👋 Greetings earthlings,

Emma here, co-founder of Spark AI. There's a question I see flicking behind every agency leader's eyes when we talk about AI: "where are we with this compared to everyone else?" Of course we're all comparing ourselves to our peers – it's only human. But many of you don't ask us this directly - so we wrote the latest Spark Report to help you answer that.

Now in it’s 4th edition, The Spark Report is all about how agencies can move from lots of activity around AI to actually using it for advantage. We use the Spark AI Maturity Model™ to map that journey, based on the patterns we see on the ground across the 70+ agencies we've worked with – from early experimentation through to genuine innovation. It's designed to help you work out where you are and then suggests what your next focus should be. I have a neat little summary for you below.

Also in today’s newsletter: join us for our our first Spark Session to see AI in action - proper demos of agency workflows. We are going live on Thursday with Tuncarp's Chris Murphy - sign up link below. Check out why BarkleyOKRP won Campaign US AI Creative Studio of the Year and big updates from Gemini, ChatGPT and Claude.

Quick links:

• Four stages of AI adoption - where are you?

• AI in action - join us live this Thursday with Tuncarp

• All the tool updates - Claude Design, ChatGPT & Gemini

• Campaign US AI Creative Studio of the Year

The four stages of AI adoption

Stage 1: Experimentation – "Everyone's having a go"

There are no rules. People are using different tools, often on free accounts – whatever they've found that works for them. There are enthusiasts taking things further, but most are just using it like a glorified search engine – typing in a question and accepting whatever came back. Data is being uploaded to unprotected plans. Client Service teams are terrified of getting the ‘how are you using AI?” question from their clients. When procurement asks for your AI policy, there’s a scramble to pull it together. Lots of people are using AI, and there's energy and curiosity, but no structure, no strategy, and little control.

Stage 2: Adoption – "We've standardised the basics"

There’s a plan. The agency has chosen its tools and rolled out professional, data safe plans. There’s been some training, and everyone knows how to use platforms for more sophisticated work. Someone’s in charge of the AI programme, and they’ve written robust guidelines. People are starting to use Projects and creating custom GPTs or Gems for repeated tasks. Most people are using AI to help them with their work most of the time, and using AI to help with client facing work is commonplace. The agency feels confident, but aware there’s much more they could be doing.

Stage 3: Optimisation – "AI is in the workflow, not alongside it"

According to the Spark Report, only around 15% of agencies have reached this stage. Workflows are supported by custom AI assistants (GPTs, Gems, Agents), and simple automations streamline admin. Creative teams using node-based workflows for production-ready output. Client service teams are comfortable talking to clients about how the agency uses AI, pricing models are changing and it’s a core part of pitching and proposals to win new work. Time is set aside for building and testing new workflows. The shape of teams and what they spend their time on starts to change. It’s hard to imagine how work used to get done.

Stage 4: Innovation – "We're designing new ways to work"

Very few agencies are here yet, and that's fine. They’re using proprietary data and methodologies to build AI platforms that deliver value to their clients in new ways, and generating revenue from services that wouldn't exist without AI – rethinking what the agency actually sells. They’re not just creating assets, they’re creating the systems that create the assets. They’re no longer defined by their outputs “we’re a social media agency” but by their impact “we connect companies to their customers”. AI has helped reinforce their moat, lock in their clients, and is a source of competitive advantage.

Where most agencies actually are

The Spark Report data shows 83% of agency staff say they're confident with AI. But we keep seeing that confidence is mostly built on Experimentation. People are prompting individually, getting decent results, and assuming the agency is further along than it is.

When you don't know what adopting AI or optimising for it in your business actually looks like, experimenting with AI feels like progress. It's a classic case of not knowing what you don't know – which is totally understandable right now – and it's why so many agencies overestimate where they are.

AI maturity in action – live this Thursday

If reading about the AI Maturity Model left you wondering what moving through it actually looks like inside a business, here's your chance to see it.

This Thursday we are very excited to be hosting the first in our new Spark Sessions series – where we invite agencies to show the world what AI maturity actually looks like in practice. Up first: a live demo with Chris Murphy, Founder and MD of Tuncarp. Chris will walk through their creative operating system – from a brief landing to a production-ready asset going out the door – showing how brand guidelines travel with the work, how IP protection is built into the process, and how approvals run across teams and clients. Hosted by my brilliant colleague Emma Jackson.

