Ecommerce / Retail · Deep dive

Daydream

Julie Bornstein's second attempt at the same idea — an AI fashion shopping agent that reads plain-English or an uploaded photo and finds the outfit across 8,000-plus brands, three years after Pinterest bought her last one.

emerging

The question that decides it: Can a fashion-only affiliate agent build enough repeat shopper habit and click-through conversion to defend a double-digit referral take rate before Google Gemini, OpenAI's ChatGPT checkout, and Amazon Rufus fold vertical fashion discovery into agents that already own the traffic and the transaction?

My take ★★ Very interesting ★ Interesting − Maybe ✗ Not interesting

Snapshot

Daydream is an AI shopping agent built exclusively for fashion. A user describes what they want in plain language — “a revenge dress for a party in Sicily” — or uploads a screenshot, and Daydream returns products from a catalog of ~2 million items across 8,000-plus brands and 200-plus retail partners. It monetizes as an affiliate: purchases happen on the brand’s own site, and Daydream takes a referral fee. Founder Julie Bornstein is a 25-year e-commerce veteran whose last company, The Yes, was an AI fashion app Pinterest bought in 2022 — making Daydream a rare thing: a second attempt at the identical problem by the same person, funded with a $50M seed before a line of code shipped. It is very early — no disclosed revenue, no disclosed user count, and every large AI platform now racing at the same behavior.

Founding story

Bornstein says she first imagined a true personalized search-and-discovery layer for shopping while running e-commerce at Nordstrom in the early 2000s, then spent two decades getting closer — Urban Outfitters, CMO and Chief Digital Officer at Sephora, COO of Stitch Fix (scaled to ~$1B in under six years), and finally The Yes, the AI fashion startup she co-founded in 2018 and sold to Pinterest in 2022, where she led shopping strategy before leaving.

The trigger for Daydream was ChatGPT’s late-2022 arrival, which convinced her the missing piece — a machine that could parse messy, taste-loaded requests — had shipped. She recruited people who had lived the problem with her: Richard Kim, who ran Pinterest shopping strategy after The Yes deal; Lisa Yamner Green, a fashion-tech partnerships veteran from Google, The Yes, and Condé Nast; plus product lead Dan Cary and technical co-founder Matt Fisher. Forerunner and Index co-led a $50M seed announced June 2024 (closed December 2023), with GV, True Ventures, and model Karlie Kloss participating — a comment on how much investors will pay for a proven fashion-commerce operator at the peak of the agentic-AI theme.

The uncomfortable subtext: Pinterest already bought this thesis once, in the form of The Yes. Daydream is a wager that the version of the idea that failed to stand alone in 2018-2022 works now because the underlying models are dramatically better — not because the market has proven it wants an AI stylist.

How it works

A first-time user builds a “Style Passport”: name, sizing, price range, favorite brands. From then on Daydream learns from searches, clicks, and saves. The core surface is a chat box deliberately oversized to invite long, natural requests rather than keyword strings. Type a prompt and the screen fills with product tiles taking up roughly two-thirds of the page; the agent moves to a left rail and asks follow-up questions — a “Say More” button lets the shopper add constraints (“different fabric,” “same style, lower price,” “nothing sheer”). Users can also upload or screenshot an image — an Instagram outfit, a runway look — and Daydream reverse-searches for the closest buyable matches.

Under the hood is an ensemble, not one model: around a dozen small, specialized language models plus large frontier models (reportedly OpenAI and Google Gemini) and proprietary components, each tuned to a fashion attribute — silhouette, fabric, fit, occasion, color, shopper behavior. The differentiator is not the chat interface anyone can copy, but the data layer: partner brands hand over raw catalog feeds, and Daydream enriches them with fashion-specific attributes so the agent can answer taste-and-occasion questions generic product search cannot.

Crucially, Daydream holds no inventory and processes no payments. It is a discovery-and-referral layer: when a shopper is ready, they click through to the retailer’s own site to buy. Bornstein has said an agent that completes the checkout on the shopper’s behalf is a future ambition, not a current feature.

