Yes There Are Agents - and they are Coming! Is your Site Ready?

Brett Tabkeon

We have seen that search gave way to AI answers, and AI answer boxes are already giving way to actions. In the next click-less moment, AI agents will book your flights, reorder groceries based on an image of your fridge contents, and surface the exact paragraph you need without ever bothering to load a webpage.

That leaves all of us site owners with a new urgent question: will an agent bother to visit you at all? What does this mean for websites?

I’ve been tormented with those thoughts lately. On one hand it feels like "game over dude, go home" and on the other, it seems like fresh opportunity. So I started digging under the hood of my own sites and pages, asking, “Is my site ready for agents?” I realize the answer is for most assuredly “not yet”. So before we shut up shop and call the retirement broker, let’s once again get ahead of the curve and rebuild our little corners of the web for the agent-first era-before the rest of the internet is scrambling to do the same.

Wait, I know some think this is premature, that we are not near this eventuality yet. True, nothing has taken the net by storm yet, but it sure seems like a given.

Do you know what these browsers have in common?

Opera Aria, OpenAI, Perplexity Commet, Arc, Zen, Brave, Dia, Fellou, Edge CoPilot, DuckDuckGo, Chrome

They are baking in AI and Agents. None of them have struck "killer application" status yet, but they will. It will happen either through sheer attrition by providing 50 actions you might like to do, or by solving just one issue that reduces a moment of browser pain.

I am not alone in thinking that AI Agents are the next wave of optimization for sites to consider. Duane Forrester (the former Project Manager at Bing who launched Bing Webmaster Tools) and now running a new marketing startup UnboundAnswers:

There are many opinions as to what's going on or going to happen as generative AI systems grow and expand. But one thing that's hard to argue is the sheer volume of users suddenly using these systems to get answers to questions. 700 million active weekly users on ChatGPT.


AI Agent Capabilities

A good starting point is to look at what functions the current crop of browser agents are implementing :

  • Info Retrieval & Summarization
  • Transactional & Booking
  • Communication
  • Account & Profile Mgt
  • Monitoring & Alerts
  • Collaboration with Other Agents
  • Automation & Orchestration

As always, early movers get the default advantage in these things. This is not going to be a perfect one-size-fits-all with Agents magically figuring out how to navigate websites. At their heart, LLM's are pattern-matching algorithms that will look to successful task repetition as their guide. That means default APIs and standardized software will be understood by LLM's first.

Once an agent finds and perfects a source to 'book a table at a restaurant near me' and your site fits that bill, it has you in their system and the other dozen competitors sites still running on something custom and are invisible.

Training data windows are short lived items. Currently, most of the LLMs are trained on data six months to a year old. Even those with search engines bolted on, are surfacing old data.

Agents will rely on trusted, tried-n-true successful completions moving forward. Much of the web is not going to be there when these things tip. Those that are late to the party, may even be locked out of the party as it could take a year for them to get into the next LLM that agents will be based on.

Goals for a Website Operator Prepping Now:

  1. Be Agent Readable: Ensure your data, content and APIs can be understood without human parsing. APIs are the key here.
  2. Stay in the transaction loop: Again - APIs that allow an agent to complete your tasks. Book a reservation, order a service, and even talk to customer service.
  3. Leverage Agents as a distribution channel: Start to think of agents as a type of RSS feed or social media channel as a way to reach a potential audience.
  4. Control the narrative: it's up to you to rep your company. Make sure to only surface facts about your biz (hrs, specs, price$, reviews) so that they pull relevant data going forward.
  5. Future Proofing: Get your agent future in place now. Agents that adopt your APIs today, will be there tomorrow.
  6. Track and measure those interactions.: add tracking for api calls, structure data pulls, and known agent visits. Do not - do not block AI agents today (and encourage competitors to do so).

Will Scott of Search Influence says:

Our approach to Website Agent Optimization focuses on structured, machine-readable data and integration points that make content easy for agents to discover, understand, and act on. That includes:

  • Full schema coverage beyond basics such as product, offers, reviews, FAQs, how-to, and organization markup, all validated and monitored.
  • Mapping the agent journey we're not just looking at Schema and metadata for discovery, they can be a roadmap for agents.
  • Semantic content alignment to ensure entity relationships are clear, reinforced, and linked across the Knowledge Graph.

We're designing for direct-to-agent delivery in parallel with traditional SEO.

Structured Data is Key

Structured data is the machine-readable layer that tells agents what something is, not just how it looks or what you think of it. In the current search engine world, schema.org markup can help SE's present rich snippets. In the agentified future, it is the guide they’ll pull from to build responses.

What agents will use structure data for:

  • Direct fact extraction: Agents will need clean, unambiguous signals for names, prices, addresses, hours, specs, availability, etc., been-here-done-that.
  • Task execution: If a user says, “Book a table at 7 pm for four,” the agent needs the reservation schema to connect to a booking API without parsing through human-language content.
  • Context linking: Schema lets agents map your entity to known knowledge graphs, avoiding guesswork or misattribution.

Structured Data can also be sourced out of your ADA compliance - making accessibility and schema not merely as compliance but as foundational to AI interaction. Ben Fisher, Founder, Steady Demand writes:

Making a site as ADA compliant is important, if you are ADA then you are doing schema, alt texting links and images etc.. In an agentic world this data must be readable visually and in the code to assist the agents with finding what they are looking for.