All show, no tell. Register for Thursday's session here:

Tool Updates

Claude Design

Anthropic launched Claude Design on 17 April, the day after Opus 4.7 shipped. It's a separate product to the Claude chat app. You describe what you want, Claude builds an initial version, and you refine it with inline comments and direct text edits. It's aimed at designers and non-designers alike – founders, marketers, sales teams. Decks export as pptx or push straight into Canva, where they remain editable. Wireframes export as HTML, or hand off to Claude Code to build the working version. It’s in research preview for Pro, Max, Team and Enterprise subscribers.

The bit that matters for agencies is the brand system. Claude Design ingests your design files and applies your visual language across everything it makes. Your fonts, your colours, your components – baked in.

Worth knowing who it's going up against. Figma is the direct target, with a dominant share of the UI/UX market. In fact, Anthropic's CPO Mike Krieger resigned from Figma's board three days before launch, and Figma's stock dropped 7% on the day. Canva, by contrast, is now a partner not a rival (interesting in itself) – their CEO publicly endorsed the integration. If you are wondering about tools like Lovable and Bolt, they sit in an adjacent category (working MVPs, not just designs), with Claude Code as Anthropic's answer.

Anthropic isn't alone here in moving towards being a product company. Google has had a very similar stack since last year – Stitch for design, Antigravity for building working software, and AI Studio for developers to prototype with Gemini. OpenAI is taking a different path – turning ChatGPT itself into the platform that everyone else's tools (including Figma and Canva) plug into, with Codex as their answer to Claude Code on the developer side. The big AI companies are all expanding into the tools you actually use, just with different strategies.

I've been in Miami all week so haven't had a chance to have a proper play with Claude Design, but writing this newsletter this morning I gave it 10 minutes and had a wireframe of a "Spark AI Maturity Tracker" app – with our brand pulled in from wearespark.ai and it’s knowledge of what we do already inherent in Claude. it took two prompts to get the brand right and there are still some Claude giveaways (those all-caps mini-titles get everywhere). The Reddit chatter backs this up – people are reporting a "sameness" problem in the designs without pushing it hard. If you've had graphics popping up on LinkedIn this week and thought they all looked oddly similar, you weren't imagining it.

Writing PowerPoint slides has always been the dream of AI for agencies, and lots of tools have emerged to try and address that need. The best to date is probably Gamma, but it still needs a lot of input to make the output great.

Matthew, who leads our creative programme, had a crack at it creating a 10 slide deck with Spark's slide template (the image above is my quick go this morning). Matthew was able to get closer than anything else we've used to try this but it still wasn't perfect and we burned through £60 in credits! So you may want AI to draft your deck for you but how much is that worth? Maybe for a high-value client presentation £6 a slide is worth it. For an internal project status update, maybe not….

This is brilliant for first-pass wireframes, internal decks, and getting ideas out of your head and in front of people. So it is a serious drafting accelerator, not a finished deliverable. Treat it as the new starting line, not the finish line.

Open AI strikes back

Claude has been getting a lot of love over the last few months, ever since Anthropic dropped Opus 4.6 and Cowork, and had a high profile bust-up with the Pentagon. But never discount OpenAI. They’ve been toiling away and have made two big releases in the last week that put them right back at the frontier.

ChatGPT Image 2.0

Last summer ChatGPT’s new image model, GPT Image 1, made waves by being the first high-fidelity image model that genuinely responded to natural language (cue Studio Ghibli cartoons all over Instagram). Suddenly this made creating multiple versions of a similar image, swapping backgrounds or combining images much simpler than it had been before. But compared to traditional diffusion models it was slow.

A few months later Google stole the crown with the wonderfully named Nano Banana, which is now on its third iteration, and has quietly become the backbone of many creative AI workflows. We use it all the time in Figma Weave.

But last week ChatGPT launched GPT Image 2.0 and it looks like it might have stepped ahead once again. Compared to Nano Banana Pro images look more detailed, more photo realistic, and generate faster. And text is now much stronger, even fine-grained text on products or pages is rendered accurately. You can actually ask it to render the page of a book, and the text is sharp and accurate enough to read it.

