What the AI company hiring 60 marketers tells the rest of us.

Here’s a thing I’ve been chewing on.

Anthropic, the company behind Claude, one of the leading AI models, is currently hiring somewhere north of 60 people into marketing and brand. Field Marketing Managers in the US. Partner Marketing Leads across EMEA. Product Marketing Leads. Segment Marketing Leads. Events Managers. Video Directors. A Head of Comms, Brand and Marketing Recruiting. Roles spread across San Francisco, New York, London, Dublin, Munich, Paris, Tokyo, Sydney, Bangalore and Zurich.

The AI company is building a proper, globally distributed, full-stack GTM function. From scratch. Well remunerated. Fast.

Now contrast that with a conversation I had with a founder a few months back. He told me, with a straight face, that he was going to run his marketing function on two AI agents and a contractor. I asked him who was going to decide what to ship next quarter. He said the AI would. I wished him luck and moved on. He’s not the only one. Variations of that conversation have been a recurring theme in CEO meetings over the last 18 months. “We’re looking for a CMO but we’re really leaning into AI. We think we can run marketing leaner, maybe with a couple of generalists and a good stack. Thoughts?”

My thoughts are this. If AI were ready to replace the function, the people building AI wouldn’t be hiring sixty marketers.

Worth saying upfront – – this isn’t unrelated to the CMO churn problem I wrote about last summer. The same CEOs who burn through marketing leaders every 18 months are often the same ones reaching for AI as the next silver bullet. The pattern is the underlying issue, not the tool.

Before we go further

I’m not anti-AI. I use it every day. I’ve built marketing functions from zero in multiple different commercial contexts, and the tooling I’d use to do that today is unrecognisable from what I had three years ago. Used properly, it is genuinely brilliant.

• Research and competitive analysis at a fraction of the time it used to take.
• First-draft copy, messaging variations, campaign concepts to iterate on.
• Data crunching and reporting that would have taken a junior analyst a week.
• Personalisation at scale for email and ad campaigns.
• Synthesis of customer calls and qualitative research into usable insight.

If you’ve built the right stack and trained your people to use it, a small marketing team today can move faster and deliver more than a big one could five years ago. That part is genuinely true, and anyone telling you different is behind the curve.

So why the concern? Because saving time and removing people from an organisation are not the same thing. CEOs and decision makers keep mixing them up.

The Klarna example worth looking at

You’ll have seen the Klarna story. Between 2022 and 2024 they replaced roughly 700 customer service staff with an AI assistant built in partnership with OpenAI. In 2023 they cancelled their marketing contracts too, on the view that AI could handle creative and brand. The CEO did the rounds telling anyone who’d listen that AI had made their human workforce obsolete.

Klarna spent two years telling the world they’d cracked it. By mid-2025 they were quietly putting the band back together. Rehiring marketers. Rehiring humans for customer service. Sebastian Siemiatkowski admitting publicly that the AI-only approach had delivered “lower quality” and that customers must always have the option to speak to a real person.

Read that last bit again.

The AI boosters will point out that the models Klarna deployed in 2023 are a generation behind what’s available today, and they’d be right. But the deeper lesson isn’t about model capability. It’s about what happens when cost saving, not customer or brand outcome, becomes the sole driver of the decision. That mistake is repeatable at any capability level.

Klarna is now the cautionary tale being cited in boardrooms. Executives pitching AI workforce strategies are being asked how their plan avoids the Klarna outcome. That’s a decent filter question for anyone selling you “AI replaces marketing” as a business case.

The SDR thing is happening all over again

The Klarna pattern is repeating in sales development right now. The autonomous AI SDR category peaked in hype through 2024 and 2025. Plenty of companies piled in, cut their human SDR functions, and deployed the likes of 11x and Artisan as full replacements.

The numbers surfacing now are telling, even allowing for the fact that the vendors reporting them often have their own commercial angle.

• UserGems reporting 50 to 70 percent annual churn on AI SDR tools.
• Gartner projecting that more than 40 percent of agentic AI projects will be cancelled by the end of 2027. Forecasts are forecasts, but they shape procurement behaviour regardless.
• Hybrid models (AI augmenting human SDRs rather than replacing them) reportedly generating 2.8x the pipeline of the full-replacement crowd.