Product and business overview

There are three things Daydream has built. First, the consumer agent — the daydream.ing web app (public beta June 2025) plus a design-forward iPhone app (November 2025). Second, the catalog and enrichment engine — the ingestion and attribute-tagging pipeline that turns raw brand feeds into something an LLM can reason over; this is the actual moat-in-progress. Third, the partner network — around 200 retail partners representing 8,000-plus brands and ~2 million products at launch, from mass (Uniqlo, Nike, Madewell) to luxury (Chloé, Khaite, Emilia Wickstead) and multi-brand retailers (Nordstrom, Net-a-Porter, Mytheresa). The cheapest onboarding budget band runs up to $150, a quiet signal the target customer is affluent.

The strategic read: Daydream positions as a top-of-funnel channel brands cannot buy their way to the top of — visibility depends on catalog data quality, not paid bids — which is how Bornstein sells wary partners. Her framing is a new acquisition channel that cuts their dependence on Google and Meta, not a marketplace disintermediating them.

Business model and pricing

Daydream is free to shoppers and free for merchants to join. Revenue is affiliate: a referral fee booked when a Daydream user completes a purchase on a partner’s site. Bornstein has not disclosed the exact rate but has said it sits above standard fashion affiliate-network fees (typically 5-10%) and below what luxury marketplaces charge (which can exceed 30%). Call it low-to-mid double digits, unconfirmed.

That model is elegant and fragile. Elegant: no inventory risk, no payments liability, near-zero marginal cost per recommendation, and a non-threatening pitch to brands. Fragile: affiliate revenue only exists if the click-through converts on the partner site — a step Daydream does not control or see cleanly — and the take rate is exactly the line item every large platform can undercut or absorb. There is no subscription, no ad load (“no ads, no pay-to-play” is marketed as a feature), and no first-party checkout yet. Monetization is nascent: one revenue line, no published GMV, resting on a behavior — habitual AI-agent fashion shopping — that does not yet exist at scale.

Traction over time

The honest version: a company with a launch, a catalog, and prizes, but no disclosed commercial metrics. What is on the record:

No user counts, GMV, revenue, retention, or app-download figures have been published as of mid-2026, and no follow-on round has been confirmed (a separate, unrelated company also called “daydream,” an AI SEO agency, raised a $15M Series A in April 2026 — not this company). That absence is itself the traction story: eighteen months post-funding and a year post-launch, the metrics that would prove the behavior is real are not yet being shown.

Market analysis

The tailwind is enormous on paper. Mordor Intelligence pegged the agentic-AI-in-retail market at roughly $60B in 2026, growing ~29% annually. The forward numbers are staggering and wildly dispersed: Bain projects US agentic commerce at $300-500B by 2030 (15-25% of e-commerce); Morgan Stanley models $190-385B of US agentic-shopper spend by 2030; McKinsey floats up to $1T orchestrated US retail revenue and $3-5T globally by 2030. AI-in-fashion specifically is forecast to reach ~$9.45B by 2030 (Research and Markets, ~40% CAGR).

The counter-signal is in the same reports. A July 2025 Forrester survey of 700 consumers found only ~a third willing to pay through an answer engine at all, citing privacy. Analysts note agentic adoption moves fastest in commodity and optimization purchases (groceries, supplies), while identity-laden categories — fashion foremost — are where consumers most resist handing off the decision. That is the paradox in Daydream’s TAM: vast, but fashion may be the slice most resistant to the very autonomy the market is sold on.