Now available in Figma Weave, Flora, and other node-based tools too. This should be a welcome upgrade to many workflows.

ChatGPT 5.5

For the last 6 months or so, the consensus among the agency people I talk to was that Claude was simply the better model for serious work. That consensus held up to about Wednesday, when OpenAI shipped GPT-5.5 and matched or surpassed Opus 4.7 across most of the benchmarks that mattered.

So is the difference real?

If you used GPT-5.5 to draft a client email, then used Claude to draft the same email, then asked five colleagues to spot which was which, most of them couldn't. The frontier models are now close enough that, for the kinds of tasks creative and brand teams do most often, the difference between them is dwarfed by the difference between someone who has learned to use AI well and someone who hasn't.

So the pace of model releases reinforces our message to agency leaders: stop worrying about what platform you’re on. The models are all good, and they're all getting better all of the time. The right move is to pick one platform, get it properly rolled out, train your team on it, and let the upgrades come to you.

The performance gap that actually moves your output is not ChatGPT versus Claude. It is "we have a process for this" versus "we don't". Map new process with AI. Then build out the GPTs, projects, or Gems that support the process. This is what will bring you real productivity. The models will keep catching up to each other.

Gemini App and Personal Intelligence

Meanwhile Google has been moving forwards too. Gemini finally has a native Mac app (available direct from Google, not the App Store), and it is the last of the big three AI tools to arrive on desktop after ChatGPT and Claude. Hit Option + Space from anywhere and it's there, no tab switching needed. 

The downside of using the app as opposed to the web version? None of your Gems are available. This seems a big miss in our opinion, as that makes it useful only for the personal user rather than the AI-enabled business user. Just rolled out a suite of Gems to support the way you work? You won't find them in the app.

Gemini is definitely feeling like it's had a little bit less love than Claude and ChatGPT recently, so it wouldn't surprise me if we get a major update soon. It seems like an age since Gemini 3 dropped back in the autumn and was for a while the best model out there.

There's no reason why Gemini can't be a killer app - if they made it easier to connect to external data sources without having to be on an Enterprise account, it allowed for projects that learn from context, as well as the Gems that stick to their instructions, and most of all allowed it to spin up Google Docs, Sheets, and Slides directly in the chat, we’d be there. Here's hoping.

Netflix open-sourced VOID

Netflix released VOID – a video AI model that removes objects from footage and reconstructs how the remaining scene would physically behave without them. Remove a car from a collision and the other vehicle drives on down a clean road. Remove a person jumping into a pool and the water stays still. It's free, Apache 2.0 licensed, and in human preference tests it was chosen over Runway 65% of the time. One for your production or creative tech teams to look at.

Read more | Access VOID on Hugging Face

Work to check out

BarkleyOKRP won AI Creative Studio of the Year in Campaign US. They built an AI-generated 1990s radio universe for Slice Soda, – 40 original songs, synthetic DJ hosts, a month of live FM in Los Angeles. 120 million impressions at 60% more efficient CPM. For The Salvation Army, they created a system that auto-generates ads for thrift items the moment they sell.

The creative work is impressive, but the bit I want to highlight is what sat underneath it. CEO Katy Hornaday established:

• An agency-wide AI upskilling programme alongside an advisory committee to guide responsible adoption across departments.

• They appointed a dedicated SVP of Creative and AI.

• Their guiding standard: AI earns its place only when it makes the work better.

This is the pattern we see every time – the agencies producing the strongest AI-enabled creative work are the ones that invested in structured capability building across the whole team. Leadership commitment, role-specific training, governance, and a clear creative standard. The tools follow. (It's exactly the progression the Spark AI Maturity Model™ describes – and what our 3 month AI Accelerator is designed to build in your business.)

Read the full Campaign article (you need to be a subscriber I’m afraid)

If you haven't read the Spark Report yet, grab your copy here – it's the most comprehensive picture of where agencies actually are with AI that we've put together, with practical next steps for every stage of the maturity model.

That's all for this edition - bumper as usual and I hope you found it helpful.

See you next time!

Emma

Co-founder, Spark AI

About Spark AI

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

👉 Find out more 

Not a subscriber yet?