Table of contents
AI gives you the most probable answer, fast. That one fact explains both what it is brilliant at in marketing and where it will quietly make your work worse.
Two talks at the September 2026 Lenny & Friends Summit came at this from opposite sides. Elena Verna, who leads growth at Lovable, made the case for what AI lets one person do. Katie Dill, Head of Design at Stripe, made the case for what it can’t. Put together, they give marketers a practical way to decide which work to hand over and which to keep.
In short: AI is good at producing a competent, average version of almost any marketing task, quickly. It is not good at deciding what is worth doing, what makes your brand different, or when the work is actually good. Use it to raise the floor, and keep people on the ceiling.
Why AI defaults to average
Verna’s framing is blunt. She calls AI “average intelligence”: an average marketer, engineer, analyst and designer available whenever you need one.
“So, only one or two areas of exceptional craft are needed, and then the good enough everywhere else to get it to the point where you can learn.”
Elena Verna, Head of Growth, Lovable, at Lenny & Friends Summit
Dill describes the same limit from the other side. Large language models are good at telling you what has been popular and what has worked before. They are less good at telling you what is original, or what fits your brand and your customers. Her example was a restaurant website that the audience guessed was a software company. It sold Korean barbecue.
Both are describing a model that predicts the likely next thing. That is a strength when “likely” is what you need and a weakness when you need to stand out. We covered the mechanics in why your AI agents all have the same marketing idea.
What AI is good for in marketing
Covering the skills you don’t have
Most marketers are strong in one or two areas. AI fills in the rest to a “good enough” standard: a landing page layout when you are a copywriter, a SQL query when you are a content lead, a first-pass design when you have no designer. Verna runs pricing and packaging at Lovable herself, from research and prototypes to deploying changes and analysing results.
The point is not that the output is great. It is that it gets a campaign live, so you can learn from real customers rather than from a meeting. We wrote about the career and team side of this in the high-impact marketer and the generalist marketer.
Getting to a live campaign faster
Verna’s argument is that marketing teams don’t do things because they know the right answer. They do them to find out. The old route ran through an idea meeting, a brief, design, build, review and approval, often taking months. AI shortens that to idea, build, learn.
For marketers, that means more campaigns through the plan, launch, analyse loop in the same time. Speed only matters if each campaign is small enough to analyse and you actually look at the result.
Cutting the coordination
“When the cost of building falls below the cost of coordinating, the org chart should change.”
Elena Verna, Head of Growth, Lovable
A lot of marketing time goes on hand-offs: briefing an agency, waiting for a designer’s queue, getting sign-off. AI removes some of those steps. Verna’s warning is that buying AI tools without changing the approval chains around them wastes most of the gain. In her words, “when experiments are cheap, approval chains are the expensive part.”
Exploring more options
AI makes trying things cheap, so you can try more of them. Dill’s team designed the opening animation for the summit with AI’s help, and the designer went through 56 iterations to get it right. She points out that without AI he might not have attempted such a complex scene, or tried so many viewpoints. For marketers, the equivalent is testing ten headline angles or five creative concepts where you used to brief two.
Applying standards you have already written down
AI is good at following rules, as long as the rules reach it at the right moment. Stripe first gave its AI tools an MCP server with its design documentation. Dill says the results weren’t specific enough: three people could type the same prompt and get three different results. A command-line tool built on the design system, with full templates and flows rather than just components, made the output far more consistent.
The marketing parallel is your brand guidelines. A PDF nobody opens does nothing. The same rules written as skills that load when an agent writes copy are much more likely to show up in the output. See context engineering for agents for how to do this.
What AI is not good for in marketing
Having a point of view
“You have to have a point of view. If you don’t, AI will give it for you.”
Katie Dill, Head of Design, Stripe, at Lenny & Friends Summit
Without a point of view, AI fills the gap with the generic one. At Stripe, that point of view is optimism, and it shapes the colours, the writing and what they choose to write about. Your brand needs its own answer to “who do we want to be to our customers?” before AI can help express it. That answer is the basis of your standards, and it has to come from people who have noticed what your customers need, not just what they say.
Knowing when it’s good, not just done
Dill compares AI output to a microwave burrito: you go from hungry to lunch in 90 seconds, and the speed makes you overlook some serious flaws. A polished-looking ad or landing page appears in seconds, and it feels finished. That doesn’t mean it solves the problem or sets you apart.
The quality filter used to sit throughout the process, because teams could only resource one idea in 20. Now you can build all 20, so the filtering has to happen after the build, when saying no is harder. Dill’s answer is an editor: someone who looks at the whole customer journey and asks whether it hangs together.
In a Stripe design review, a colleague asked what the quality standard should be for something made with AI. Dill’s answer was that customers don’t care how it was made, only whether it is good. The ad in question came back with 17 fixes.
