Agentic-Commerce Infrastructure
"Agents are becoming the storefront. Without a new foundation, brands vanish from the shelf—and agents take forever to be wrong."
Intent
→
Discovery
→
Evaluation
→
Checkout
→
Fulfillment
Most of the decision happens before checkout—when a buyer is comparing options and deciding what's acceptable on price, delivery, and returns.
Click = the handoff (discovery → evaluation on PDP)
Humans do the work: browse, compare, interpret
Win the click → you get a chance
Shoppers compare ~3 different stores/marketplaces before buying[1]
Today, the click is the handoff: discovery earns a click, then evaluation happens on the PDP.
Humans do the work—browse, compare, interpret—so if you win the click, you at least get a chance to compete.
Transition: Now the handoff is moving.
Handoff moves into the agent (evaluation before PDP)
Agent returns a shortlist inside chat
Distribution becomes 'be in the agent' (catalog integration)
Early market: requirements multiply (N agents = N requirements)
45% take 6–12 months to adapt product data to a new channel[5]
As shoppers move from links to answer engines, the agent starts doing evaluation inside the conversation—often before the user ever reaches a PDP.
In that world, distribution isn't 'win the click.' It's 'be in the agent'—and because this is early, requirements will multiply across agents.
Transition: That creates new marketplace dynamics.
Agents can explore, but time is limited → results must come fast
Exploration becomes retrieve → rerank → shortlist (not 30 PDPs)
Index + retrieval infra becomes the "shopping shelf"
Shortlists concentrate outcomes; ranking matters more
Top 3 results get ~64% of clicks (Amazon analog)[2]
If key attributes and policies aren't structured, you may not qualify
The hard part is keeping many integrations correct and current
Wrong data pushes platforms to demand higher correctness, increasing merchant burden
Even strong models look unreliable when the commerce layer is incomplete and inconsistent
Variants, duplicates, and inconsistent policies make decision-quality comparisons impossible
Scraping and calling dozens of APIs per query doesn't meet latency and reliability requirements
We're Building The Missing Layer
Turn messy catalogs into structured, comparable offers + machine‑readable policies — with provenance, timestamps, and freshness monitoring
A fast, reliable retrieval API that returns decision‑ready candidates (not links) with the signals agents need to choose confidently
An open publishing + discovery substrate where merchants publish once and agents integrate once — so the ecosystem can proliferate without bespoke integrations or closed‑catalog lock‑in
Supply (brands)
Subscription for agent-readiness + syndication/refresh + monitoring/diagnostics
Demand (agents)
Usage-based retrieval API + paid SLAs (latency, freshness, provenance)
Enterprise (B2B)
Private deployments + custom integrations (internal catalogs, procurement, regulated workflows)
Expand
More destinations + freshness tiers + deeper category/vertical packs
Existing budget: product data (PIM/PXM) + feed/syndication + integration ops
Context: AI in retail projected ~$96B by 2030[4]
Initial supply SAM (wedge): ~100k mid-market/enterprise merchants × $20–50k/yr = ~$2–5B
Expansion: as more agents launch, 'integrate once' becomes the default path
Market sizing is easiest to see as budget consolidation. Merchants already spend on product data systems, syndication, and ongoing integration ops—we replace and unify that spend for the agent era.
Near-term, the wedge is supply-side subscriptions for 'integrate once → syndicate everywhere.' Over time, demand-side usage expands as agent shopping query volume grows.
Start with: differentiated D2C brands (they win on decision-driving details: materials, fit, compatibility, delivery, returns)
Promise: integrate once (storefront + PSP agnostic) — connect to your existing stack
Distribution: publish to agent platforms (ChatGPT first; more agents coming)
Ops: keep facts current as requirements evolve (price/inventory/policies refresh)
Velou — $9M (2024) (catalog intelligence)
Daydream — $50M (2024) (AI-native shopping/discovery layer)
Nekuda — $5M (2025) (agent payment authorization/guardrails)
Rye — $14M seed (Oct 2022) (universal checkout API)
Skyfire — $9.5M (2024) (agent payment rails)
Co-Founder
We've shipped AI agents and enterprise-grade integrations, and built conversion-critical commerce experiences. Now we're applying that playbook to agentic e-commerce end-to-end.
[1] Baymard Institute — cart abandonment benchmark (~70%):
https://baymard.com/lists/cart-abandonment-rate[2] Algolia — e‑commerce search impact (2.4× buy likelihood; 2.6× spend):
https://www.algolia.com/blog/ecommerce/e-commerce-search-and-kpis-statistics[3] Salesforce — consumer AI shopping / agent adoption signals:
https://www.salesforce.com/in/news/stories/new-research-shows-how-ai-agents-can-step-in-as-consumer-trust-slips/[4] Mordor Intelligence — AI-in-retail market sizing ($14.2B → $96.1B, 2025→2030):
https://www.mordorintelligence.com/industry-reports/artificial-intelligence-in-retail-market[5] Lunar.dev (Report 2024) — integration troubleshooting time (~36% of eng time):
https://www.lunar.dev/report-2024[6] Aviso — agent latency expectations (<2s for "natural" workflows):
https://www.aviso.com/blog/how-to-evaluate-ai-agents-latency-cost-safety-roi