The customer service inquiry arrives at 3 a.m. Pacific time. A skincare subscription needs to be paused. Inventory for a Korean beauty serum is running low. An Instagram influencer campaign underperformed and the budget needs reallocation. At most startups, these problems wait until morning. At Light Anchor, they're resolved before sunrise—by software.
Sangha Park and Chase Kim are running five direct-to-consumer brands from San Francisco with what they describe as a "staff of zero." Not zero employees in the metaphorical sense of a lean operation, but something closer to the literal: an organizational chart populated entirely by AI agents with job titles borrowed from corporate America. There's a CEO agent that sets weekly priorities and controls discretionary spending. A general manager handles profit-and-loss statements and supplier negotiations. A marketing director generates ad creative and courts influencers. An engineer builds the tools the others need.
It's the kind of setup that sounds either visionary or faintly ridiculous, depending on how much faith you place in large language models to make judgment calls about customer acquisition costs and brand positioning. Park and Kim, who spent years building conversational AI at Sendbird before striking out on their own, are betting on the former. Their pitch, laid out after joining Y Combinator's Spring 2026 cohort, is admirably blunt: consumer brands can now scale "at the speed of compute, not headcount."
Perhaps more striking is what they claim they're not doing. According to the founders, Light Anchor isn't selling software to e-commerce operators. It's not positioning itself as infrastructure. The founders are becoming the operators themselves, with algorithms occupying the C-suite and real products shipping to real customers who have no idea a human never approved the Facebook ad that found them.
An Org Chart of Prompts
The architecture is deceptively straightforward. Light Anchor's "CEO" agent receives objectives—grow Seoul Dispatch's subscriber base, launch a new supplement line, test stoic philosophy apparel—and distributes them downward with a $10,000 weekly discretionary budget per brand. That agent only escalates to Park and Kim when something falls outside preset parameters, which the founders describe as rare but won't quantify.
Below the CEO sits a GM responsible for merchandising, inventory planning, and sourcing decisions. It's this agent that determines when to reorder K-beauty serums from suppliers in Seoul or which supplement formulations to test next based on margin analysis. A separate marketing agent generates ad creative in multiple formats, writes copy, and manages outreach to micro-influencers. An engineering agent builds internal tooling—customer dashboards, analytics interfaces, workflow automations—so the other agents can execute faster.
According to the company's Y Combinator launch post from May, this wasn't theory first. Park and Kim validated the approach by taking on more than 50 Shopify-related gigs on Upwork, where they earned a 4.9 rating before transitioning to operating full storefronts. The timeline they claim: three weeks from concept to live brand.
Whether that pace is sustainable—or desirable—is an open question. But it's undeniably fast.
Five Storefronts, Varying Levels of Proof
The portfolio, as it exists in late spring, spans K-beauty subscriptions, supplements, apparel inspired by Stoic philosophy, custom meme-based T-shirts, and skincare. Seoul Dispatch, the most developed of the bunch, ships monthly beauty boxes starting at $59 to customers globally. Since launching in early May, it has accumulated 14 reviews on Trustpilot between May 1 and May 16 with a 4.4-star average—a thin but real customer base.
Slow Haste, the skincare line, lists six SKUs priced between $28 and $42, marketed with the tagline "Ingredients with receipts." The branding leans clinical: serums and oils positioned on transparency and sourcing rather than aspiration.
The other three brands—The Half Life (supplements), The Stoic Supply (apparel), and Meme Tees (custom shirts)—appear on Light Anchor's homepage but haven't surfaced widely in public search results, which isn't unusual for newly launched DTC storefronts still in their first weeks of operation.
All five share a common platform, with brand-specific "memory" and decision logs accumulating in what the founders describe as a central operational record. The theory, outlined in a March blog post, is that each store's performance data feeds into the setup and targeting of the next one. Speed compounds. Mistakes, presumably, do too.