Standard APIs for Structured Data

Agents will probably be trained:

  1. ... for API interactions. These will be OpenAPI specs, Json Schema, OAuth/OIDC and other SDKs. They will learn common vernacular, params, auth patterns and errors.
  2. ... on the most popular software packages out there. That would be WordPress, and then the most popular plugins available. WP Rest API is critical.
  3. ... to follow each step. So walk through your customers journey to create the same flow for the agent: search, find restaurant, ck availability, book.
  4. ... to read feeds: sitemaps, RSS, product Json feeds, webhooks, Etags

It begs the question, what will agents actually be doing? Below is a curated list from far-to-many-hours digging through LLM's, reading futurists, watching The Matrix for the 10,000th time, and reading blog posts about agents.

Dixon Jones CEO of Inlinks says they are all-in on structured data and schema:

Because of our technology (WAIKAY stands for What AI Knows About You) we already consider it critical for our site to be ready for AI Agents, but we are addressing issues by no longer assuming that the visitor to our site will be human. This means making sure that we have very clear feature pages with a clear, unambiguous sentence for each feature, ideally clustered into logical topics. The objective is to make it easy for AI Agents to compare our offerings with those of our competitors. The AI Agent will then be more likely to recommend our offering if we tick more boxes in feature comparison tables.

Potential Agent Action Website Prep Steps
Answer questions from your site Add accurate, up-to-date structured data (JSON-LD schema.org); use semantic HTML; keep metadata consistent. (biggest thing you can do)
Summarize pages  Break content into logical sections with <h2>/<h3> headings; use <article> and <section> tags; avoid embedding key text in images. Headers seem to be key here
Compare products or services Include standardized product/service schema; expose pricing, availability, and specs in machine-readable formats; avoid hiding key info behind scripts. Stock apis are key. All this is risky though if you are not a price and/or performance leader.
Book reservations Implement reservation schema; provide an API endpoint or booking URL; ensure booking forms are bot-friendly with clear field names. Use one of the stock services that book sites.
Purchase items Offer secure, documented purchase endpoints; mark product details with schema; support tokenized agent transactions (OAuth and API keys).
Subscribe or sign up Simplify sign-up forms; allow POST requests from known agents; provide subscription endpoints.
Schedule apts Use Event and Offer schema settings to expose available slots via an API or .ics feed; make sure to support time zone metadata!
Send a msg or inquiry Have a machine-usable contact form API; mark required fields in form HTML; rate-limit to prevent spam abuse. again, use the most popular product/software you can.
Request quotes or proposals Provide a /quote endpoint with structured request fields; document accepted parameters for agents.  Stock RFP's will rule the day. If you do RFP's - dig in here deep.
Log in & retrieve private data Support OAuth2 or secure API token flows; allow session-based agent access; encrypt responses (HTTPS). Tricky - not sure of answer here.
Update account details Provide documented PATCH/PUT endpoints; validate inputs server-side; log all agent changes.
Download documents or statements Store files at predictable, secure URLs; mark them with rel="download"; serve with proper MIME types. An easy win.
Watch for changes to content Publish RSS/JSON feeds; implement Last-Modified and ETag headers; expose change logs or diff feeds. Notifications will be key.
Monitor compliance signals Include accessibility metadata; validate schema regularly; expose compliance status in a machine-readable format.
Event Notifications Provide webhooks or event APIs; document trigger conditions for other agents.
Pass data to other agents Use standardized formats (JSON, CSV); mark data licensing clearly; offer chunked or paginated output for large sets.
Chain multi-step tasks Make key workflows modular; allow partial task execution via API; document step dependencies.
Trigger external workflows Support outbound calls to integrations (Zapier, Make, etc.); publish integration documentation.
Batch-process queries Accept batched parameters in API calls; respect rate limits; return structured batch responses.
Sync info between systems Support two-way sync via APIs; implement change detection; secure data transfer with encryption and authentication.

What you can ship now so agents “just work”

a) Publish an OpenAPI spec

  • Host it at /openapi.json or link it from a manifest.
  • Include examples and error bodies/msgs.

b) Add a simple agent manifest

c) Expose your actions in JSON‑LD

  • Use potentialAction with EntryPoint to point at your API routes.
  • Keep prices, hours, stock, and policies machine‑readable.

d) Offer a sandbox endpoint

  • A test tenant or demo mode with fake cards and seed data.
  • Agents learn faster when they can run end‑to‑end safely.

e) Tight auth and quotas - this is big

  • OAuth2 with scopes per action.
  • Reasonable rate limits, retry‑after headers, and 429s when needed.

f) Clear, actionable errors

  • Helpful status codes and messages: what went wrong, what to try next, which field is missing.

g) Logs Logs Logs

  • Detect agent user‑agents and client_ids.
  • Log tool calls, success, time to complete, and common failures.

h) A simple checklist for “trainable" sites

  • Publish an OpenAPI spec with examples.
  • Add /.well-known/agent.json that points to it.
  • Advertise actions in JSON‑LD.
  • Support OAuth2 with narrow scopes.
  • Provide a sandbox and test data.
  • Return clean HTTP errors with hints.
  • Log by client_id and ship monthly changelogs.
  • Document rate limits and terms for AI usage.

The takeaway is blunt: the next web milestone isn’t a page view, it’s an agent completing a task. If your site can’t be discovered, understood, and executed by an AI in milliseconds - via APIs, machine-readable schema, human-free, friction-free endpoints-it simply won’t exist in the user’s journey. Rebuild now: publish an OpenAPI spec at /.well-known/, mark every action with JSON-LD, sandbox-test every flow, and log every agent call. Early movers become the training data; latecomers vanish from the model.