News

gettyAI agents are increasingly able to search, compare options, navigate digital services and take actions online on a user’s behalf. As these tools become more capable, websites and apps may need to serve not only people navigating screens themselves but also software interpreting information and interacting with digital systems for them.

That shift raises a new design question for businesses: How should digital experiences change when the visitor may not be a human customer but an AI agent acting on a customer’s behalf? Here, members of Forbes Technology Council share one way businesses should rethink their websites, apps or digital services as AI agents become a more active part of online discovery, decision-making and transactions.

Separate Price Quotes From Inventory ReservationsLet agents compare prices without tying up stock. An agent may fill 10 carts while deciding what to buy. If each cart reserves inventory, customers ready to pay can find those items unavailable. Separate quotes from reservations and ensure unused holds expire. Your website shouldn’t say “sold out” because a bot is still making up its mind. – Abdullah Rashid, ClearTraced

Direct Agents To APIs And MCP ServicesDirecting AI agents away from unstructured websites to callable APIs should be the first priority. These APIs will now act as a form of prompt engineering for the agents. If you deliver these APIs via the Model Context Protocol, it is crucial to focus on the tool descriptions to influence how the agent behaves and understands how your business operates. CAPTCHA your human forms and direct agents toward your MCP services to control the interaction. – Siddhartha Rao, Oracle

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?

Design For Evidence Over PersuasionBusinesses should design for evidence, not just persuasion. AI agents won’t rely on brand familiarity, intuition or “feel” the way people often do. They will compare what they can access: facts, pricing, capabilities, policies and proof. Clear, structured and verifiable claims will increasingly determine which products or services agents recommend and choose. Include fewer “contact us” sections to get access to information, because an agent might penalize a service it needs to wait on. – Samuel Kaluvuri, ApyHub

Design For Humans And AI AgentsBusinesses should start designing for both humans and AI agents. Today, websites are mostly designed for humans, with beautiful CSS and media; humans can digest such content very easily and quickly. For AI agents to understand and use a website, they need the website to be easy to parse, in JSON and markdown format, while also having a clear and consistent structure. – Anand Chaudhary, Paragon

Verify Agent Identity And PermissionsWe will need to distinguish legitimate agents acting with a user’s authorization from malicious bots, impersonators and autonomous agents operating beyond their permissions. Agentic interactions introduce a much more dynamic trust problem. The next generation of digital services should therefore be both agent-friendly and agent-aware: easy for authorized AI agents to interact with while continuously verifying identity, intent, permissions and behavior. – Joydeep Mukherjee, Akamai Technologies

Replace Anti-Bot Barriers With Agent-Friendly NavigationClear LLM.txt files, markdown files and a clear sitemap are all key to ensuring smooth agent navigation and experience. Agents are the consumers of tomorrow, and we need to rethink all the anti-bot precautions from yesterday. – Haz Hubble, Pally

Build Catalog Intelligence For Agentic ShoppingClaude for Commerce validates that agentic shopping shouldn’t be ignored. But it’s important for brands to know that general LLMs are reasoning engines, not retail brains. Whoever owns the best catalog intelligence will “win,” regardless of which LLM hosts the conversation. While general models excel at dialogue, conversation alone doesn’t drive conversion. Brands should enable inventory controls, margin-driving merchandising and personalization. – Zohar Gilad, Fast Simon Inc.

Rethink Ad Value And Security For Bot TrafficBots already represent more internet traffic than humans. That has implications for both demand and security. In terms of demand, businesses must ask fundamental questions like, “What are ads worth when the audience is a bot?” In terms of security, infosec teams will need to defend what agents see across a growing set of attack surfaces, not just what humans see. – Ryan Woodley, Netcraft

Update Every Page An Agent Might ReadYour redesign should consider that agents will read everything, including the boring pages and the forgotten docs. A person prioritizes the content they review, but an agent considers everything, and if the content does not make sense, moves on to the next player. Yes, you need to be agent-friendly, but first update what’s already there, because your worst-maintained content can be a deal-maker. – Akash Pugalia, TP

