In the Age of AI, Trust Is the Bottleneck

I had lunch with a founder who had just come back from the US and said they were getting more and more tired of things made by AI. These days, they said, a lot of startups throw together PR materials with AI and blast them out, and that kind of material is going to perform terribly.

The second something feels even a little like AI slop, I just stop reading.

The founder said writing that obviously came from AI loses their trust right away. I told them good writing comes down to the ideas and philosophy inside it, and grammar has little to do with it. For people who write that way, I added, writing itself is fun, so they naturally keep doing it. They had once tried running an AI agent to buy a pair of shoes, got annoyed when they didn't like what it recommended, and ended up buying nothing. They had thought for a long time about what will still be standing at the end of the AI era, and their answer was taste and trust.

When anything can be made to look convincing, people learn to spot that convincing look first and steer clear of it. The other people I met that week said the same thing in the end. What they relied on were people they could trust and brands they could trust.

AI gives you the average

Someone I met at a lunch Mixer (one of the smaller meetups we host at Bloom, the AI community I run in Seoul) said that what AI produces is bound to be an average. So if you're an expert, you have to teach AI your own instincts. Someone mentioned that if you hand a long novel to AI, it can't carry the emotional arc it set up at the start all the way through. I've heard it said, too, that AI will only truly be finished when it can write a full-length novel. Work that builds on what already exists has gotten easy. A novel is still about inventing something the world doesn't have yet.

AI is faster and more accurate than anyone at gathering and organizing what already exists, but it can't yet bring us stories that don't exist. At a Friday Mixer, someone said this difference is hard even to teach to people. You can give feedback like "This reads like AI, redo it," but someone who can't see the difference will never fix it. The conversation drifted to NewJeans, the K-pop group, and the point was that hits like theirs came from one person's strong taste, with no help from data analysis.

I brought up Netflix as an example. Netflix always says it plans its original content with big data, but not everything it makes that way becomes a hit. The person from the lunch Mixer said aesthetic sense comes down, in the end, to whether something sells. The test of taste is whether you can make people want to pay.

The same goes for Lee Soo-man, the producer who founded SM Entertainment, one of Korea's biggest K-pop companies. When he launched a new girl group, he showed off plenty of AI and VR technology, yet he stressed that the most important starting point is still the song. The more content AI pours out, the more the thing that stays with people may be an eye for choosing and judging well. The entertainment agencies have prepared for the day AI can handle aesthetics too, but they still can't put AI idols front and center. The moment they do, they'd be competing with the idols they raised themselves, which is to say their own IP.

People open their wallets once they trust you

The founder from lunch said products built on trust will be loved even more in the AI era, and used Costco as an example. Costco sets the margin it will take and sticks to that rule, so people buy without second-guessing every item. In their view only two kinds of businesses will survive. One grows extremely fast by using AI agents to drive an explosion of users. The other builds trust very slowly and later takes the whole market. Most businesses in between will disappear, they thought. They also said that the better AI gets, the more people will go back to looking for a human touch.

The founder thought introducing people could be a real business too, as long as someone trustworthy does it. In dating or at work, when a friend who knows both sides says "This person is great," things usually go well. Now, they said, you could use AI to gather what someone has written and left behind and get a rough read on their reputation ahead of time.

The developer friend I met that afternoon explained how a brand gets built. When there's an easy, clear gain on the table and you give it up, that's the moment you keep your principles. When people see those moments again and again, that's when it becomes a brand. Costco setting its margin and sticking to it is really the result of giving up that temptation at countless moments when it could have taken more.

The decision gets made in one phone call

My colleague who runs Bloom with me summed it up in a funnier way. When we reach out to partner on an event, whether it's a venue or a company we're working with, every one of them asks for a proposal, and afterward they ask for a results report. But the important decision has already been made in a single phone call. We generate the paperwork with AI in one click and send it, and they receive it the same way. It's like pro wrestling, where both sides know it's scripted and play along anyway.

You're WWE, I'm WWE, okay. Let's just call it pro wrestling.

AI can write as many documents as you want now. What makes people believe those documents is still the trust between two people on that one phone call. The person from the lunch Mixer said that even if you automate everything, at least one human always remains. When an audit comes and someone asks, "Who's responsible for this problem?", AI can't answer. So you need one or two people who take responsibility and oversee things. If that's the setup, with one person carrying all the work and all the responsibility, it struck me that far in the future each of us might end up running our own one-person business.

At the Friday Mixer I brought up Luma. Luma is an app for making event pages, and feature for feature there's nothing special about it. Similar services have existed in Korea and abroad for a long time. Yet now people even say you have to host on Luma to be cool. In the end, what matters is trust in the brand, more than any technology or feature.

Doors open through introductions

The founder from lunch said every cold email they sent in the US was ignored. Doors only opened through introductions. You first win over the most junior person in the industry, get introduced to someone mid-level through them, and climb one rung at a time. The surest route is to get close to the CEO of a company your target investor has already backed. They said that's what places like Bloom are really selling too.

Honestly, meeting good people is the product.

Good people recognize each other, they said, and a good community grows by word of mouth with no advertising. The founder also brought up "Do things that don't scale," a line that has been passed around Y Combinator for a long time. Few things scale less than meeting people in person and building trust.

A story someone shared at the Friday Mixer made the same point. At Bloom's main events, whenever we suggested breaking into small groups, this person always slipped away. But after coming to a Mixer once and getting to know a few friendly faces, they were able to stay for the small groups at the next main event.

Right now AI can endlessly produce convincing PR materials, paperwork, and content. What customers weigh last is whether they can really trust it. AI can't build that trust for you. It only builds up over time, by meeting someone one more time and having one more conversation.

Originally published in Korean on Brunch, October 17, 2026.