Competitive intel

The competitive set is not scrappy startups; it is every platform with a chat box. Google can turn fashion discovery into a native Gemini answer and has already wired agentic checkout in — it owns the traffic Daydream must otherwise buy. OpenAI embedded Instant Checkout into ChatGPT (800M+ weekly users) via its Agentic Commerce Protocol with Stripe in late 2025; Bornstein’s defense is that generalists handle spec-driven buys but not taste, which OpenAI is actively trying to disprove. Amazon Rufus added auto-buy but, per Jassy’s own 2026 admission, returns generic, sometimes mispriced picks — leaving the fashion-taste gap open, for now. Perplexity added PayPal-powered Instant Buy across 5,000+ merchants. Pinterest, at ~570M monthly users, is the sharpest irony: it bought The Yes and inherited Bornstein’s prior team and technology, so it competes with her using her own last company. And LTK / ShopMy already own creator-led fashion affiliate discovery — the same rails and the same trust layer, sourced from humans rather than an agent.

Daydream’s honest edge is focus and data — a fashion-only catalog enriched with attributes no generalist bothers to model, run by operators who have built fashion e-commerce four times over. The honest risk: “focus” is a feature the platforms can add; distribution is not.

History and evolution

• 2018 — Bornstein co-founds The Yes, an AI fashion shopping app.
• Jun 2022 — Pinterest acquires The Yes; app and site are sunset, team folded into Pinterest shopping.
• 2023 — Bornstein leaves Pinterest; founds Daydream with Kim, Cary, Fisher, and Green. Seed closes in December.
• Jun 2024 — $50M seed announced, co-led by Forerunner and Index, with GV, True Ventures, and Karlie Kloss.
• Jun 25, 2025 — Public beta launches on the web with ~200 partners and ~2M products.
• Nov 18, 2025 — iPhone app ships; catalog framed as 10,000+ stores.
• 2025 — Named to TIME’s Best Inventions of 2025.
• 2026 — Headcount around 60; no follow-on raise or commercial metrics disclosed. Big-platform agentic-checkout launches (OpenAI, Amazon, Google, Perplexity) intensify.

What people say

The case for. Early users and reviewers describe genuine “the search is over” moments. App Store reviews and hands-on write-ups (Fashionista, CNN, a widely shared review by stylist-blogger Lizzy Helena Brown) praise two things: natural-language search nailing oddly specific briefs — modest-dress constraints, occasion-plus-budget combinations — that filters cannot express, and screenshot-to-shop turning social inspiration into buyable products. A TIME Best Inventions of 2025 nod added credibility. The investor case is a bet on the operator: nobody has built consumer fashion e-commerce more times than Bornstein, with a team that ran this exact playbook at The Yes and Pinterest. Brands reportedly like that traffic lands on their own sites with no ad money reordering results — a cleaner deal than Google or a marketplace.

The complaints. The skepticism comes from serious places. Puck questioned whether anyone actually wants AI to shop for them, capturing doubt that a “digital stylist” is a product people return to. Amazon’s Andy Jassy told analysts in 2026 that most AI shopping agents — his own Rufus included — deliver generic, mispriced, under-personalized results, a blunt admission the category has not cracked usefulness. Forrester (July 2025) found only ~a third of consumers willing to transact through an answer engine, and analysts flag fashion as the category most resistant to autonomous buying because identity is on the line. Reviewers note practical limits: stack too many negative constraints (“no sheer,” “nothing tight”) against real inventory and results degrade. The loudest critique is strategic — Daydream is a thin discovery-and-affiliate layer chasing a behavior Google, OpenAI, and Amazon can subsume the moment they decide fashion is worth a vertical, with no first-party checkout, no proprietary demand, and a take rate any of them can undercut. There is no employee-review record yet, and — tellingly — no disclosed usage or revenue to rebut any of it.

Outlook: the open question

For Daydream to matter, one thing must become true that is not yet proven: that fashion shoppers will form a repeat habit around a specialized agent and click through often enough to convert — before the platforms that own the traffic make vertical fashion discovery a free feature of an agent people already use. If the habit forms, the enriched fashion catalog and operator pedigree compound into a defensible taste-and-data layer, and the affiliate model — zero inventory, near-zero marginal cost — becomes a high-margin traffic business brands gladly fund. If it does not, Daydream is a beautifully executed feature waiting to be absorbed.

