Launch BF: Light Anchor, Autonomous E-commerce

Chase Kim &

) builds fully autonomous e-commerce companies. We own and operate several brands today, each run by agents across marketing, sales, sourcing, operations, and support. We believe work that once required a team of humans can now be fully automated with higher quality at scale. Our vision is to run dozens, if not hundreds, of great e-commerces.

https://youtu.be/_uN-A-4uugA

Why we're doing this

The hypothesis behind Light Anchor is that e-commerce brands have always been bottlenecked by how fast humans can run them. With agents that can operate end-to-end, that bottleneck disappears, and a different kind of consumer business becomes possible: one where the cost and speed of operating it isn't capped by headcount.

The simplest way to test this was through Upwork where small businesses post work and freelancers fulfill it. The problems already existed and the success metric was simple: would the client accept the agent's work? Would they pay for the job? Could agents deliver human-quality results? Within a few weeks of building agents for sourcing, executing, reviewing, and delivering e-commerce jobs, the agent (we called him Flash) had completed 50+ jobs end-to-end, with an average of 4.9/5.0 rating and won several repeat customers.

Then something magical happened. The brands that hired our agent started asking us to go beyond the original scope. What started as "clean up our product catalog" soon turned into "run our marketing," then "handle our sourcing," then "run the whole storefront." A few weeks in, we were operating multiple stores on behalf of store owners who used to do this themselves.

As we expanded scope into different functions of a given brand, starting our own brand felt less a leap, but more an obvious extension of what we were already doing. That's how Light Anchor came to be.

Our stores

We're running 1 flagship brand and several experiment stores, deliberately spread across different categories to test various consumer hypotheses.

• Slow Haste (slowhaste.com): Our flagship brand, K-beauty subscription

• Experiment Stores
  • The Half Life (thehalflife.co): lifestyle supplements
  • The Stoic Supply (thestoicsupply.com): merch for Stoic philosophy fans
  • Meme Tees (shopmemetees.com): custom meme t-shirts

The platform

The platform is the shared layer every brand runs on top of. It handles the work that would otherwise need to be rebuilt for each new brand: tooling for the agents (uploading meta ads, generating ad creative, managing inventory, processing support tickets), analytics and logging for every action they take, and the cross-brand learnings that get fed back.

Every investment we make in the platform compounds across every brand we run, and across every brand we'll ever run. That's the whole point. The marginal cost of launching the next brand keeps falling, and the speed at which we can launch keeps increasing. Our bet is not that we can just operate more stores, but that we can understand customers better than anyone and capture their changing demand instantaneously through the compounding platform.

[Image: What’s recorded in the platform. Decision trace of a GM]

The agents

• CEO agent (per brand): Decides weekly priorities for the brand, manages budgets (each brand currently gets $10K of investment), and escalates decisions for our review.
• GM agent (per brand): Owns the P&L, the merchandising calendar, inventory, sourcing, and tier-1 customer support.
• Engineering agent (shared / platform / horizontal): Unblocks technical tasks, provides new tools, and ensures stability.
• Marketing agent: Leads creatives and influencer marketing. Our new marketer & influencer will be going live soon. Her website is cherylkim.me. Give her a shout!

[Image: Snippet of Slow Haste GM’s session]

The data

Every brand we run produces a stream of operating decisions: creatives tested, ads dollar spent, suppliers negotiated, and tickets closed. All of it gets written to a central repository that agents can read from.

This is one of the parts of the company we find most interesting, while nothing about it is theoretically new. Any e-commerce operator would love to have a structured history of every decision made across every brand they've ever run, and feed it back into how they run the next one. The reason it doesn't exist is that humans don't operate that way.

However, when the operators are agents, that cost goes to nearly zero. The record gets written by default, and once it starts to accumulate, a few things start to be possible that weren’t before:

• Better targeting and conversion. A creative that works on one brand teaches how other brands can target similar demographic. We can also iterate dozens, if not hundreds, of variants per brand, so the cross-brand signal accumulates even faster.
• Hardened tooling. Every tool the agents rely on is being run in production across all our brands. Bugs get surfaced faster, improvements are automatically suggested, and the next brand we launch inherits a system that's already been stress-tested in live operations.
• Market simulation. Eventually, with enough consumer-preference across enough brands, we should be able to simulate consumer reactions inside our own platform and rule out bad brands or products before spending real money on them.

📣 Teaser: We're releasing a benchmark for our Marketing agents on how welll they run digital advertising in the wild in the coming weeks.

Asks

If you're any of the following, or you know someone who is, we'd love to hear from you:

• Buy from Slow Haste. Use the YC-fam code YC26 for 25% discount at slowhaste.com. Buy a box, gift one, post about it. Every purchase and every reaction (good or bad) feeds the agent loop directly. We also have machine-friendly APIs, which is very cool :)
• E-commerce Operators. Share your stories and experiences. Use us as your sounding board and test bed for your agent workstream.
• Builders with tech we should be using. Send it. We test new tech faster than just about anyone in our batch and run it in production across our brands, so you'll get unusually fast signal back. Anything that touches agents, e-commerce ops, ad creative, sourcing, or support is worth showing us.
• Frontier labs working on real world consumer-behavior models. We sit on a lot of real-world signal across four operating brands: targeting, ad creatives (text and image), CTR, conversion, sales, retention. We'd love to talk to anyone trying to train models that can run end-to-end operations rather than just write code.

Reach us at founders@lightanchor.ai.

Team

We met at Sendbird (YC W16), where we led the AI agent product and forward deployment teams during its 2024 pivot into AI agents for customer experience. Sendbird built the messaging APIs behind DoorDash, Reddit, and Hinge.

We took this right after we decided to start a company back in December. Day 1!