The details that show you cared
“One of the things that so offends us about AI ‘slop’ images is a sense that the details don’t matter. The cup is green, but it may as well have been blue.”
Nabeel Qureshi, “what makes art great”
Dill quotes this essay in her talk. AI picks details that are plausible, not ones that mean something. In marketing, that is the example in a case study that could belong to any customer, the stock phrase in a subject line, or the illustration that matches no idea in the copy. Customers notice when details are chosen on purpose, and they notice when they aren’t.
Owning the result
Dill’s third warning is that work this cheap starts to feel disposable. Nobody decides who maintains the page, updates the nurture sequence or retires the landing page once the campaign ends. AI can produce the asset. It can’t be accountable for it. Someone on your team has to own each campaign from launch to retrospective.
A starting split
Every team will draw the line differently. This is a reasonable place to start:
| Hand it to AI | Keep a person on it |
|---|---|
| First drafts of copy, ads and emails | The positioning and point of view behind them |
| Variations: headlines, angles, creative concepts | Choosing which variation to put live |
| Work outside your specialism, to a “good enough” standard | Your one or two areas of real craft |
| Applying brand rules you have written down | Writing and updating those rules |
| Pulling and summarising campaign data | Deciding what the result means for your goal |
| Critiquing a draft (a second agent) | The final edit and sign-off |
How to avoid zombie marketing
Dill opened with post-war architecture: builders copied modernism without its intent, and cities filled with identical, characterless buildings. Her worry is that AI-made products go the same way. Marketing is at least as exposed. Her recommendations translate directly:
- Improve your inputs. Don’t ask for “a landing page for a Korean barbecue restaurant”. Give it what you believe, what good looks like and your source material: customer interviews, win/loss notes, your best past campaigns.
- Encode your standards where the work happens. Brand voice, terminology and examples belong in skills and templates the AI loads, not in a shared drive.
- Push past the first draft. Treat the first output as the start. Ask what is generic about it and fix that.
- Use a second agent to critique. An adversarial reviewer catches the obvious problems before a human editor sees the work. Our guide to AI evals for marketers covers how to make this repeatable.
- Name an editor. One person decides whether the work is good, not just done, before it goes live.
- Protect the strange. If you lead a team, give people room to try things that don’t look like everyone else’s. Differentiated design has practical examples.
“Yes, AI can help us manufacture monotony, but it is also the greatest creative catalyst we’ve ever had.”
Katie Dill, Head of Design, Stripe
The bottom line
AI makes average marketing cheap. That is useful: it covers your gaps, gets campaigns live sooner and removes a lot of waiting. But because everyone now has the same average on tap, average is no longer enough to stand out. The scarce skills are a point of view, judgement about what is good, and the care that shows up in the details. Use the time AI saves on those, and on running more campaigns, rather than just making the same things faster.
About Growth Method
Getting the split right between AI and people is the difference between running more campaigns and producing more generic ones.
Growth Method is the practical way to adopt AI in marketing. Our agents suggest campaigns from your own data, then help you plan, launch and analyse them, with no AI vendor lock-in, security taken care of and humans in the loop. Nothing changes in your tools until someone on your team approves it, so the judgement stays with you.
We are on-track to deliver a 43% increase in inbound leads this year. There is no doubt the adoption of Growth Method is the primary driver behind these results.
Laura Perrott, Colt Technology Services
Get started free, and let AI raise the floor while your team works on the ceiling.
Frequently asked questions
What is AI good for in marketing?
AI is good at producing a competent, average version of almost any marketing task quickly. That makes it useful for covering skills you lack, getting a campaign live faster, cutting hand-offs between teams, exploring more creative options, and applying standards you have already written down.
What is AI not good for in marketing?
AI is weak at anything that depends on being specific to you: setting your brand’s point of view, judging whether work is good rather than just finished, choosing the details that show care, and owning the result over time. Those stay with people.
Why does AI marketing content all look the same?
Large language models predict the most probable answer, which means what has been popular before. Without a clear brief, brand standards and specific source material, they drift towards the same generic output for everyone. Katie Dill of Stripe compares it to the identical buildings of the post-war building boom.
What does “AI raises the floor, not the ceiling” mean?
It means AI makes average work cheap and fast, so the minimum quality anyone can produce goes up. The best work still depends on human judgement, taste and a point of view. Katie Dill argues teams should use the time AI saves to raise the ceiling as well.
How do I stop AI marketing output from being generic?
Give it specific inputs (your positioning, your customers’ words, your source material), encode your brand standards where the AI actually works, push past the first draft, use a second agent to critique the output, and make one named person responsible for editing before anything goes live.