The bear case is concrete: monetization is a single undisclosed affiliate line dependent on off-platform conversion Daydream cannot see or control; the behavior it needs is the one surveys say consumers most resist in this category; the competitors are every trillion-dollar platform with a chat box, one of which (Pinterest) runs on her last company; and eighteen months after a $50M seed, the settling metrics — users, retention, GMV, realized take rate — are not being shown, usually a sign they are not yet good. The bull case is equally concrete: the most credible operator in fashion commerce, real and free brand partners, and the vertical-specialist slot in a market McKinsey sizes in the trillions. The falsifiable test is retention and click-through conversion at scale over the next 12-18 months. Until Daydream shows it — or a platform folds fashion in first — the second act is unproven.

How a challenger would attack it

Daydream is itself the challenger, which makes its soft spots unusually visible: no owned demand, no owned checkout, no shown metrics. A new attacker would not rebuild the chat box — that is commodity — but would fix the two things Daydream deliberately left out. First, close the loop: Daydream books an affiliate fee only when a click converts on a partner site it cannot see or control, and Bornstein has deferred agent-completed checkout to “future ambition.” A challenger that launches with first-party checkout on ACP/Stripe rails captures conversion data, remembers sizes and returns, and offers brands attributable, guaranteed sales instead of referred traffic — worth more than a low-double-digit affiliate cut. Second, buy the habit instead of waiting for it: LTK and ShopMy prove fashion discovery already routes through human creators with trust attached, so a challenger seeds the agent inside creator storefronts rather than a standalone app that must manufacture repeat visits from zero. Third, exploit the metrics silence — eighteen months post-seed with no disclosed users, GMV or retention, Daydream’s brand pitch (“a channel you can’t buy placement in”) is unverifiable; a rival that publishes conversion benchmarks to partners wins the 200-retailer network on evidence, and those partnerships are non-exclusive feeds, not contracts.

Same playbook, new buyer

The enrichment engine is the asset; fashion-for-the-affluent is just its first customer. Daydream’s real build is the pipeline that turns raw catalog feeds into taste-attribute data an LLM can reason over — and its $150-max onboarding budget band shows it aimed that engine at affluent US women first. Two shifts look stronger. One: home and interiors — the same “taste, occasion, and budget” query structure (“mid-century console for a small rental entryway”), an even worse filter experience on incumbent sites, higher order values, and no Pinterest-owned agent yet dominating buyable results. Two: the same fashion agent sold B2B — licensing the attribute-enrichment and conversational-search layer to multi-brand retailers (the Nordstroms and Mytheresas already in its network) as white-label site search, converting partners into paying customers with revenue that doesn’t depend on habit formation. Daydream won’t follow either path easily: its founders’ entire moat is fashion pedigree, its brand and Style Passport UX are consumer-facing by design, and pivoting to B2B infrastructure would concede the thesis its $50M seed was priced on — that a standalone consumer fashion agent becomes a destination.

Sources and further reading

• Daydream rakes in $50M seed funding to build an AI-powered search engine (TechCrunch, June 2024)
• Retail veteran Julie Bornstein’s Daydream AI fashion shopping agent launches (Fortune, June 2025)
• Julie Bornstein’s Daydream Is Leading The Agentic AI Fashion Shopping Revolution (Forbes, June 2025)
• ‘Revenge dress for a party in Sicily’: using AI to make online shopping hyper-personal (CNN Business, July 2025)
• Daydream Launches Design-Forward iPhone App to Advance AI-Driven Fashion Search (PR Newswire, November 2025)
• Why the AI shopping agent wars will heat up in 2026 (Modern Retail, 2026)
• Does Anyone Actually Want AI to Help Them Shop? (Puck, 2026)
• Pinterest acquires AI-powered shopping startup The Yes (TechCrunch, June 2022)
• Agentic Commerce Impact Could Reach $385 Billion by 2030 (Morgan Stanley, 2026)
• Agentic AI in Retail and E-Commerce Market Size (Mordor Intelligence, 2026)