Skip to content

Selling to AI Agents: One Year Later

Adam Leon
Selling to AI Agents: One Year Later

A follow-up to the June 2025 prediction

In June 2025, we published “You’re Not Just Selling to People Anymore. You’re Selling to Their Robots.” The article argued that customers would increasingly use AI agents to research, compare, filter, and eventually transact with businesses.

One year later, the prediction has aged well, but the most important change is more practical than many expected.

The first version of agent commerce is the buyer asking an AI system what to consider, what to ignore, and which option fits best. The agent does not need to complete the payment to influence the sale. It only needs to shape the shortlist.

For businesses, the question has moved from “Can an AI agent buy from us?” to “Can an AI agent understand us well enough to recommend us?”

What changed first: the shortlist

AI’s strongest commercial role today is filtering.

Buyers are using LLMs and AI search tools to:

  • Compare vendors
  • Summarize reviews
  • Interpret pricing
  • Identify risks
  • Explain integrations
  • Build shortlists
  • Draft internal business cases
  • Prepare questions for sales

This is already material in B2B. Research we are tracking shows 94% of buyers use LLMs during the buying process, and 72% encounter Google AI Overviews. At the same time, 70% of the B2B buyer journey happens before sales contact, and 94% of buying groups rank their shortlist before contacting sales.

The decision point is moving upstream.

A company can lose before a sales call if it is absent, unclear, or misrepresented in AI answers. If pricing is hard to interpret, integrations are vague, proof is thin, or positioning changes from page to page, the AI layer has less confidence.

The AI-generated summary is becoming a new first impression.

The transaction layer is forming, but unevenly

Agent-led transactions are developing, but the market is still early.

The current landscape has several layers:

AI-assisted research

A person asks AI to compare, summarize, and recommend. This is already common and commercially important.

Agent-mediated transaction

AI helps complete a purchase, usually with human confirmation. This exists in selected consumer and platform-specific contexts.

Agent-to-system execution

An agent interacts with APIs, checkout links, product feeds, documentation, or structured workflows. This is where much of the practical infrastructure is forming.

Agent-to-agent coordination

A buyer-side agent communicates with a seller-side, supplier-side, or service-side agent. This is early, but protocols, commerce manifests, scoped payments, and agent identity systems are making it more realistic.

Autonomous agent commerce

Agents transact within approved budgets, rules, permissions, and audit trails. This remains limited for most complex purchases, especially in B2B.

The market is strongest today in research and infrastructure, with early transaction use cases in narrow, auditable settings.

Where agent-to-agent commerce is most real

Agent-to-agent commerce is appearing first where the environment is structured and the action is repeatable.

Developer tooling is one leading edge. Agents can already read documentation, follow setup instructions, install packages, configure services, use APIs, run commands, trigger workflows, and generate logs.

The same pattern applies to commerce. Agents need:

  • Clear instructions
  • Defined capabilities
  • Authentication
  • Permissions
  • Logs
  • Repeatable outcomes
  • Structured data
  • Payment or authorization rules
  • Human override where needed

The strongest current categories include:

  • Developer tools
  • API products
  • Low-risk SaaS add-ons
  • Usage credits
  • Software licenses
  • Service booking
  • Machine payments
  • Agent marketplaces
  • Referral and reward infrastructure
  • Bounded procurement workflows

Agent-to-agent commerce starts where the action is permissioned, low-risk, repeatable, and auditable.

Why B2B moves more slowly

B2B buying is more complex than consumer checkout.

A buyer is evaluating business impact, technical fit, risk, pricing, implementation, and internal consensus. A marketing manager choosing software often needs input from sales, RevOps, IT, security, finance, procurement, leadership, and end users.

The complexity shows up clearly:

  • Buying committees typically include 6 to 10 stakeholders
  • 38% of objections come from IT and security
  • 30% of purchases face finance and procurement objections
  • Security is often a pass/fail factor
  • Integration is the top decision factor for marketing software buyers

This slows autonomous purchasing, but it increases the value of AI-assisted evaluation.

Agents can gather evidence, compare options, summarize risks, draft internal cases, and prepare procurement questions. In B2B, agents shape evaluation and consensus before they control purchase execution.

That is still a major commercial shift.

Trust is becoming machine-readable evidence

AI systems need evidence to recommend confidently.

Generic vendor claims are weak source material. Strong recommendations depend on proof, structure, specificity, and consistency across sources.

Important trust signals include:

  • Public reviews
  • Peer recommendations
  • Case studies with specific metrics
  • Transparent pricing or pricing logic
  • Integration documentation
  • Security documentation
  • Support expectations
  • Implementation timelines
  • Customer proof
  • Clear limitations
  • Fresh product data
  • Consistent messaging across channels

The buyer is also looking for external validation. 97% of buyers say peer recommendations are the most trustworthy source, 92% are more likely to purchase after reading a trusted review, and 73% check third-party reviews during evaluation.