Plenty of businesses are now quietly rebuilding human SDR functions they dismantled 18 months ago. Expensive, awkward, and hard to hide from the board.

What AI isn’t reliably doing today

I’ll caveat this because the goalposts move every six months. The claim isn’t that AI will never do any of this. It’s that today, in typical enterprise deployments, this is where the wheels come off.

Reading context it hasn’t been told. AI works with the information you give it. It doesn’t know your board’s risk appetite this quarter, the politics between sales and product, whether your CFO is about to get squeezed on runway, or that the CRO is quietly interviewing. Good marketing decisions depend on signals that mostly live outside the prompt window.

Pushing back with skin in the game. Yes, modern models can critique a brief when you ask them to. What they won’t do is spend political capital defending the critique in a boardroom, or recognise that what’s needed right now is a hard conversation rather than another campaign. A senior marketer earning their salary is the one reframing the ask and protecting focus. AI executes. It doesn’t fight for a direction.

Brand and tone that actually lands in your market. AI will generate content that reads well and is strategically forgettable. Tone that sounds like three of your competitors. Claims that look impressive but don’t survive a sales demo. Fine-tuned models help. They don’t close the gap to a practitioner who knows what makes your buyer stop scrolling.

Pulling the whole function together. Marketing is brand, product marketing, demand gen, sales enablement, PR, events, customer marketing and ops, all pulling in the same direction. Someone still has to decide what to drop and what to push, given this quarter, this CEO, this market. That someone has a salary. It’s the most valuable part of the job.

Relationships. With sales leaders, customers, press, analysts and partners. The ones you lean on when a launch is late or a deal is wobbling. Those aren’t sitting in any prompt library, and won’t be for a wee while yet.

The counter-argument, and where I land

The fair counter is: “This all shifts. Models keep getting better. Agentic systems are closing the gaps. Your piece will age badly.”

Some of it will. That’s fine. The question isn’t where AI will be in 2028. It’s what you’re deciding right now, with today’s capabilities, and what the cost is if you get it wrong.

Today the cost of dismantling a marketing or SDR function and rebuilding it 18 months later is enormous. Lost pipeline. Damaged brand. Rehiring costs that often exceed the original “savings.” Board credibility gone. Klarna has already shown us the maths.

A simple test before you automate anything

If you’re a CEO or a CMO trying to decide what’s safe to automate and what isn’t, three questions are worth asking before you sign anything off.

1. Is this work, or is this judgement? Drafting a follow-up email is work. Deciding whether to send it, to whom, and what it should actually say given the deal context, is judgement. Automate the first. Be very careful with the second.

2. Who pays the price if it goes wrong? If you pay (extra time fixing AI output), the cost is visible and you’ll learn quickly. If your customer pays (off-brand messaging, irrelevant outreach, lower-quality service), you’ve quietly transferred the cost to the people you’re trying to keep. That’s a slow-burn brand bill that doesn’t show up on this quarter’s P&L, which is precisely why CEOs keep doing it.

3. Can you reverse it in 90 days? Cancelling a tool licence is reversible. Letting go of three senior marketers and dismantling your demand gen team is not. The harder it is to unwind, the more evidence you should need before doing it.

Most of the AI replacement decisions going wrong right now would have been caught by question three alone.

Three things to take away if you’re a CEO

One. If the company literally building the AI is hiring 60 marketers, your instinct to run marketing on a skeleton crew plus a tool stack deserves a second look. They’ve done the maths. They know what the tools can and can’t do.

Two. Automate the work, not the judgement. The most expensive mistakes in this space come from confusing the two.

Three. The cost of getting this wrong isn’t just money. It’s 12 to 18 months of lost pipeline, a damaged brand, and the fun of explaining to your board why you’re rehiring for the roles you just cut. Ask Klarna.

And if you’re a marketer reading this

Here’s what I actually do. Take it or leave it.

• I have a personal AI workflow I use every day. Not as a party trick. As the way I work. If you’re not, the marketers underneath you already are, and they’ll be doing your job in two years.
• I run experiments that might not work. The marketers who’ll be valuable in 2028 are the ones who learned, in 2026, what AI is actually good for in their specific context.
• The ones who get fluent first move fastest, and they make the rest of the team better by osmosis.

Don’t throw the baby out with the bathwater. And don’t fall asleep in the bath either.

Godspeed