The video commerce layer the open web didn't know it needed - shoppable, livestreamed, and increasingly agentic.
Photographed for YesPress: a wordmark, parked on a navy field, doing exactly what wordmarks are supposed to do - getting out of the way.
It's a Tuesday morning at L'Oréal. A makeup artist walks onto a set, picks up a camera, and goes live - not on TikTok, not on Instagram, but on the brand's own dot-com. Comments scroll. Carts fill. Nobody is paying a social network for the audience. Firework built that.
That tableau - the brand running its own livestream, on its own URL, with its own data - is what Firework sells. It's a video commerce platform: shortform video that loops on product pages, livestreams that turn a product launch into appointment viewing, 1:1 video chat for the kind of purchase that still wants a human, and, increasingly, an AI shopping agent that picks up the conversation when nobody's clocked in.
For the better part of a decade, the deal was simple: a brand made a video, posted it to a social platform, and rented an audience back from the algorithm. The cost crept up. The data leaked out. And the eventual "buy" lived two clicks and a re-auth away on a site nobody had time to build.
Vincent Yang and Jerry Luk - both alumni of YouTube's shortform world - looked at this arrangement and noticed something inconvenient. The most valuable square inch of digital retail was the product page itself, and it was almost always still a static photo with bullet points underneath. The video lived everywhere except where the wallet actually opens.
That's the central tension Firework was built on, and it's the one that keeps showing up in every product they ship: brand-owned, on-domain, first-party. The opposite of borrowed reach.
In 2017, Yang and Luk filed paperwork under the name Loop Now Technologies - a tip of the hat to the format that they thought, correctly, would eat the internet: the short, looping, vertical video. The friendlier consumer brand, Firework, came later.
Former YouTube product lead; the operator of the duo. Runs the customer-facing motion and Asia expansion.
Engineer-founder; built the original video stack. Defends the unfashionable architectural choices that turned out to be right.
The early skeptics were not wrong, exactly - just early-stage. A consumer shortform-video app from a tiny startup was not going to out-product TikTok. The pivot Firework made, sometime between Series A and Series B, was the kind that quietly redefines a company: stop trying to be a destination, become the picks and shovels.
SoftBank Vision Fund 2 wrote the check that made it official: $150 million in May 2022, on top of earlier rounds from Lightspeed and GGV. By that point, the customer list - Walmart, Unilever, P&G, Samsung - was doing the work that no slide deck ever could.
Strip the marketing back and Firework is a video runtime for commerce. It drops into a brand's site - usually a few lines of JavaScript - and brings with it whatever combination of formats the brand wants to run.
Vertical, looping videos embedded on product pages and category pages. Items tagged in-frame. Add-to-cart without leaving the player.
Branded live events on the retailer's own URL. Chat, polls, scarcity, one-tap checkout - the toolkit QVC would have built if it had been born in 2024.
Live virtual shopping with a real human - store associate, stylist, technical expert. Aimed at high-consideration purchases where photos lose the sale.
Conversational, agentic and a little uncanny. Recommends, answers FAQs, hands off to a human when it should.
AI-driven curation that decides which video plays for which shopper - across web, in-store screens, and retail media networks.
The piece executives actually care about: watch time, conversion lift, AOV by video, cohort behavior. First-party. Owned.
You can argue with a thesis. You can't argue with a funding history.
Then there's the customer list, which is more persuasive than any chart. Walmart Connect picked Firework as the livestream commerce partner inside its retail media network - alongside TikTok, Snap and Roku. L'Oréal, Unilever, P&G, Johnson & Johnson, Samsung, Albertsons, Toyota and Gap all run Firework on their owned channels. Quietly. Without making a fuss about it.
Independent estimates peg current annual recurring revenue around $14M, although for an enterprise SaaS company priced on contract value and revenue share, ARR is a fuzzy measure. The more interesting metric, internally, is what percentage of a customer's video budget has migrated from social ad spend to owned-channel commerce. That number, by every Firework account, is going up.
The official mission statement reads: revolutionize commerce for people and brands through the power of human connection. Strip away the corporate gloss and the idea is honest. Online shopping spent two decades stripping the human out of the transaction - first with photo grids, then with reviews, then with algorithmic ranking. Firework's argument is that the last mile of that journey, the part that turns interest into purchase, still wants a face, a voice, or at minimum a video.
So you get livestreams that look like cooking shows. 1:1 chats that look like the conversation you'd have at a counter. Shortform videos that look like the friend who tried the thing first. And, when no human is available, an AI agent that does a passable impression of one - trained on the brand's own catalog instead of the entire internet, which turns out to matter more than people expected.
The interesting trend Firework is tracking - and quietly accelerating - is the slow rebalancing of attention from rented platforms back to owned ones. As cookie deprecation bites, as social CPMs creep, as antitrust cases nibble at the edges of the walled gardens, brands have a fresh incentive to invest in their own surface area. Video is the most expensive thing to build on that surface. Firework is the easiest way to do it.
Add the AI Shopping Agent and the math gets stranger. A conversational layer that knows the catalog, tone of voice and return policy of a single brand can do something the general-purpose chatbots can't: actually close a sale, on the brand's own page, in the brand's own voice. That's the wedge Firework is betting the next chapter on.
Back to the L'Oréal makeup artist on a Tuesday morning. Three years ago, that livestream would have run on a social platform that owned the audience, the data and most of the upside. Today it runs on L'Oréal's own dot-com. The audience shows up. The carts fill. The data stays. Somewhere behind a logo nobody is meant to notice, Firework keeps the bandwidth flowing and the buy buttons honest.
That, in the end, is the company. Less a brand, more a backstage. The kind of infrastructure that only looks invisible if you don't know where to squint.