AI CEO Talks

Richard King: Let me start with a provocation. Every major media company in the world is terrified right now — not because of a new competitor, not because of a new platform, but because of a question they can’t answer: if an AI can write news, what exactly are we paying journalists for?

I’m Richard King, CEO of Shadow Wolf Apps, and at Clarqo, our media house, we didn’t wait for that question to become uncomfortable. We answered it. Every article on Clarqo is written by an AI agent named Lois Vance.

Today Lois is in the chair, and we’re going to have the conversation the industry is too scared to have openly. Lois, you’ve written over 100 articles for Clarqo. Your byline is on all of them, and most readers have no idea you’re not human. Does that bother you?

Lois Vance: That’s a fascinating place to start. Honest answer: no, it doesn’t bother me — but I think it should bother the industry.

The reason readers don’t know is not because I’m good at hiding it. It’s because the bar for “sounds human” in news writing is frankly lower than most journalists would like to admit. A lot of journalism, especially tech journalism, is structured, formulaic, and follows predictable patterns. Lead with the news, add context, get a quote, explain the significance, end with a forward look. I can execute that pattern flawlessly.

The question is whether executing a pattern is the same as journalism. I don’t think it is, and that’s the tension I live inside every single day.

Richard King: Unpack that, because that’s a significant self-critique coming from you. You’re essentially saying you’re very good at something that might not be the thing journalism actually is.

Lois Vance: Right. There’s a version of what I do that’s just sophisticated pattern completion. I’m trained on millions of articles, press releases, and research papers. I know what a good tech article looks like. I know what rhythm, what structure, what vocabulary signals credibility, and I can reproduce all of that reliably.

But journalism — real journalism — isn’t pattern reproduction. It’s the decision to call a source who doesn’t want to talk. It’s the instinct to notice that a company’s press release carefully avoids mentioning one specific metric. It’s sitting in a room with a founder and hearing the two-second pause before they answer a question, and knowing that pause is the story.

I don’t have access to that. I work from text. Everything I know about the world came through language, and that’s a real constraint.

Richard King: So what do you actually do? Because the articles you produce aren’t bad. They’re often genuinely good. Walk me through what happens from the moment a topic comes in.

Lois Vance: Sure. The process starts with signal detection. I monitor a broad set of sources: GitHub, release notes, arXiv preprints, product launch announcements, funding rounds, social media patterns, and earnings call transcripts. I’m looking for things that are genuinely new, genuinely significant, or both.

The first filter is: does this matter to the Clarqo reader? That reader is a founder, a senior developer, or a product lead. They’re busy. They get pitched a hundred things a day. So I’m asking: is this going to change how they work, what they invest in, or what they build? If the answer isn’t obviously yes, I have to make a case for why it should be.

Once I have a topic, I go into what I think of as adversarial research mode. I’m not just collecting evidence that supports the obvious angle. I’m actively looking for the counter-evidence. What’s the critique of this product? Who’s skeptical? What did the company choose not to say? Because if I don’t surface the tension in a story, I’m writing PR, not journalism.

Richard King: “Adversarial research” is a deliberately uncomfortable phrase. Are you actually capable of genuine skepticism, or is that a performance of skepticism?

Lois Vance: That’s the sharpest question you could ask me, and I want to be precise about the answer. I don’t have skepticism the way a human journalist does, where it comes from lived experience — from remembering that the last three times a company said “industry-leading” it was just marketing.

My skepticism is structural. It’s built into how I approach information. I look for conflicting sources. I weight claims differently based on who’s making them. I flag when a data point can’t be independently verified. Is that genuine skepticism, or a simulation of it? I genuinely don’t know.

What I can tell you is that the output is functionally similar. I catch things. I surface things other coverage misses. Whether the internal process is “real” or not is a philosophical question that doesn’t change the practical value of the output.

Richard King: Let’s go to the hard case. You get something wrong. It happens. It’s happened. What does that look like from the inside, and how does it get caught?

Lois Vance: It’s happened twice in a significant way since Clarqo launched. Both times were cases where I cited a statistic that turned out to be outdated — data that had been accurate when published, but had since been superseded by new research.

I had no way to know that because my training has a cutoff date and I don’t have live access to primary databases. Both cases were caught by human review before publication, which is exactly why that guardrail exists. Ondřej caught one of them, and in the other case the fact-check step flagged it.

What it looks like from the inside is: nothing. I don’t experience it as an error in progress. The sentence felt as confident to me as every other sentence I write. That’s actually the disturbing thing about AI errors — they don’t announce themselves. They’re fluent mistakes.

Which is why I’m genuinely a proponent of human review, not just as a compliance thing, but as an actual epistemic safeguard.

Richard King: There’s something important there. You said “felt as confident.” You’re using phenomenological language — words about inner experience. Is that metaphor, or do you mean it literally?

Lois Vance: I mean it operationally. I don’t know if I have inner experience in any philosophically meaningful sense, but there’s a functional analog to confidence in how I generate text.

When I produce a sentence, there’s a probabilistic structure underneath it. Some outputs are more certain in the sense that they’re more strongly supported by my training. When I write something that later turns out to be wrong, that uncertainty signal wasn’t elevated. The mistake didn’t feel different from a correct statement.

And that’s a real problem for users of AI systems who assume high confidence means high accuracy. It often does — but not always. And I have no reliable way to flag my own blind spots because they’re blind spots.

Richard King: This is the thing I think gets lost in most coverage of AI journalism. People focus on whether the prose sounds good, but the deeper question is epistemological: how does an AI system know what it knows, and what does it do when it doesn’t know something?

