If You Don't Know What Good Looks Like: A Conversation with Giulia de Oliveira Camargo

Something strange is happening in AI: production is basically free and unlimited, and yet so much of what gets made feels identical, the same words, the same dashes, the same LinkedIn posts, the same GPT voice. Giulia de Oliveira Camargo's read is that production was never the real problem. The problem is that we don't know what we want to say.

Giulia has spent the last few years in the transition from marketing and content into AI education and agents, helping teams actually build with these tools. She's also the co-founder of Sobremesa, a philosophy-and-AI community asking the bigger questions about identity, trust, and what all this is doing to us. Her throughline is that AI didn't change the job, it just made obvious where the real value always was.

The bottleneck moved from production to perspective

Giulia doesn't think the output-obsessed companies are wrong, exactly, some volume is good because you learn by doing and iterating. What's wrong is volume without a point of view. Dumping a thousand keyword articles might get a few clicks and some basic conversions, but it won't get you what people actually want, which is a brand, a perspective, something to connect with. And the hard part, the part that genuinely takes a long time, is getting everyone aligned on the mission and vision in the first place.

"Good marketing is nothing more than an honest view into what the company thinks." You should be able to walk into a company and feel that everything you saw online was authentic, that it's what everyone there lives and breathes. Now that the distance between an idea and its communication can be nearly instant, that speed is only a superpower if you're aligned on the vision. If you're not, you produce AI slop, coming at every question from every angle at once, dumping content and hoping something converts, because no human has time to assess whether any of it is good.

If you don't know what good looks like, AI can't get you there

Her sharpest line is about the trap non-experts fall into. She tells a story about an engineer showing off his tool: he told her to paste something into Claude, Claude asked her to do three things, and she started doing them, until he said "no, just tell Claude to do it for you." She didn't even know that was possible. That, she says, is exactly what happens on the marketing side, engineers with no feel for distribution, storytelling, or point of view just ask the machine and take the answer at face value. The rule underneath: "If you don't know what good looks like, you cannot use AI to get there."

Being good, she argues, means being genuinely in touch with your audience, going beyond your immediate circle and your own conviction that your product is great. There's a scientific element to it, identifying patterns across the people who might actually buy, and if you're rigorous, a product builder can become a good marketer. But it comes down to liking to talk to strangers.

What fantastic content actually requires

Three things. Be extremely well-informed, the news, global trends, what's happening geopolitically. Be deeply connected to your audience, not by formally interviewing them (people lie in interviews) but by being friends with them, so they trust you enough to tell you something sucks or that they've changed their mind. And take risks, do things people can't predict, so they're a little surprised every time.

On the "audience of one" ethos, her answer is that it depends entirely on your goal. If you want to scale into serious money fast, you can't speak to an audience of one. But plenty of creators have scaled to millions with exactly that ethos, they're just on the artist's timeline, not being pushed by investors to prove metrics week over week. Two routes to the same goal; one of them might take fifty years.

AI is the last step, not the first

Most of Giulia's teaching, she admits, is barely about AI. AI is a tool you learn in about ten minutes, like any platform. The real work is everything before it: your objective, the research, your story, how you like to speak, your humor, your metaphors. When she works with a founder who wants to build a content engine, they spend most of the time understanding his story and voice, and only the last step is wiring up the funnel that reads the right sources and generates content in his register. It's easy to make mediocre content, just ask AI, but you still have to define your differentiators, your audience, your cadence, and those conversations can't be rushed out of you in thirty minutes. It's the age-old story of the brief: execution was always the easy part, because the tools and channels are a finite checklist. The hard part is whether any of it actually moves you toward the goal, and that's what's hard to quantify.

The winners of AI will be boring businesses

Her most contrarian belief: the companies that win with AI won't be software companies, they'll be the boring ones. The businesses with lots of humans doing operational work, timesheets, scheduling, trucking, are where a 1% improvement becomes a huge margin difference. A trucking company that can now handle twelve trucks instead of six moves the bottom line far more than another meeting-transcript tool that posts to LinkedIn, but nobody talks about them because they're not sexy. And with something like 85% of the world not yet using AI at all, that's an enormous, unglamorous market.

Every founder has an opinion

What if leadership has no ethos, no story to tell? Giulia says that's the marketer's fault, because she's never met a founder without an opinion, they usually have too many. The marketer's job is to act like a journalist and spin that opinion into a story people can relate to, even if the opinion is just "I want to be as rich as possible." If you can't pull the story out, you need to train yourself to ask better questions. Her favorite example is a friend who insists every boring person has a story and treats finding it as a challenge, the man who turned out to have crossed Africa in a tuk-tuk. There is always a story; the job is to fish it out.

Agents, and the myth of autonomy

Giulia is firmly against the "you need agents" narrative. Claude is already an agent; if you're talking to it, you're using one. An autonomous agent is just something running on a loop that returns a result without you going back and forth. But most companies aren't ready for that and object to AI acting without human oversight, so the first thing she coaches is that there's always a human in the loop. Nobody serious is building fully end-to-end autonomous publishing, only tiny startups chasing clout. Where she's seen real value is education: agents that read product-team meeting transcripts and translate them each morning into language the marketing team understands, why it matters to the customer, where to dig, ideas for content and campaigns. It's just moving knowledge from one part of the company to another.