Marketing as Public Experiment
In a move that feels part transparency stunt, part infrastructure play, Light Anchor has built AdBench—a public dashboard that pits large language models against each other in live Meta advertising campaigns. A snapshot from early May shows $54.01 spent across 1,543 impressions and 40 clicks, yielding a 2.59% click-through rate, part of an ongoing performance study. Four models—Claude Opus 4.7, Gemini 3.1 Pro, GLM 5.1, and GPT 5.5—ran creative variants for Half Life supplements, with per-model performance breakdowns visible to anyone on a waitlist.
It's an unusual level of disclosure for a startup trying to prove its brands can work. Unless, of course, the real product isn't the brands at all but the system that runs them. Park and Kim argue in their thesis post that the bottleneck in e-commerce has never been ideas or capital but what they call the "undocumented execution layer"—the grind of daily operations that doesn't make it into playbooks. Agents, they suggest, learn by doing the work, not by reading about it.
That's a testable claim. Whether it's true at $54 in ad spend or $54,000 remains to be seen.
A Suddenly Crowded Category
Light Anchor isn't alone in chasing agentic commerce. In February, Genstore announced AI-native retail with autonomous agent teams. Shoplazza followed in April with what it described as the world's first AI-native commerce operating system. Triple Whale released Moby 2 in May, an "AI operating system for ecommerce" that promises autonomous execution across media buying, creative production, and operations. Crisp, another entrant, claims thousands of brands using its retail-focused agents.
The infrastructure beneath all of this is standardizing quickly. Google unveiled a Universal Commerce Protocol for AI agent shopping in January. McKinsey published an automation curve for agentic commerce the same month. The Linux Foundation now hosts an Agentic AI Foundation, a signal that the space has moved from fringe to institutional legitimacy.
What separates Light Anchor—at least in positioning—is the decision to own the brands rather than sell tools to other operators. It's a bet that the best way to prove agentic commerce works is to run actual stores, accumulate messy operational reality, and build systems that survive contact with customers who don't care whether a human or an algorithm wrote the product description.
The Sendbird Alumni Go Solo
Park and Kim met at Sendbird, the messaging platform that emerged from Y Combinator's Winter 2016 batch and later focused on customer experience agents by 2024. They led AI agent product development and forward deployment there before leaving to start Light Anchor in early 2026. The company raised a pre-seed round from Korean investors Krew Capital and ASQ in February, then entered Y Combinator's Spring cohort in March. CB Insights and Dealroom data indicate total funding around $500,000, though some sources report conflicting figures, likely including the standard YC safe note.
Upwork CEO Hayden Brown publicly congratulated the founders on their launch and noted a partnership "to explore the frontiers of AI agents" on the platform. Light Anchor also appeared at Shoptalk 2026 and ran a session on OpenClaw, an open-source browser automation tool, though details on adoption remain scarce.
The Scale Question No One Can Answer Yet
The operational claims are bold but early. Seoul Dispatch, the flagship, launched in May and has 14 verified reviews from its first two weeks of operation. The other brands are weeks or days old. The agent system manages a $10,000 weekly budget per brand, which puts the entire operation's scale in the five-figure range—a proof of concept, not a unicorn.
It's also unclear how much human oversight remains tucked inside escalation loops, supplier negotiations, or creative approval processes that don't make it into the public narrative. In their public communications, the founders describe the setup as involving low human intervention, though not zero. And the model only functions if customer acquisition costs stay low enough that algorithmic targeting and compute-generated creative can sustain unit economics without the brand equity or repeat purchase behavior that typically take months or years to develop, not three weeks.
What Light Anchor has built is legitimately unusual. Most agentic commerce startups are selling picks and shovels to existing operators. Park and Kim are attempting to be the operator, compounding learning across a portfolio of disposable brands until the system accumulates enough operational knowledge to launch something that might actually matter.
Whether that strategy works hinges on questions no one can answer yet: how much of commerce is repeatable process versus irreducible human judgment, and whether AI agents can learn the difference before exhausting their budgets. The founders have given themselves a tight window to find out. Their customers, scrolling through skincare serums at midnight, probably don't realize they're part of the experiment.
