AI & agent readiness · running in production
A growing share of buying decisions now begins with a question put to an AI system rather than a search box, and the software asking those questions is moving closer to acting on the answers. AI & Agent Readiness is the assignment for making a business legible to that layer: one verified account of what the company does, published where machines look for it, in the formats they read—plus a bounded, owner-controlled way for an agent to ask a question and get a real answer, with no ability to commit the business to anything.
Operate it yourself · no form, no signup
The surface described on this page is live and open. Type a question and read the answer with the source of every claim — then ask it for a price or a delivery date and watch it refuse.
Before a business asks AI systems to recommend it, those systems need a current, verifiable, machine-readable account of what it actually does.
When this is the right assignment
The facts exist, but they are scattered across a site, a directory listing and a social profile that disagree—so answer engines reconstruct the company from whichever fragment they found, and often get it wrong.
A person can browse the website and understand the offer. An agent sent to evaluate vendors has no maintained source for what the business does, where it works, what it charges toward or how to reach a human.
Marketing copy, structured data and third-party listings have drifted apart over years, and nobody inside the business can say which one is current or who is responsible for correcting it.
What the work can include
How the work moves
Score the current state against the readiness ladder and inventory the approved public facts, proof, service boundaries, locations and contact paths. Nothing becomes a machine-readable claim until a person has confirmed it is true.
Connect the human-facing pages and the agent-facing documents so they describe the same business, each claim tied to the page it came from, without inventing authority the business has not granted.
Add the interaction contract that lets an approved agent ask a defined question, with everything it cannot do written into the contract itself. Then verify the live endpoints and set the dates and rules for review.
Proof you can open in a browser right now
River Cade Concepts is not describing a roadmap. Every document below is live on this domain as you read this, and the same system is published on three other properties the firm’s owner controls. Open them, read them, and judge the work rather than the pitch.
The readiness assessment this work begins with, run against any domain from a public tool. Every finding names the URL fetched and the status returned, a check that could not complete is reported as unknown rather than missing, and nothing is stored about the site.
Run the checkA question box wired straight to the live endpoint. Every reply shows the claims it used and the pages they came from, and asking for a price or a date returns a refusal rather than a guess. Nothing on that page is prepared: if the endpoint is down it says so.
Ask the agentWhat the business does, written so a machine can read it — every statement paired with the page a person can check it against. Nothing appears here that is not also on the site.
/.well-known/business-capabilities.jsonThe pointer that tells an arriving assistant where the real record lives and which questions it answers. It follows a published external convention rather than one we invented for ourselves.
/.well-known/ard.jsonThe same business truth in the plain format language models read directly, so an answer engine is working from a maintained source instead of reassembling the company out of marketing copy.
/llms.txtWhat an agent may ask for, and what it will never get. Anything that commits the business goes to a person; anything outside the boundary is refused and recorded rather than quietly dropped.
/.well-known/agent-interaction.jsonOn 28 August 2026 this site moved its entire public hosting to infrastructure the owner controls. Every document above kept serving from the new host without being rebuilt, and the surface check found no conformance defects. A presence that cannot be moved is not really owned.
The same readiness work is published on a creative firm, a farm, a local service business and a commerce and editorial brand — owned ventures, each labelled as owned work.
See the portfolioThis is not a disclaimer added to the marketing. It is published inside the machine-readable contract itself, where an agent reads it before making a request—and it is why an owner can turn this on without wondering what was agreed overnight. Anything beyond these limits stays a decision a person makes.
Relevant proof
Every system below is live. Open it and judge it.
A farm identity rooted in the ground it is restoring.
Read the case study →Commerce · Editorial · MediaBlack Market Cattle CompanyNot a collection of channels. One connected brand world.
Read the case study →Local service · Lead generationShovel & DirtA local operation made useful, memorable and ready to measure.
Read the case study →Consumer software · Information designIs That Really Food?Dense product information turned into a decision surface.
Read the case study →Concept projects
Businesses we invented so we could design them properly. Complete published sites, each labelled a concept on every page.
Before you inquire
It is live. The discovery documents, the capability record and the bounded request contract are published on this domain and on three other owned properties, and the interaction endpoint enforces its documented refusals against real requests. The links in the evidence section above are the system itself, not screenshots of it.
No, and that is deliberate. An approved agent can ask a defined question and receive an answer drawn from published claims, or have a request queued for a person. It cannot negotiate, quote, confirm a schedule, take payment or commit the business to anything. Those limits are published in the contract the agent reads, not merely promised here.
No, and treat any provider who guarantees them with suspicion. Nobody controls what a model says about a business. What a business can control is whether an authoritative, current, machine-readable account of it exists to be used. This work builds and maintains that account.
Answer-engine and generative-engine optimisation name the same underlying shift: being represented accurately where answers are generated rather than where links are ranked. This is the foundation that makes it possible—a consistent, current, sourced account of the business, aligned across the human pages, their structured data and the agent-facing record.
The business does. Everything is generated from a record of your own facts and exports as a package you keep, so the presence can move hosts or providers without being rebuilt. River Cade Concepts moved its own public hosting in a single day, and the published documents kept serving unchanged from the new host.
No. Agent-facing truth is strongest when the human-facing website, message and evidence are already coherent—an agent reading a confused business publishes a confused summary of it. River Cade Concepts can build the complete presence or add this layer to a sound existing system.
Have a business that is better than it looks online?