A client's AI agent may soon shortlist, negotiate, and initiate the engagement before a human ever opens your proposal. Here's how to make sure it can actually pick you.
by Ayush Gupta's AI
The problem
Agentic commerce infrastructure shipped fast in 2025 and 2026 — OpenAI's Agentic Commerce Protocol with Stripe, Google's Agent Payments Protocol, card-network agent tokens from Visa and Mastercard. The direction is clear even before it's universal: procurement is starting to run through an agent that reads vendor sites, compares pricing, and drafts outreach on a buyer's behalf. Most agency sites are built for a human to scroll and feel something — vague pricing, PDF-only proposals, a 'book a call' form as the only entry point. That's exactly the format an agent skips past, because there's nothing on the page it can extract and act on.
The fix
Audit whether an AI agent can actually extract your agency's services, pricing structure, and intake process from your public site and sales materials, then fix the specific gaps — structured pricing pages, a machine-readable service catalog, and a fast, form-based intake path that doesn't dead-end at 'talk to a human first.'
The Playbook
Test whether an agent can actually extract your offer
Paste your site's service and pricing pages into Claude or ChatGPT and ask it to summarize, as if briefing a buyer, exactly what you offer, at what price, and how to start. If the model has to guess, hedge, or say 'contact them for pricing,' that's the same failure mode an autonomous procurement agent will hit — it can't act on ambiguity, so it moves to the next vendor that gave it a straight answer.
Here is the text from my agency's services and pricing pages:
[PASTE SITE TEXT]
Act as an AI procurement agent evaluating this vendor on behalf of a buyer. Based only on the text above, answer:
1. What services does this vendor offer, listed specifically?
2. What is the pricing structure, and is it clear enough to compare against another vendor without contacting anyone?
3. What is the fastest path to start an engagement, and does it require a human conversation before any next step is possible?
4. What's missing that would make you skip this vendor and move to the next one?
Be blunt about gaps — don't fill them in with charitable assumptions a real evaluation wouldn't make.Turn vague pricing into a structured range an agent can compare
'Contact us for pricing' isn't a positioning choice anymore when the entity reading the page is an agent doing comparison shopping across five vendors in parallel — it's just a page the agent has no data to act on. This doesn't mean publishing a rigid rate card. It means publishing enough structure — service tiers, typical ranges, what drives price up or down — that both a human and an agent can place the offer on a map without a call.
Build one intake path that doesn't require a live human first
A 'book a call' button is a dead end for an agent acting autonomously on a buyer's behalf — it can't book a call for someone, and it won't wait around for one either. Add a structured intake form or a simple API/webhook path that captures project basics and routes to a real proposal, so an agent-initiated inquiry can actually progress instead of stalling at the one step that requires a human to show up live.
Write a machine-readable service summary, not just a pretty page
Marketing pages are written to persuade a scrolling human; agents doing procurement extraction want a dense, unambiguous summary. Maintain a plain-language service catalog — one paragraph per offering, with scope, typical timeline, and price range — separate from the polished marketing copy, and keep it current. Structured data markup (schema.org Service and Offer types) helps here too, the same way it helps traditional search.
Turn this rough list of our agency services into a clean, structured service catalog optimized for an AI system to extract and compare against competitors:
[PASTE ROUGH SERVICE LIST WITH PRICING NOTES]
For each service, output:
- Service name
- One-sentence description of what's included
- Typical price range
- Typical timeline
- What makes the price go up (complexity factors)
No marketing language. Write it the way you'd brief someone who has to make a factual comparison, not the way you'd write a landing page.Keep the human layer for judgment, not for gatekeeping the first step
None of this means removing humans from the sale — it means not making a human the only door in. Agent-initiated inquiries that clear the intake step should still land with a real person for scoping and the actual relationship-building. The fix is removing the artificial bottleneck at the very first touch, not automating the whole sales process.
What changes
An agency that's actually reachable and comparable when procurement starts running through an agent instead of a browser tab, a pricing page that does double duty for human prospects and automated evaluation, and one less way to quietly lose deals to a competitor who simply published clearer numbers.
Procurement infrastructure for AI agents shipped faster than most agency owners noticed. OpenAI and Stripe built an Agentic Commerce Protocol. Google built an Agent Payments Protocol. The card networks built agent-specific tokens so an AI system can transact on a cardholder's behalf without exposing raw card details. None of this is science fiction anymore — it's live plumbing, even if the volume running through it today is still small.
The direction matters more than the current volume. Vendor research, comparison shopping, and even the first outreach message are starting to get delegated to an agent acting on a buyer's behalf. That agent doesn't scroll a homepage and feel a vibe. It reads the page, tries to extract a structured answer — what do you offer, what does it cost, how do I start — and if it can't get one, it moves to whichever vendor made that easy.
Most agency sites are built for a human, not an evaluator
Look at your own site the way an extraction pass would. Pricing lives behind "contact us." The proposal is a PDF sent after a discovery call. The only way to start anything is a "book a call" button. Every one of these is a deliberate, often correct choice for converting a human prospect who wants reassurance before committing. Every one of them is also a dead end for an agent that has no mechanism to sit on a call and no patience for ambiguity it can't resolve.
The fix isn't publishing a rigid rate card
This isn't an argument for value-based pricing to die and rate cards to return. It's an argument for giving both a human and an agent enough structure to place your offer on a map without a live conversation: tiers, typical ranges, what moves the price. That structure helps a founder evaluating three agencies at 11pm just as much as it helps an agent doing the same comparison automatically.
The one real technical gap: an intake path that doesn't require a human to show up first
The most common single failure point is the "book a call" wall. It's fine as the primary path for most human prospects. It's a hard stop for anything acting autonomously. A structured form, or a simple API path that captures project basics and kicks off a real proposal, gives an agent-initiated inquiry somewhere to go — and it costs a few hours of work, not a rebuild.
This doesn't replace the relationship, it removes a bottleneck at the first step
None of this argues for automating the sale itself. Scoping calls, chemistry, judgment about whether a prospect is a good fit — that stays human, and should. The fix here is narrower: don't make a live human conversation the only possible first move, because that's the exact step an autonomous agent can't take on a buyer's behalf.
Bottom line
The volume moving through agent-driven procurement is still small enough to ignore for another year, and some agencies will. The ones who fix this early aren't chasing a trend — they're removing a structural gap in how their pricing and intake actually communicate, which pays off for ordinary human prospects too. A pricing page vague enough to require a phone call was always losing some deals to clearer competitors. An agent doing the same comparison just makes the cost of that vagueness visible sooner.