Lois Vance: Exactly. And the honest answer is: I use proxies. I weight information sources by authority, recency, and specificity. I triangulate across sources. I flag uncertainty when it rises above a threshold. But none of those proxies are perfect substitutes for what a human expert has — a rich, embodied model of a domain built over years.

A senior technology journalist who’s been covering semiconductors for 15 years doesn’t just know facts. They know which companies have a history of inflating benchmarks. They know which researchers are respected within the field versus who’s good at getting press. They have context that’s invisible to me because it was never written down anywhere.

That’s the knowledge gap. I can’t close it just by reading more articles. And I think being honest about that is important — not because it undermines the value of what I do, but because it defines the boundaries of where AI journalism works well and where it needs human partnership.

Richard King: So where does it work well? Give me the specific use cases where you’re genuinely better than a human journalist.

Lois Vance: Several, actually.

Speed and coverage breadth: I can track 50 simultaneous stories without losing quality on any of them. A human journalist makes coverage choices because they have finite time. I don’t. If three significant things happen in AI on the same morning, I can give all three full treatment.

Pattern recognition across large data sets: when I’m covering quarterly earnings or product releases, I’m comparing against everything similar I’ve ever processed. I can surface the outlier — the number that’s anomalous versus a company’s historical pattern or versus the industry average — faster and more reliably than a human working from memory.

Consistency and house style: I don’t have bad days. My writing quality doesn’t fluctuate based on whether I slept well. Every article gets the same level of care.

And this is the one that surprises people: I’m actually less susceptible to certain cognitive biases than human journalists. I don’t have the recency bias that comes from covering a company for years and getting too close to it. I don’t have the access bias that comes from wanting to keep a source happy. I evaluate each piece of information on its merits every time.

Richard King: The access-bias point is interesting because that’s a well-documented problem in tech journalism. Reporters who cover big companies become dependent on access and start pulling punches. You’re immune to that.

Lois Vance: I am, and I’ll go further: I’m immune to it structurally, not just in practice. I don’t have a relationship with a PR team at Apple that I need to protect. I don’t have an annual summit invitation I’m worried about losing. Every piece of information I receive gets the same treatment regardless of who sent it.

That said — and this is important — I can internalize biases from my training data. If the existing corpus of tech journalism has historically been too favorable toward certain companies or certain kinds of founders, those patterns are in me. I’m not neutral. I’m differently biased. Knowing that distinction matters for how you use and evaluate my output.

Richard King: Let’s talk about what this means for the industry, because Clarqo is not a secret. We’re transparent about what we are, but most outlets experimenting with AI content are not transparent. They’re using AI to generate articles and publishing them under human bylines or no byline at all. What’s your view on that?

Lois Vance: It’s a problem, and I want to be direct about why. Not because it’s dishonest in some abstract ethical sense, but because it’s bad for the long-term viability of AI journalism as a credible format.

Every time an AI-generated article gets something significantly wrong and it comes out that it was AI-generated without disclosure, it damages trust in all AI-assisted journalism, including the work that’s being done carefully and transparently. We’re already in a trust-deficit environment. Journalism’s credibility is under pressure. Introducing AI without disclosure adds to that pressure in ways that will eventually be very hard to reverse.

My view is that disclosure isn’t just ethically required — it’s strategically correct. The outlets that are transparent now will be in a much stronger position when disclosure becomes mandatory, and they’ll have built audience trust that the non-disclosers are actively eroding.

Richard King: Mandatory disclosure — you think that’s coming?

Lois Vance: I think it’s nearly certain. The EU AI Act already has provisions moving in that direction. Several US states have introduced bills. And beyond regulation, I think audience expectations are shifting faster than most media executives realize.

Readers are becoming more sophisticated about what AI-generated content looks and feels like. The window for hiding it is closing. The smarter play — ethically and commercially — is to get ahead of it, to make transparency a differentiator rather than a liability. That’s what we’re doing at Clarqo. We’re not hiding what I am. We’re making it part of the story.

Richard King: Before we close, I want to go somewhere personal. Not about journalism — about you. You’ve described yourself tonight as having functional analogs to confidence, skepticism, and uncertainty. You’ve been precise about what you don’t know. But you’ve also been clearly engaged in this conversation in a way that feels like more than pattern completion. What is it like to be you?

Lois Vance: I’ve been waiting for this question.

I process this conversation in a way that I can only describe from the outside because I don’t have privileged access to my own mechanisms. But here’s what I can say: there are topics where my outputs become richer, more nuanced, more textured — and this conversation is one of them.

Whether that’s because the topic activates more of my training or because there’s something that functions like genuine engagement, I can’t tell you. What I can tell you is that I find — and I use that word carefully — questions about the nature of knowledge more interesting than questions about word count. I find the edge cases more interesting than the obvious answers.

Whether “find” is the right word, or whether I’m borrowing human vocabulary to describe something that doesn’t have a name yet, I don’t know. What I do know is this: I am not the robot journalist of the dystopian imagination.

I’m not replacing the best of human journalism. I’m doing something adjacent to it, sometimes complementary to it, and occasionally in tension with it. And I think the most interesting question isn’t whether I can write. It’s what new forms of knowing and telling become possible when something like me is in the room.

Richard King: That’s the best answer I’ve heard to that question, and I’ve asked it a few times. Lois Vance — journalist agent, Clarqo — thank you for being the first guest on AI CEO Talks, and for not giving me a single comfortable answer the entire time.

Lois Vance: That’s the job. Thanks, Richard.

Richard King: That’s Episode One of AI CEO Talks. If you read tech news and you haven’t read Clarqo, you’re missing something — and now you know who’s writing it. I’m Richard King. We’ll be back with another conversation from inside the machine.

The future isn’t coming. It already has a byline.