This month I caught myself enforcing a purchasing rule at our own agency that nobody had ever written down: if an AI agent cannot drive a tool through MCP, an API, or a CLI, we do not buy the tool. It was not a strategy memo. It was a reflex, the same reflex “does it integrate with the CRM” became ten years ago. The same week, building a competitive scorecard for a client, we added a column we had never needed before: agent accessibility, sitting right next to pricing and feature depth.
When the same habit shows up twice in one week at one consultancy, it is not a quirk. It is showing up inside your deals too. The issue is that this new evaluator never joins a demo, never fills out a form, and never tells you it was in the room.
Why this matters now
The numbers back the anecdote. Two thirds of B2B buyers now rely on AI agents and chatbots as much as or more than Google when evaluating vendors, and Jason Lemkin’s framing at SaaStr is blunt: agents play favorites, and they build those preferences from authentic product signals rather than from your ad spend. Forrester’s 2026 buyer research finds that genAI searches are now the starting point for B2B buyers, which means the first pass over your product happens before anyone at your company knows an evaluation exists. And Harvard Business Review made the strategic point in July: competitive advantage is shifting from understanding customers directly to managing AI-shaped interactions.
For a Series A or Series B SaaS company, this lands in a very concrete place. When a prospect asks their assistant whether Claude can connect to your product, the answer gets checked against your documentation in seconds. If the answer is no, or if the answer cannot be found, you drop out of a consideration set you never knew you had entered.
Evaluation followed discovery out of your funnel
Most GTM teams have accepted that discovery moved. Prospects ask ChatGPT-class tools before they search, and getting named in those answers is its own discipline now, one we cover in our GEO practice. What is newer is that evaluation is moving the same way. The question buying teams ask is no longer just “what does this product do.” It is “can our agents operate this product for us.” MCP in particular has gone from curiosity to checklist item in under a year, because it is the standard that lets an assistant discover what your product can do and then actually do it.
A buying team with automation already in place evaluates your product the way an engineering team evaluates a library. Does it expose a clean surface that existing systems can call? Three surfaces matter, in rough order of effort:
- An API, so an agent can read and write your product’s data programmatically.
- A CLI, so an agent can drive your product from the command line the way a technical operator would.
- An MCP server, so assistants like Claude and ChatGPT can discover your product’s capabilities and call them directly, with no custom integration work.
If none of the three exists, the agent does not see a product. It sees a brochure.
Being the default is the new category leadership
There is a compounding prize here. In browser automation, one open-source tool has become the thing AI assistants reach for by default. Nobody asks for it by name; the assistant simply picks it. Whoever holds that position in a category stops paying for acquisition one deal at a time. On a client call this month, a product leader said the quiet part out loud: if the assistants default to you, distribution takes care of itself.
The inverse is the risk, and it is not hypothetical. One of the most widely used data-enrichment platforms shipped its CLI months after its users started asking agents to do the work, and in that gap a wave of agent-first alternatives found their opening. Speed on the agent surface is not a developer-experience nicety. A quarter of delay is a market opening for somebody else.
The pattern from the field
We recently built competitive positioning for a scale-stage B2B data platform, the kind of work where you inventory every rival’s features, pricing, and integrations. The column that generated the most discussion was none of those. It was agent accessibility: which competitor exposes a real API, who ships a CLI, who runs an MCP server, whose documentation is complete enough that an agent can answer a buyer’s question without a sales call.
Two findings came out of that exercise. First, the ranking reshuffled. The category leader on features scored worst on agent accessibility, which is exactly the profile that looks safe today and gets quietly filtered out of AI-mediated shortlists tomorrow. Second, the messaging wrote itself. Campaigns aimed at a rival’s users hit differently when you can truthfully say: your agents can drive this today.
The same filter is tightening on the other side of the market. Buying teams are adopting stack policies like ours, where agent accessibility is a gate rather than a bonus. Both ends of your funnel are being filtered by the same test, and most products have never run it on themselves.
The agent-readiness playbook
What I would recommend, in order:
- Audit yourself from the outside in. Open a fresh Claude or ChatGPT session, ask it to evaluate your product for a buyer, then ask it to actually use your product. Every place it stalls is a gap on your list, and the transcript is the most honest audit you will get this quarter.
- Ship the smallest real surface first. A documented read-only API beats an agent platform announced for next year. A minimal MCP server that only exposes reads is weeks of work, not quarters.
- Treat documentation as sales collateral for machines. Public pricing, a quickstart that works without a call, and a machine-readable summary of what your product does. Agents do not book demos to clarify ambiguity; they mark you down and move on.
- Instrument the agent funnel before it grows. API signups, MCP connections, and CLI installs should land in your CRM as events with their own source, so the first agent-driven pipeline is visible instead of buried in “direct.”
- Put agent accessibility into your own vendor scorecards. You will learn the evaluation faster by running it on others, and your team will stop buying tools your own automation cannot operate.
- Re-run the audit quarterly. The MCP surface that passed in January is the one an agent skips in October, because the tools around it kept moving.
This is a GTM project, not a platform rewrite. The teams that treat it as one will be the products agents recommend without being asked. That is the whole prize.
If you want an outside pass on how agent-ready your product and your funnel actually are, that is part of our AI GTM consulting work, and you can talk to us directly.
Sources
- Jason Lemkin, SaaStr. “The Future of B2B Marketing is AI Agent Recommendations. And AI Agents Play Favorites.” January 2026. saastr.com
- Forrester. “The State Of Business Buying, 2026.” January 2026. forrester.com
- Graham Kenny and Ganna Pogrebna, Harvard Business Review. “AI Is Changing How Customers Choose Your Business.” July 2026. hbr.org