The AI Native Influence Stack
The most valuable influencer in the AI-native era is not on LinkedIn, is not a paid creator, and is usually an employee of the company they evangelize. Influence flows from the build, not the brand budget.
The Setup
U.S. influencer marketing is on track to spend $13.7 billion in 2025. Nearly a quarter of brands now allocate more than 40% of their marketing budgets to creator partnerships, and the dominant playbook is familiar: external creator, multiple sponsors, TikTok Shop, LinkedIn for B2B, traceable ROI.
Now look at the AI-native shelf. Claude Code reached a billion dollars in annualized revenue within six months of going public. Anthropic crossed thirty billion in annualized revenue by April 2026, more than tripling its run rate in four months. Almost none of that growth came from external creators. It came from the team's own engineers shipping in public.
The contrast is the article. In the AI-native era, the influencer is internal, technical, and permanent. The audience is the builder, not the buyer. The channel is X and GitHub, not LinkedIn or a brand TikTok account. The unit of trust is a merged pull request, not a sponsored post. Hardly any of the standard influencer-marketing dashboards measure this kind of influence at all, which is exactly why so many brands miss it.
I've been watching this pattern accumulate for two years across the people I learn from daily. There are nine of them in this article. Their channels do not look like a marketing budget. They look like a build queue.
The Closed Feedback Loop
Boris Cherny is the canonical example. He created Claude Code at Anthropic, shipped it publicly in mid-2025, and now runs the team. He has 426,000 followers on X and roughly 500 connections on LinkedIn. The asymmetry is not an accident. It tells you exactly where his customers live.
What makes Boris's channel work is the cadence, not the count. A bug surfaces in an X thread. He replies inside the hour. The fix lands in the next release. The thread closes with a screenshot of the merged PR or the changelog entry. The customer walks away with their issue resolved and a public artifact of how Anthropic's leadership engages with builders. The next person watching that thread updates their priors about the product without ever talking to a sales rep.
Boris's February 2026 conversation with Lenny Rachitsky on the "What Happens After Coding Is Solved" episode broke 367,000 views on X within twenty-four hours of release. The most quoted line was not about Claude Code's roadmap. It was that Boris had not written a line of code by hand since November and was still shipping ten to thirty pull requests a day. Influence here is the side effect of a working loop: build, ship, post, listen, fix, post again. Marketing sits inside the loop, not outside it.
This pattern explains why Anthropic now reports more than a thousand customers spending over a million dollars a year, double the count from two months earlier. Nobody bought Claude Code because of a billboard. They bought it because they watched the engineer running it answer their question on a Saturday.
The Channel Inversion
Andrej Karpathy puts the joke right on his LinkedIn profile: "I don't use or check my LinkedIn, I only have an account for searching/hiring. Please use my X/email to contact." He has 280 LinkedIn connections. He has 2.3 million X followers. His pinned post reads, "The hottest new programming language is English."
That joke is the entire shape of the channel inversion. LinkedIn rewards polish, headlines, and hiring funnels. X rewards code, crisp opinions, and screenshots of training loss curves. GitHub rewards the only thing that actually compounds: working software in public. Karpathy's nanoGPT sits at 57.2K stars as of this week. nanochat at 52.5K. Each repo carries a tutorial, a YouTube companion, and the implicit promise that you can read the code and reach the bottom.
Addy Osmani runs the same playbook from inside Google. His agent-skills repository sits at 22.9K stars, encoding the workflows and quality gates senior engineers actually use, written so a junior can run them tomorrow. He has more than 260,000 LinkedIn followers, but the LinkedIn following is downstream of eighteen books and a decade of GitHub artifacts. The book and the repo are the durable surfaces. The social channels are amplifiers.
GitHub stars are the trust currency in this stack. They aggregate slowly, they cost nothing to grant, and they cannot be bought without somebody noticing. Three thousand stars on a tool that fits a real workflow beats three hundred thousand impressions on a thought-leadership post, every quarter, every time. If you want to know which way developer attention is flowing, do not read the LinkedIn algorithm. Read the trending tab on GitHub.
The Meme as Industry Catalyst
Karpathy coined "vibe coding" in early 2025. By November, Collins Dictionary had named it the Word of the Year. By the time Claude Code's marketplace crossed nine thousand plugins in February 2026, vibe coding was the de facto job description for an entire role inside frontier labs. One sentence on X moved a category.
Jensen Huang has been doing this for a decade at higher altitude. "AI Factory" gave every CIO a procurement frame. "Accelerated computing" gave every CFO a permission slip. His 2026 GTC keynote ran past two hours, kept the audience through the overrun, and turned a single demo into a market-moving moment for his entire supply chain. The format is consistent every year. Inspiring future-state opening, hardware and software middle, robots on the closing stage. The format is the meme.
Memes work because they compress a worldview into something that survives a hallway conversation. They route around the slide deck and the press release. They are also nearly impossible to back-engineer through a brand agency because they have to come from somebody who actually believes the worldview they are compressing.
In an AI-native company, the people most likely to coin the meme that moves your category are the people who built the product. Get them on stage. Get them on X. Stop sending them to media training.
The Closed Loop Is an Org Design Choice
Most enterprises trying to copy the AI-native influencer pattern miss the most important part. The closed feedback loop is not a content strategy. It is an organizational design choice.
The evangelist has to be technical enough to fix the bug, senior enough to ship the fix, and visible enough that customers know who to tag. That role does not bolt on. It has to be designed into the shape of the company from early. Anthropic and the AI-native cohort are running the play at higher velocity than the prior generation because the channels move faster and the artifacts are smaller, but the structural choice is the same. The person who builds is the same person who posts is the same person who replies.
