AI surfaces
AI surfaces — conversational AI products, AI-driven search and answer engines, AI assistants embedded in apps — are where users increasingly ask to buy things. The destination is the AI surface itself: a chat thread, a generative search result, an in-app assistant. The agent responds to a purchase intent by calling commerce APIs in the background and presenting the result back inside the conversation.
The Firmly platform turns any AI surface into a transactional one without per-merchant integration on the AI surface’s side.
What this category covers
AI surfaces as a destination category includes:
| Surface type | What it does |
|---|---|
| Conversational AI products | Chat-based AI products where commerce shows up alongside other interaction types |
| AI answer engines and generative search | Search surfaces where results need to become transactable, not just informational |
| In-app AI assistants | AI integrated inside another product (workplace, productivity, lifestyle) that needs commerce as one capability |
| Autonomous agents | AI that completes purchases on the user’s behalf with explicit consent and constraints — see Agentic Pay |
Why AI surfaces are a Firmly destination
Conversational commerce introduces challenges an AI surface usually doesn’t want to solve in-house:
- Multi-merchant discovery — a single conversation might span products from many merchants; one Firmly integration reaches all of them, with no per-merchant build on the AI surface’s side.
- Conversation-aware error handling — when the cart is blocked (out of stock, address invalid, payment declined), the agent needs to recover gracefully inside the conversation, not break the user back to a form.
- Consent and disclosure — the user has to know what they’re authorizing, especially for autonomous purchases. See Consent & Disclosure.
- Protocol heterogeneity — different agent stacks speak different protocols. Firmly speaks UCP, MCP, ACP, and direct REST.
- Payment for cardholder-not-present — when an autonomous agent purchases on the user’s behalf, network-token-based mandate payment is essential. See Agentic Pay.
Which Firmly solutions apply
| Solution | When to use it |
|---|---|
| Agentic Commerce | The primary solution for AI-surface destinations. Designed end-to-end for AI-agent-driven commerce — protocols, errors, consent, payments. |
| Branded Commerce | When the AI surface routes the user to a partner-branded checkout (e.g., for a guided purchase flow). |
Recommended starting points
When you integrate (see Status today for current availability), these are the defaults to reach for:
| Choice | Default | Why |
|---|---|---|
| Auth | Browser session | Per-conversation authentication; user-scoped session token |
| Protocol | MCP for chat / agent-call surfaces; direct REST for backend agents | MCP exposes Firmly tools the LLM can call iteratively; REST is the canonical fallback |
| Integration pattern | Embedded checkout or deep link / headless | Embedded for visual rendering; deep link / headless for chat-only |
| Payment | Card via JWE; Agentic Pay for autonomous flows | Standard for user-present; mandate-based for user-absent |
Status today
As a destination category, AI surfaces are a roadmap buildout. The underlying agentic-commerce platform — protocols, payment, error handling, consent — is production-ready and documented at Agentic Commerce. Surface-specific buildout (per-AI-surface integration patterns, MCP server templates, agent-stack guidance) is scoped per engagement.
What to think about
- Discovery scope. AI surfaces typically curate the merchant set rather than expose every merchant on Firmly. Discuss the scope during onboarding.
- Error patterns. Conversational commerce has unique error states (stock changed mid-conversation, payment declined, address invalid). See Errors & Recovery for the patterns.
- Going live readiness. Before flipping production traffic, work through the Going Live checklist.
Related
- Agentic Commerce overview
- Agentic Pay — mandate-based payment for autonomous purchases
- Consent & Disclosure
- Protocols overview — UCP, MCP, ACP, REST