Almost every marketer starts with AI the same way. They use it for content (a blog post here, a batch of images there, maybe one of those slightly unhinged AI videos), or as a smarter Google that answers their questions. Both are good places to start. They're also where a lot of teams get stuck, because they train you to treat AI as one more tool you open when you need something written or looked up.
The biggest mindset shift I see across the marketing teams we work with at 10x Company comes right after that, when a team understands that AI is infrastructure, something that feeds several layers of your strategy and several layers of your tactics at once. And once that clicks, the questions change. Instead of asking which tool writes the best LinkedIn post, you ask what your whole team, and your agents, need to know and share.
Stop using AI like a flashlight
The picture I keep coming back to is electricity. Most teams still use AI like a flashlight: on when they need it, off when they don't, connected to nothing.
You need to be thinking about AI as the new electricity. Most people are still using AI as a flashlight. On and off, on and off.
You see the flashlight thinking most clearly when leadership decides it's time to "go AI first." The company buys a subscription to some AI tools, runs a few workshops and calls it a transformation, a flashlight on every desk. And it doesn't work, because whatever electricity does for a building comes from the wiring behind the walls, the connections that let you plug in something new and have it run.
When I say wiring, people often hear automations. Automations come later. The wiring is your context, your knowledge, your data and the governance around them, what we call the Marketing Brain. It's the fundamental context and data layer behind any agentic work, and it gives the whole team, humans and agents alike, a single source of truth. Build that first, and then you can scale, outpace your competitors and stand out for real.
The good news is that only a few have built a well-functioning Marketing Brain layer so far. Most companies are still clicking the flashlight on and off, so the teams that wire themselves in now get a real head start.
Wire the whole team in, or get silos
The two use cases I see running on that wiring again and again are competitive intelligence and voice of the customer. Both come down to analyzing large amounts of mostly external data, in the context of mostly internal company data. So, everything your competitors publish and everything your customers keep saying, read against what you already know about yourself. The value is in how fast you turn that analysis into a campaign, a piece of content or an event.
None of it works unless the whole team is plugged into the same brain. When one person runs everything through their personal ChatGPT, someone else uses a different tool with different instructions and a third keeps the good stuff in their head, you get exactly what you'd expect: chaos and silos and just randomness, basically.
Each person does get faster on their own. But driving faster without a destination is worth less than moving slowly toward a clear goal, so individuals feel more productive while the marketing team as a whole gets more confused and more overwhelmed.
Be context first
The way out of that confusion is shared context. If AI is the electricity, context is what runs through the wires. Context is everything your AI needs to know about your company and how you work: your positioning, your customers, your voice, the decisions you made and why, what you already tried, your competitors and where they're heading, and your product truth. In most companies it lives in scattered versions of scattered documents and in people's heads, and it gets updated only when someone pushes for it, because nobody owns a process for keeping it current.
Stop thinking about being AI first, AI native, whatever those buzzwords are. Start thinking about being context first.
We've been building AI systems and agents for companies for a while now, and more than the model, more than the tools, the thing that decides whether they work is context. Combine it with skills (documented methodologies for specific tasks), tools and connectors (the plugs that link AI to your other apps), and your people and your agents finally work from one source of truth. It has a lovely side effect, too. Once knowledge is written down clearly enough for an agent, your colleagues understand it better as well, so human-to-human communication improves along the way.
Context is also the part you control. You can't control the model, what it says or the quality of what comes out, but you can guide it through context. With the right context and the right marketing brain architecture, I get excellent results even from models that aren't the newest, and I don't need Fable for everything.
So I keep saying we're going to be polyamorous with models. It won't be one model for everything. Soon most teams will run several providers side by side, OpenAI, Anthropic and perhaps xAI, and each provider already splits the work by tier. With Anthropic, for example, Opus can carry most of the tactical and strategic work and hand the grunt work to Sonnet, while Fable comes in for the highest tier of strategy and creativity. Models come and go, and the context underneath them is what makes each one useful.
Slop still has an author
All that context doesn't mean you hand the keys to the machine. Bad content and lazy ideas existed long before AI, since human beings weren't producing perfect work right up until it arrived. AI just gives us a convenient new thing to blame, and shaking off the responsibility feels nice. The slop is still made by people, though, sitting at their computers, looking at something mediocre and deciding, "You know what? This is good enough. Publish."