Publish Business Capabilities In DNSYour website was built for eyes, and agents do not have any. Before loading a page, an agent asks DNS where to go. Publish the answer there: a DNS-AID listing of what your business can do, DNSSEC to sign it so it cannot be forged, and TLSA to prove the destination is really you. Otherwise, a lookalike answers for you. – TK Keanini, DNSFilter

Optimize Content For Generative EnginesA good starting point is understanding the principles of generative engine optimization and following best practices such as providing clear, structured product information and maintaining authoritative, up-to-date content written in question-answer format that AI systems can easily understand and use. GEO does not replace SEO, but it is becoming an equally important element of online discovery. – Pawel Rzeszucinski, WebPros

Measure Agent Traffic SeparatelyFix your measurement before agents break it. Yes, build the machine-readable product layer. But the bigger miss is when half your “visitors” are agents shopping at machine speed, your human-centric analytics will quietly lie to you. Attribution, conversion rates and funnel metrics are all trained on behavior patterns agents don’t follow. Instrument agent traffic separately before it corrupts every number you report. – Ambarish Majumdar, Meta

Verify Agent Authorization ChainsStop assuming a visitor is a person, and decide how you will tell an agent acting for a real customer from an agent acting for someone else. Most sites authenticate a session, not the chain of delegation behind it. When an agent places an order, what matters is who authorized it, with what scope and whether that can be verified afterward. That is an identity problem, not a front-end one. – Aviv Mussinger, Kodem

Provide Verifiable Information And ConfirmationsDesign for verifiable decisions. Give AI agents clear, structured information about prices, availability, terms and permitted actions, then return explicit confirmation of what an action accomplished. A service should help an agent determine whether it meets the user’s intent. That requires reliable evidence at each step, including when an action fails or needs human approval. – Casey Kindiger, Droma (formerly Grokstream)

Design For Agent Actions, Not Just ClicksBusinesses must stop designing only for clicks and start designing for AI-driven actions. Websites should become machine-readable, API-friendly, transparent and permission-aware so AI agents can discover, compare, decide and transact confidently. The winning digital experience will serve humans beautifully while giving intelligent agents a clear path to act. – Bindu Madhavi Mangalampalli, Cotiviti

Make Customer Experiences Machine-ReadableStart asking, “Can an agent understand and safely act on what our customer sees?” Product details, pricing, policies and available actions need to be unambiguous to machines as well as people. As more journeys begin with an AI intermediary, discoverability may depend as much on machine-readable experience as traditional UX. – Prajkta Waditwar, Box Inc.

Make Action Consequences ExplicitBusinesses should make the consequence of an action explicit. An AI agent should be able to read what a purchase, cancellation or upgrade will change, including price, timing and terms, before committing. If that outcome is unclear, ambiguities a person might catch can turn into mistakes that scale at machine speed. – Nicholas Domnisch, EE Solutions

Design For The Full Customer LifecycleDesign beyond checkout is key. An AI agent will not only buy; it will monitor delivery, claim warranties, switch plans, cancel renewals and resolve disputes. Businesses should expose the entire customer lifecycle as safe, reversible actions with clear deadlines, fees and evidence. The competitive question becomes: Can a customer’s agent improve the decision after purchase, or does your service trap it in a human-only support maze? Make post-sale freedom a product promise. – Jagadish Gokavarapu, Wissen Infotech

Keep The UI While Adding An Agent Entry PointOriginally, I thought UI was dead and everything would live in agent chat. Then we realized humans are visual. People say an agent can draw any screen, so you don’t need a UI. That’s wrong. Our brains want a stable layout we can glance at, not a long chat every time. Otherwise, everyone builds their own view, and the company loses one shared picture of what’s correct. So UI stays. An agent is the extra door for what the main flow can’t cover. If your service isn’t available through secure MCP, you don’t have a service. – Andrii Stetsenko, Wyllo