Claude, Gemini, ChatGPT. These names dominate our Slack as much as client briefs. But here’s what keeps us up at night: How do we harness AI’s innovation without losing what makes us human? While being conscious of what it could mean for communities, or the environment? As a B Corp agency, we’re finding that the answer isn’t choosing between innovation and responsibility. It’s doing both, carefully.

How much energy does your prompt really use?

The uncomfortable truth? Nobody knows exactly. Per-query energy use swings wildly with model size, prompt length, and deployment. The IEA’s 2025 report shows data centres as the primary growth story through 2030, with AI load the key driver. Microsoft links a 23.4% emissions rise to AI infrastructure. Google reports emissions up 51% since 2019. These aren’t small numbers.

But here’s where it gets interesting: the “how” of our usage matters enormously. Studies on prompt-level emissions show orders-of-magnitude differences based on model choice and prompt design. Think of it like this: using GPT-4 for a simple task is like driving a truck to get milk from the corner shop. Translation: right-sized models plus prompt hygiene plus caching beats “bigger, longer, more” every time.

At Circul8, we’re treating prompts like carbon: reduce, reuse, right-size. We default to the smallest model that meets the brief. We schedule heavy jobs in greener grid windows. We cache results obsessively. It’s not perfect, but it’s intentional. In our weekly AI Hot Minutes, we track and share these practices across the team.

Will AI take your job? It’s complicated

WPP, a global advertising powerhouse, is ramping up AI adoption through its in-house platform WPP Open and using it to produce high-quality campaigns more efficiently. Yet the shift is reshaping its operations and causing layoffs. Goldman Sachs projects 300 million jobs globally face automation exposure. Scary stuff. But MIT and Stanford research reveals something crucial: AI augmentation (working alongside AI) delivers 2-3x productivity gains versus pure automation. The “with AI, not instead of” model isn’t just nicer. It’s more profitable.

The World Economic Forum reports 69 million new roles emerging to manage AI, while 85 million face displacement. The math isn’t clean, and that’s the point. Countries with highest AI adoption maintain lowest unemployment but highest inequality between AI-skilled and traditional workers. The lesson? Deployment without development widens gaps.

So is using AI ethical if it displaces jobs? For agencies, the responsibility cuts both ways. Yes, we automate routine tasks. That’s what frees us to create better ideas for clients. But firms that pair automation with upskilling see 3x better retention and 2x revenue growth. The business case aligns with the ethical one. At Circul8, every AI implementation starts with “how does this make our people and our work better?” not “who does this replace?”

What we’re actually doing about it

Talk is cheap. Here’s what accountability looks like in practice. Every week, our Head of Digital & Design runs an AI Hot Minute, distilling what’s signal versus noise in AI developments. From new image gen updates to smarter prompting techniques, we stay sharp while staying grounded. It’s not about chasing shiny objects, it’s about continuous improvements.

We’ve committed to dedicating 5% of AI efficiency gains to team up-skilling. If AI saves us 10 hours, 30 minutes goes to learning prompt engineering or AI strategy. We share our automation roadmap quarterly with no surprises, no blindsides. Where roles evolve, we evolve together through training, shadowing, or transition support.

Most importantly, we keep asking uncomfortable questions. When we brief AI into a campaign: who benefits? When we automate a process: what happens to the time saved? When we chase efficiency: at what cost, and to whom? These aren’t rhetorical. They’re the conversations that shape every AI decision we make.

The path forward

We’re not choosing between staying current and staying conscientious. We’re proving they’re the same choice, made carefully, prompt by prompt. Every efficiency gain could be someone’s livelihood. Every model training draws from the grid. Every automation decision ripples through communities. We can’t optimise our way out of these tensions, only navigate them transparently.

This isn’t virtue signaling. It’s about building the agency we want to work in three years from now. One where AI amplifies human creativity rather than replacing it. Where efficiency gains fund upskilling rather than downsizing. Where success means roles retained, carbon reduced, and capabilities raised.