If you are building an AI-native product right now, the question is not "who is our influencer agency?" The question is "who on the team is technical, senior, and present in the channel where our customers live?" That person is your distribution. If nobody on the team can answer to that description, you do not have a distribution gap. You have a hiring gap.
Recommended by LinkedIn
The Founder-Distributor Pattern
Garry Tan, President and CEO of Y Combinator, runs the most aggressive version of this pattern in the early-stage market. He posts on X constantly, surfaces portfolio launches in the same channel, and the YC megaphone now functions as the primary discovery surface for AI-native startups. A "talked-about-by-Garry" mention has become an early-stage signal that founders chase the way they used to chase TechCrunch.
What makes Tan a clean illustration of the builder-president archetype is that he ships open source himself. GStack landed on GitHub in early 2026 and now sits at 83K stars and 12.1K forks: an opinionated set of twenty-three Claude Code skills that turn a solo engineer into a virtual startup team. GBrain followed on April 9, 2026 under MIT license, currently at 11.3K stars. It is Tan's actual daily AI memory system, ten thousand markdown files, three thousand people pages, thirteen years of calendar data, twenty active cron jobs, all running against his own knowledge graph. He did not commission a thought-leadership piece about agentic memory. He shipped his own.
Elon Musk runs a louder version of the same loop on the company side. xAI shipped Grok 3 in February 2025, integrated it into the X timeline, rolled it into Tesla vehicles by mid-2025, and announced an Office plugin track in April 2026. SpaceX's acquisition of xAI in February 2026 consolidated the operating model. The product, the platform, and the founder's posting account run on the same surface. Releases land in users' feeds within minutes of going live. The marketing channel is the product.
Both leaders show what happens when the CEO is also the distribution. Skip the politics, watch the structure. The release cadence and the posting cadence collapse into one thing.
The Analyst Tier
Not every influencer in this stack is a builder. Dylan Patel runs SemiAnalysis, the only newsletter most AI infrastructure teams pay for. His coverage of the semiconductor supply chain, training runs, and compute pricing routinely sets the priors that every frontier lab argues against in private. When Patel publishes a number for the cost of a GPT-class training run, every CFO at a model company recalibrates against it the next morning.
The analyst tier matters because it gives the closed feedback loop a counterweight. Builders are biased toward their own roadmap. Patel is paid to be biased toward the truth. In a category that moves this fast, both signals are necessary. The credible analyst becomes a kind of public-square referee, and the AI-native era has produced very few of them. That scarcity is itself the moat.
The Thesis Bet
Leopold Aschenbrenner published "Situational Awareness" in June 2024. The essay ran fifty-one thousand words across one hundred sixty-five pages and predicted the decade of AI ahead. He then did something almost no essayist does. He stood up a hedge fund and traded against his own thesis.
Situational Awareness LP grew from roughly $254 million in Q4 2024 to more than $5.5 billion in U.S. equity exposure by Q4 2025, across nearly thirty holdings. The top of the book reads like a literal financialization of specific essay passages. Bloom Energy as the largest position, on the bet that fuel cells become primary on-site power for AI data centers. CoreWeave through both common stock and a six-hundred-percent expansion of call options, on the AI cloud compute thesis. Core Scientific and Applied Digital on the crypto-miner-pivots-to-AI-hosting trade. The essay-to-portfolio mapping is one of the tightest you will see in modern markets.
Peter Steinberger sits at the other end of the spectrum, betting with code instead of capital. He created OpenClaw in late 2025, an open-source autonomous agent execution engine. As of this week the OpenClaw GitHub repo stands at 364K stars and 74.5K forks. NVIDIA partnered with him to build a hardware-accelerated variant called NemoClaw, and Jensen Huang showed OpenClaw's growth curve on stage at GTC 2026 with the line that every company now needs an OpenClaw strategy. Steinberger then joined OpenAI, with OpenClaw continuing as an open foundation. One builder, one repo, one keynote shout-out, one industry pivot.
Both arcs share a structure. Public thesis, public artifact, real conviction backed by real consequence. Whether the consequence is a 13-F filing or a GitHub repo, the underlying move is the same: refuse to let the influence float free of the bet.
The Takeaway
An AI-native builder is the best influencer for another AI-native builder. The shape of that influence is concrete: a closed feedback loop on X, a public artifact on GitHub, a long-form essay or a hands-on book, occasional appearances on the long-form podcasts that other builders listen to. The brand budget is not the input. The build is.
If you are running marketing inside an AI-native company, the move that pays back fastest is to stop hiring external creators to talk about your product and start clearing calendar for the engineers who already built it. Give them platform support, not media training. Give them ownership of the thread, not approval workflows. Measure the channel by merged PRs and customer conversations, not impressions.
If you are an operator without a public channel yet, pick the artifact you can ship in the open and start. A skill, a repo, an essay, a benchmark, a book. The artifact is the influence. Everything else is amplification.
This is the first article in a parallel arc. The main eleven-part series covers how AI-native organizations are built. This arc covers who shapes them in public.
This is article V01 of the AI-Native series Voices arc. The main engineering arc continues at articles 1 through 11. Next in this arc: builder-CEOs and the founder-distributor stack.
Follow @manavsehgal on X for the thread version and ongoing discussion.
References
Distribution and influence is the moat" — this is the part most AI-native founders underweight in year one. When the build is cheap and fast, the differentiator stops being what you ship and becomes who hears about it and trusts you to deliver. Looking forward to this fork in the series.