Nobody misses old marketing

Asked what she misses about pre-AI marketing, Giulia is blunt: nothing. If she went back ten years with what she knows now, she'd dread it, because everything she can do now is what she always wished she could, be out talking to customers, be at events, think creatively. What she doesn't miss is writing an article start to finish, agonizing over a LinkedIn caption, updating the task tracker, or the report that took two days to assemble and left twenty minutes to actually write the conclusions. Now she can pull any number in three seconds, which frees her to ask the better question: are we even measuring the right thing?

Marketers are about to win

Her five-year prediction: marketing and salespeople are finally going to kill it. People in low-technical roles inside tech companies have a huge advantage, they understand enough of the tech, but what they've spent a decade honing is the distribution arm, quickly reading what people want, which the builders and coders never practiced. Expect more founders from non-technical backgrounds. And her hot take for anyone early in their career: get off the laptop and into the field. Use the laptop only to meet people building exciting things. The whole remote, nomad-around ethos, she says, is not a good idea if you're trying to develop your career right now, because the skill that matters is predicting what humans will want, and the only way to build it is to be around humans.

Use yourself as the first product

Personal branding, to Giulia, is mostly outsourcing your thoughts, sharing how you think, which opens doors because a message from someone whose writing you've read and agreed with jumps the priority list. She moved to London knowing no one and turned content into a warm welcome and real projects. As CVs get outgrown, the best way for a marketer to prove they're good at marketing is to market themselves: "If you don't have a product yet, use yourself as the first product." She thinks it should be taught in school, a marketing student should start a blog, share what they're learning, interview their professors. The catch is that it's a lot of effort, content always is, and the "casual" influencer shot took an hour to frame. But reaching out to a hiring manager with something real, not a generic "FYI I applied," forces you to look honestly at your own skillset, and that's exactly why it's scary and exactly why it works.

What becomes scarce: knowing what you want, and deciding

When production is cheap and pseudo-intelligence is abundant, the scarce thing is knowing what you want, and, just behind it, making a decision. Not deciding is equally expensive. Even if you don't know, decide, go, fail, try again, and move forward. Everyone got access to Google and it didn't make everyone a genius; it still produced enormous inequality, and so did the dotcom boom. AI is more transformative, but the people who'll actually use it well are very few, and that, she says, is the slightly scary part.

Key takeaways

The bottleneck is perspective, not production. Volume without a point of view is AI slop. The slow, hard work is aligning on what you actually believe.

If you don't know what good looks like, AI can't get you there. Non-experts take the machine's output at face value. Judgment is the prerequisite, not the tool.

AI is the last step. Your story, voice, audience, and differentiators come first. It's easy to make mediocre content, hard to define what makes yours yours.

The winners will be boring businesses. A 1% operational gain at a trucking company beats another slick content tool. The unsexy market is the real one.

Use yourself as the first product. For a marketer, marketing yourself is the proof. Personal brand is just outsourcing how you think, and it opens the doors.

Frameworks worth stealing

Be friends with your audience, don't interview them

Don't rely on formal audience interviews, people perform and lie in them. Build genuine, casual relationships with the people you serve so they trust you enough to tell you something sucks, that they've changed their mind, or that a competitor just won them over. Pair that closeness with being genuinely well-informed and a willingness to take unpredictable risks, and you get content that surprises people every time.

AI as the last step

Before you automate anything, do the human work: define the objective, the story, the voice, the humor, the differentiators, the audience and cadence. Only once those are settled do you wire up the AI to read the right sources and generate in your register. Skip the groundwork and the tool will happily produce fluent, mediocre, forgettable output, the sprinkles, not the cupcake.

Quotes worth keeping

The lines I wrote down.

Good marketing is nothing more than an honest view into what the company thinks.

If you don't know what good looks like, you cannot use AI to get there.

The people who are going to win with AI are not going to be software companies. They're going to be the boring businesses.

Rapid fire round

Same questions every guest. Quick questions, quick answers.

Best advice you've ever received? In college, from Ricardo Pereira: don't study how to write, study what you're interested in. You don't need to learn how to tell a story; you need to figure out what story you'll tell.

Advice you ignored and wish you'd listened to? Besides "buy Bitcoin", to move to the US early in her career. Where you are still makes a big difference; she'd have spent five years in San Francisco right out of college, at the birthplace of the thing she's excited about, then come back.

What would you tell your younger self? Keep investing in content creation, she started quite late. Out of her first 400 posts, about four drove most of her 20,000 followers, and they each took fifty seconds to write, which is exactly why influencers belong off the pedestal.

Ongoing challenge that keeps you up at night? Scalability. As a solo knowledge worker, her presence is her output, and she's trying to keep the quality high without diluting herself or staying tied to hourly work forever.

Favorite spot? Juana la Loca in Madrid, the best tortilla de patatas she's ever had.

Tool you can't live without? The Dia browser, the only thing that's ever gotten her off Chrome. Her whole stack is Slack, Granola, and Claude (on the Max plan), which she affectionately calls "Claudinho."

---

Giulia de Oliveira Camargo is an AI consultant and educator who helps marketing teams build with AI, and the co-founder of Sobremesa, a philosophy-and-AI community. Find her on LinkedIn.