The trap is thinking in tasks. A skill, an agent or an automation can now handle a lot of individual tasks, but finishing a task is a long way from high-quality leads landing in your CRM. Teams that only think in tasks forget to ask the strategic question of what they're actually sending out into the world, and what their audience will see when AI does all of it.
The fastest way to get slop is to say, 'Oh, AI can do that. You don't need the human in the loop. Let's just let AI do that, and that's it.' That's how you get something very, very mediocre.
It's also easy to forget how strange AI is. It sounds human without being human. It claims to remember things about me when it doesn't really (that memory gap is exactly why I built Total Recall). It helps me and lies to me in the same breath, and it has no common sense even though it talks as if it does. The models are getting better, and you can feel their judgment improve, but we still have to judge their judgment and keep our taste, our expertise and our humanity in the loop, because that's what connects the dots and lets us orchestrate these systems well.
The same judgment decides how you measure whether AI is working for your marketing. Most teams reach for efficiency: hours saved, posts per week, campaigns shipped. We're past that. We already know AI makes us more efficient, and it shortens our time to market, so the expectations rise with it. What matters is the result at the other end, the qualified leads in your CRM, the campaigns that learn faster, the brand people choose, and reading that takes a lot more nuance than counting hours. It's also exactly the part of the work where marketers earn their place.
Budget for tokens in 2027
Earning that place takes power, though, and electricity comes with a bill. Most 2027 plans don't have a line for it yet. We'll have to budget for tokens, which basically means paying for how much AI your team actually uses, on top of the tools themselves. For some teams that's extra usage on Claude so they can run Fable for strategic work, or more parallel automations and agent squads for more ambitious tasks. For others it's a set amount of tokens put aside for one important campaign.
For a small team, the $20 or $25 subscription may be enough for now. Eventually, though, your people will come to you asking for extra usage. "Give me some tokens, please, I need to keep working." The speed of work is only going up. We were always expected to do more with less, and with AI it's 10x that.
How much you'll need is hard to pin down. Start with what your team used this year and the times you had to top up, then at least double it, because even as models get better and the big labs make them cheaper to run, our ambition grows accordingly. In fact, it must. Set ground rules too, so everyone knows when the expensive models are worth it, for example a quarterly Fable budget for the campaign that matters most. You can also stretch the budget by cutting waste first, which is what Token Optimizer does.
Few teams budget for tokens yet, so for most of us this is still theory. It won't stay theory for long, and the teams that plan for the bill now will be the ones still running at full power when it arrives. And however many tokens you buy, someone still has to decide what they're for.
Marketing becomes the moat
That someone is a marketer, which is why I push back when marketers worry that AI will take their job. I think that's the wrong framing. We're emotional beings, so we're drawn to the most dramatic prediction, and what's actually happening is far more interesting. Jobs will change. Marketing won't be from humans to humans alone anymore, because agents will become a significant audience too, so we'll be marketing in more layers than before, and the job will keep evolving the way it has since there was a job called marketer.
I strongly believe that in this new world, marketing becomes the moat. When anyone can see a competitor's feature, copy-paste it, hit enter and ship it, what sets you apart is your brand and everything around the product.
So "Why do I need marketers?" is one of the biggest misconceptions about AI. When every company has the same agents, the same data and the same scrapers, the playing field is level. Letting an agent run your paid campaigns might be the right first step for some companies, and that's fine. Once you need your brand to stand out, to be known and to be chosen because people understand what you offer that others don't, you need human expertise in the picture, and that stays relevant even as AI models get better at knowledge work. You still have to be the expert. I also believe the people running marketing will eventually be the people running the companies.
And if you're feeling behind, stop chasing tools and trying to keep up with every piece of AI news. Start with your own reality. Ask where your knowledge lives today, where your teams are out of step with each other, what would make the work flow more smoothly and which bottlenecks you could remove. Start there. The tools will show up on their own when you really need them, like magic, but that groundwork is crucial and nothing can do it for you.
Key moments
- 01:45The big mindset shift: AI as infrastructure.
- 03:28Slop: the people pressing publish are still people.
- 04:59Marketing becomes the moat.
- 08:01An agent doing a task is just the task.
- 09:33AI as the new electricity, not a flashlight.
- 13:52Context gives your team and agents one source of truth.
- 16:27Why 2027 budgets need a line for tokens.
Alex Greenshpun is an AI strategist and builder and co-founder of 10x Company. She builds Total Recall, Token Optimizer and Outsourcerer.
