Agentic Commerce

The buyer isn't always a human anymore. That's the uncomfortable truth B2B sales teams need to sit with.

When a VP of Engineering asks their AI assistant to "find three observability platforms that support OpenTelemetry, integrate with PagerDuty, and cost under $5K/month for 200 engineers," no human is going to Google that query. An agent does the research. It visits vendor sites, checks docs, pulls pricing, and comes back with a shortlist. If your product can't be evaluated by an agent — because your pricing is hidden behind a "talk to sales" button, or your specs live in a PDF, or your site blocks bot traffic — you don't make the list.

Agentic Commerce isn't a prediction. It's already happening at the research and shortlisting stages. Full autonomous purchasing is next. The companies that prepare their infrastructure now — machine-readable pricing, MCP endpoints, structured product data — will own this channel. Everyone else will wonder why their pipeline dried up.

Agentic Commerce operates across a spectrum of autonomy:

1. Research (happening now). Agents query vendor websites, documentation, and review sites to build comparison matrices. They pull structured data from MCP endpoints, NLWeb queries, and schema.org markup. The agent does in 30 seconds what a human SDR spends 3 hours on.

2. Evaluation (happening now). Agents score options against predefined criteria — must-have features, budget constraints, integration requirements, compliance needs. They rank vendors and generate shortlists with rationale.

3. Negotiation (emerging). Agent-to-agent negotiation is the frontier. A buyer's agent communicates with a vendor's agent to explore custom pricing, discuss enterprise terms, or request trials. Both agents operate within guardrails set by their humans.

4. Transaction (emerging). For straightforward purchases (self-serve plans, commodity services), agents complete the transaction end-to-end. For complex B2B deals, they prepare everything and surface the final approval to the human.

A Head of IT at a 500-person company tells their AI assistant: "We need to replace our current email security provider. Requirements: sandboxing, URL rewriting, DMARC reporting, under $8 per user per month, and SOC 2 Type II certified."

The agent queries six vendors' MCP endpoints. Four return structured pricing and feature data. Two have no machine-readable data — the agent has to scrape their marketing sites and guess at capabilities. The agent builds a comparison table, flags that two vendors don't explicitly confirm SOC 2 certification in their structured data, and recommends a shortlist of three with confidence scores.

The Head of IT reviews the shortlist over coffee. The agent already scheduled demos with the top two vendors by calling their booking tools. Total time from request to scheduled demos: 4 minutes. The two vendors without MCP support? They never made the shortlist.

• Hiding pricing behind "Contact Sales." Agents can't fill out contact forms and wait for a callback. If your pricing isn't available through a structured endpoint, agents drop you from the evaluation. Transparent, queryable pricing wins in agentic commerce.
• Assuming agents will read your sales deck. PDFs, slide decks, and webinar recordings are invisible to agents. Your product's capabilities, pricing, and differentiators need to live in structured, queryable formats.
• Not publishing compliance certifications in structured data. SOC 2, ISO 27001, GDPR compliance — agents check these as binary filters. If it's not in your schema.org data or MCP tool response, it's a "no" to the agent.
• Blocking AI agent traffic. Aggressive bot protection that blocks OpenAI, Anthropic, and Google AI user agents means you're blocking your future customers' research assistants.
• Thinking human relationships will protect you. Relationships still matter for closing, but agents control the shortlist. If you're not on the shortlist, the relationship never starts.