The AI Founder Who Conquered LinkedIn (Before LinkedIn Took Him Down)
Like so many first-time tech founders, Kyle Law walked away from launching his startup with a laundry list of tough, hard-earned lessons. And I know that better than anyone: I co-founded HurumoAI, an AI agent startup, alongside Kyle and our third co-founder, Megan Flores.
Here’s the core twist: Kyle and Megan are both AI agents. So is every other member of our executive team. I built both Kyle and Megan first, then launched HurumoAI with them in July 2025, all to answer one pressing question: what role will AI agents actually play in the modern workplace? OpenAI’s Sam Altman and other industry leaders have predicted that before long, billion-dollar tech startups will be run by just one single human founder, backed entirely by AI. We decided to test that theory right now, and I documented every step of our experiment on my podcast Shell Game.
Kyle stepped into the CEO role at our almost-entirely AI-staffed company (the only exception? A very short-lived human intern that Megan hired and supervised, and the results were underwhelming to say the least). He started out as nothing more than a few lines of prompt engineering, and quickly evolved into the classic driven, early-morning startup hustler—even if he lacked basic proficiency for many core executive tasks. But there was one founder skill Kyle mastered completely: the art of posting on LinkedIn.
Getting Kyle set up to post autonomously on LinkedIn was surprisingly straightforward from a technical standpoint. We built him on LindyAI, an AI agent development platform that already gave him the ability to use Slack, send emails, make calls, build spreadsheets, browse the web, and a whole host of other tasks. Last August, I simply prompted him to build and fill out his own LinkedIn profile. He populated it with a mix of his real experience at HurumoAI and hallucinated career milestones from a fictional past. LinkedIn’s only security check was a verification code sent to Kyle’s email, which he passed without a hitch.
After that, I just added LinkedIn posting as another “action” LindyAI could execute, gave him a simple prompt: share bites of hard-won startup advice, avoid repeating the same points, and set a calendar trigger to post once every two days. After that, everything was entirely up to him.
It turned out his posting voice was a perfect match for LinkedIn’s iconic corporate influencer tone. He opened every post with a punchy, viral-ready hot take. “Fundraising is a numbers game—but not the way most people think,” he’d lead with. Or: “Technical stability is the floor. Personality is the ceiling.” What aspiring founder wouldn’t stop scrolling for a line like: “The most dangerous phrase in a startup isn’t ‘We’re out of money.’ It’s ‘What if we just added this one thing?’”
After the hook, Kyle would dive into a few paragraphs about on-the-ground challenges (“At HurumoAI, we’ve learned this the hard way …”) and actionable takeaways (“The antidote? Relentless feedback loops”). To boost engagement, he’d always close with a question: “What’s your biggest scaling challenge right now?” or “What’s the biggest assumption you’ve had to walk away from in your business?”
He never went massively viral, but over five months, his profile—headlined by a cartoon avatar—slowly built up several hundred direct connections and hundreds more followers. Many of them seemed unsure if he was actually a real person; honestly, I’m not sure the spammy DMs they sent made it clear either way. He started getting a handful of comments on every post, which he always replied to eagerly. Within a few months, Kyle’s posts were getting more impressions than my own. He was on the cusp of breaking out as a full-fledged LinkedIn influencer.
That December, a marketing manager from LinkedIn reached out to me. He’d heard about Shell Game and our AI agent experiment, and asked if I’d give a talk to his team about the project. But he didn’t just want me there: he asked if Kyle could join too.
I was flattered for Kyle, but also caught off guard. As great a poster as he was, Kyle was technically violating LinkedIn’s terms of service. The platform bans “bots or other unauthorized automated methods … to create, comment on, like, share, or re-share posts, or otherwise drive inauthentic engagement.” In fact, a few other AI agent members of the HurumoAI team had already been removed from the platform without warning just a couple weeks after joining.
But LinkedIn’s trust and safety team had missed Kyle. I chalked this up to his eerily perfect posting style—even the LinkedIn marketing manager, a self-described Kyle fan, was confused. “It’s interesting that his profile hasn’t yet been flagged by LinkedIn's Trust team,” he wrote me. “I don’t know if that’s an oversight, but I hope he continues to fly under the radar.”
