement – her experience now scales without her sitting in every review.

Thanks to the Brandland team for having us and to Luke for joining – see you soon Norwich.

Now, onto the round-up of all the latest updates over the past two weeks.

OpenAI released its newest model, GPT-5.6, on 26 June. The White House then asked the company to limit early access to around 20 government-approved partners, over cybersecurity concerns. It was the second time in a month a frontier model had been restricted this way – Anthropic was ordered to pull Claude Fable 5 offline in June for similar reasons.

Fable 5 came back first, on 1 July, and free access on paid plans has just been extended again, through 12 July, before it shifts to a credits model. GPT-5.6 took longer to clear. It completed the government's review and had its full public release on 9 July, over two weeks after its restricted launch. Have you tried it yet?

Read more: ChatGPT-5.6 | Claude Fable 5

Two weeks ago we told you all about Microsoft opening up Copilot Cowork to everyone. True to form, always hot on each other's heels, Anthropic and OpenAI have now leapt forward again too.

Now OpenAI is pushing their own Cowork-like tool, ChatGPT Work. It has replaced the ChatGPT desktop app and combines what was ChatGPT and Codex into a single app - just like Claude’s desktop app. You now select Work or Code from the icon on the top right, and get to work.

Jules has been using it for the last few days and likes it. Along with their new model 5.6, it’s just as powerful as Claude Cowork. So these tools are now available in Copilot, ChatGPT and Claude - and they really do change the way you work.

As we keep telling people, don’t get tool anxiety and switch from platform to platform all the time. The features will come, and focus on getting the most out of the platform you already have. Shared ways of working are more powerful than any single feature on a rival platform.

Meanwhile Anthropic hasn't stood still either – Claude Cowork now runs in the cloud as well as just on your laptop. Start a task at your desk, close the lid, and get a phone notification when it's finished or needs your input. (ChatGPT users, don't feel left out – ChatGPT Work launched with the same trick two days later. Nobody stays ahead for more than a fortnight right now.)

Read more: Claude Cowork | ChatGPT Work

Jules and I are going live on Thursday 16 July at 2pm for July's episode of our podcast What's New in Creative AI. This month, we're looking at how much the conversation around AI has shifted over the past year, and we've summarised the essentials of what that actually means for agency leaders trying to keep pace.

We've also got two guests who are worth tuning in for alone – the Head of AI at Microsoft Advertising, and the founder of Made by Humans. Between us, we get into why culture, not the models, is the hardest part of using AI well, why creative AI quietly went mainstream this year, why agencies need to redesign the actual workflow to see real benefit rather than adding AI to the one they've already got, and how AI is already reshaping business models and pricing.

Register here to watch it live or catch the recording after on YouTube or Spotify.

I'm writing this at the end of a ten-session week for team Spark – AI build days and workshops with Pearlfisher in London and New York, Everland, LeisureLabs, Big Group and the ACC. What a team, what a week! 

Until next time,

Emma

Co-founder, Spark AI

PS – if you're wondering what an AI build day with Spark looks like, hit reply and I'll tell you.

Spark AI helps you lead your team through the biggest shift since digital, with AI training, transformation and tools. We’ve worked with 70+ agencies, published the #1 bestselling book on AI for agencies, and teach at Oxford University.

👉 Find out more