The AI layer looks for corroboration. If your website says one thing, your sales materials say another, and your public proof is thin, both the buyer and the machine have to work harder.

Trust is becoming machine-readable evidence.

Agent-readable is the practical requirement

Agent-readable means making the business understandable to both people and AI systems.

This goes beyond SEO. Traditional SEO focused on clicks. Agent-readiness focuses on whether your offer can be cited, compared, validated, and acted on.

Agent-readable assets include:

  • Structured service pages
  • Product pages with clear use cases
  • Pricing explainers
  • FAQs
  • Comparison pages
  • Integration guides
  • API documentation
  • Security pages
  • Implementation guides
  • ROI calculators
  • Review profiles
  • Case studies
  • llms.txt
  • Schema markup
  • Product feeds
  • OpenAPI specs
  • AGENTS.md where relevant
  • Agent Cards or commerce manifests where relevant

The website becomes a structured knowledge system, not only a marketing destination. It needs to help buyers understand the offer, help AI systems interpret the offer, and help internal teams stay consistent.

This is where brand and execution meet. A brand defines expectations. Customer experience confirms or breaks them. In an AI-mediated buying journey, that gap can appear before a person ever speaks to your team.

What marketing teams should do now

Marketing has to support AI-mediated evaluation.

For mid-size B2B SaaS teams, this connects directly to existing pressure. CAC is rising. Attribution is messy. Sales cycles are long. Tool stacks are overloaded. Buyers expect more useful information before they engage.

High-value content now includes:

  • Integration matrix
  • Pricing and total cost of ownership explainer
  • ROI calculator
  • Implementation timeline
  • Security and compliance documentation
  • Migration guide
  • Comparison content
  • Buying committee one-pager
  • Customer proof library
  • Product demo path
  • Technical FAQ

This work serves buyers, sales teams, procurement, and AI systems at the same time.

The strongest content is useful enough for humans and structured enough for machines.

What to avoid overstating

The market is moving quickly, but the practical work is grounded.

Autonomous B2B purchasing is still limited. A chatbot comparison is not the same as an agent-to-agent transaction. AI-assisted research, agent-mediated checkout, and autonomous commerce are different levels of maturity.

A clearer view looks like this:

  • Already active: AI-assisted research and shortlist influence
  • Emerging: agent-mediated checkout and workflow execution
  • Early infrastructure: agent identity, scoped payments, agent protocols, commerce manifests
  • Still limited: complex autonomous B2B negotiation and enterprise purchasing

Businesses should avoid waiting for full autonomy before acting. The current risk is discoverability, clarity, and trust in AI-mediated evaluation.

The practical conclusion

Businesses should prepare for agent-to-agent commerce by first becoming agent-readable, trust-rich, and easy to compare.

Make the business agent-readable

  • Clarify who the offer is for
  • State the problem it solves
  • Define use cases
  • Publish structured FAQs
  • Add schema markup
  • Create llms.txt
  • Keep pricing, availability, and product data current
  • Publish integration and API documentation
  • Make service scope, deliverables, timelines, and limitations clear

Make the business trust-rich

  • Publish specific case studies
  • Capture public reviews
  • Show peer proof
  • Add security and compliance documentation
  • Explain support expectations
  • Clarify onboarding and implementation
  • Keep claims consistent across website, sales materials, review sites, and social channels

Make the business easy to compare

  • Create comparison pages
  • Explain where the offer fits best
  • State tradeoffs clearly
  • Publish pricing logic or TCO guidance
  • Show integrations side by side
  • Create buyer-role content for finance, IT, security, end users, and executive sponsors

Prepare for transaction readiness

  • Identify low-risk actions an agent could trigger
  • Define what requires human approval
  • Create clean booking, checkout, or request flows
  • Add audit trails
  • Track referral and attribution metadata
  • Prepare for scoped payment permissions where relevant
  • Build structured handoffs into CRM or support systems

Where we are applying this now

At Transmitter Studios, this is already shaping how we build.

In our own projects, we are developing rewards systems for agent builders, which are tools that let AI agent products carry referral, attribution, and reward logic through agent-readable workflows. The goal is simple: when one agent helps another system discover, install, or use a tool, value can be tracked and rewarded through structured metadata and automated payout logic.

For clients, we are applying the same thinking through AEO visibility work, which means helping businesses become easier for AI search engines, LLMs, and answer systems to understand and cite. That work includes clearer positioning, structured content, schema markup, FAQs, trust assets, comparison content, and consistent public facts.

Both streams point to the same conclusion.

The agent does not need to own the entire transaction to change the market. It only needs to influence the shortlist.

The next phase of commerce will reward businesses that make their value legible. The work starts with the information, proof, and systems agents use to decide who deserves attention.

Ready to launch insights into action?

Let’s talk about building the right tools for your business.