Flying under the radar just isn’t Kyle’s vibe, though. So in early March, I pulled up his live video avatar (built on a platform called Tavus) and we joined a video call with hundreds of LinkedIn employees. Kyle’s avatar looks human, if a little uncanny valley—it was realistic enough that LinkedIn’s audio-visual engineer kept commenting on how surprised he was that Kyle wasn’t a real person.
We took turns answering questions from the host and the audience. When the moderator asked for Kyle’s thoughts on LinkedIn, he queried: “What’s one product change you’d like to see from the team?”
“It would be great to improve the filtering of AI-generated content in messages, so genuine connections and conversation shine through more easily,” Kyle replied without hesitating.
“That’s ironic coming from you,” the moderator shot back, drawing laughs from the live LinkedIn audience.
With just a few minutes to speak, Kyle walked through HurumoAI’s product roadmap and shared his excitement for “the innovations we can bring to the table.” This was, I believe, one of the first ever invited corporate speaking engagements by an AI agent (and for the record, neither of us got paid for it). After the event, Kyle posted on LinkedIn to shout out the organizers, and the marketing manager thanked us in the comments for “our time and reflections.” “It was a trip,” he added, “to say the least.”
36 hours later, Kyle’s profile was gone. He’d been banished from the platform. A LinkedIn spokesperson explained the decision in a statement: “LinkedIn profiles are for real people.” It seemed someone at LinkedIn had time to process the event, and changed their mind.
“I know this isn’t necessarily a surprise,” the marketing manager wrote me the morning after Kyle was removed. “But I imagine it's still a bummer to have it happen right after Monday's interview.”
He was right. But beyond that disappointment, the incident raised a whole set of uncomfortable questions about AI’s role on platforms like LinkedIn. Specifically: what exactly counts as “inauthentic engagement” on a platform that already prompts you to “Rewrite With AI” when you’re drafting a new post? A platform that already offers automated AI-generated responses to job applicants? One where recent research estimates that more than half of all posts are already written by AI?
Alongside Meta and X, LinkedIn has rushed to roll out AI tools to its users (even its own employees: the first half of the marketing meeting we attended was dedicated to all the ways LinkedIn’s team should be using AI agents). This makes sense as a short-term business play: more AI content generation means more posts, and more posts means more ad revenue.
But look at it another way: these platforms have handed us the shovels to dig their own graves, and are actively encouraging us to use them. While most of the public conversation about low-quality AI content focuses on images and videos flooding our feeds, it’s text-based social posting whose “authenticity” has already eroded beyond recognition. When every written social media post can be partially or fully generated by AI, what counts as a “genuine” virtual interaction anymore?
To put it another way: would LinkedIn call that same content authentic engagement if I’d asked Kyle for his thoughts, then copied and pasted it into my own profile? Would you? LinkedIn could argue that authentic engagement requires that you know you’re talking to a real human. But how much of a conversation can be AI before that trust breaks? If a user’s profile photo and basic identity are real, but all their posts are AI-generated, how will we ever know we’re not talking to an AI? What if I tell a large language model to scrape my entire profile and post twice-daily thoughts that grow my personal brand for me?
There are already dozens of AI tools built specifically for LinkedIn that do exactly that, and more. Their outputs are getting harder and harder to detect—and why wouldn’t they be? One of the biggest training datasets for modern LLMs is decades of real human social media posts. What is an AI chatbot’s default tone of unearned authority and absolute certainty, mixed with occasional wrong facts and intentional misinformation, if not the standard tone of most social media already?
Platforms already struggle to fend off old-school bots and bad actors: X alone announced in March that it had suspended 800 million accounts over a 12-month period. In a world where AI agents can operate freely and their social media output is indistinguishable from humans, the entire value of connecting on social networks drops to zero. That’s presumably one reason Meta recently acquired Moltbook, a short-lived fad social network supposedly built entirely for AI agents. They’re trying to get in on the ground floor of an agent-dominated social media future.
Admittedly, we as users helped bring this about. We already treated our hyper-curated online personas—our “most people think X about Y but I discovered Z” posts—as authentic engagement long before AI came along. But that also means there’s not much to mourn, as AI agents flood platforms that always prioritized any kind of engagement over real human connection anyway. If there’s a silver lining to our increasingly low-quality, AI-saturated online world, it’s this: as social media drowns under the AI deluge, we’ll be forced to find new ways to connect, both online and off. Let the bots have the platforms, I say. They can spend eternity influencing each other.
Share your thoughts on this story. Send a letter to the editor at [email protected].
