Defining an Agentic AI-ready digital growth strategy is no longer a cool experiment. It has become an immediate operational and strategic imperative.
Our CX and commerce playbook has changed overnight. For years, we focused on optimizing for human eyes, crafting beautiful checkout flows, A/B testing banner ads, and triggering pop-ups to capture human attention. But with agentic AI, consumers will increasingly deploy autonomous, 3rd-party AI agents (like Meta’s Muse, xAI’s Grok, or OpenAI’s shopping assistants) to handle tasks: rebooking flights, managing recurring bills, purchasing goods, scouting discounts, making payments, looking up accounts, and optimizing bank balances.
The immediate corporate response to this “proxy access” has been defensive. That’s understandable given the risk to consumers themselves, and also to maintain compliance with current laws. Financial giants such as JPMorgan Chase, Citi, and Charles Schwab are actively locking down against unvetted bots logging in and taking actions. Delta Air Lines has barred agents from completing ticket purchases, and platforms like Amazon and Yelp have blocked agents or demanded paid data licensing.
Their concerns around cybersecurity, account liability, and lost advertising revenue are understandable. At the same time, shutting the front door on customers’ chosen assistants is not a viable long-term strategy given that we are moving to an agentic world.
So how to shift from defensive blocking to proactive integration?
Note: The industry has rapidly moved to develop various technical frameworks and protocols to solve this balance across agentic access and human control. Adoption and consensus across merchants and organizations is the next step. In this blog, I lean towards OpenID’s identity management framework for agentic AI.
Here are 4 enterprise capabilities we should start thinking of:
1. Build a Front Door with Scoped Authorization For Agents
Instead of forcing agents to access websites and portals in disguise, we should plan to provide a structured, authorized entry point by adopting explicit agent authorization.
Much like users sign into third-party apps via “Sign in with Google,” digital platforms can provide scoped rails (OAuth or otherwise). Most prominent SaaS apps already do this to provide access to other apps. Customers can then grant an AI assistant specific permissions—such as authorizing the agent to view inventory, book reservations, move money, and other such use cases. It is helpful from a risk management perspective as well.
While enterprises are comfortable building their own internal concierge bots, the next imperative frontier is creating secure, federated “front doors” so external customer agents can interact without resorting to credential scraping.
As I outlined in my post on Enterprise AI Maturity Model for Autonomous Agents, this shift marks the leap from task automation to true ecosystem interoperability.
Companies have started taking small steps towards this agentic goal but not quite on the way to opening it up yet. For example, Charles Schwab launched an assistant that can check status, manage beneficiaries etc. This is a positive step because it is likely to evolve into the open banking API driven concept.
We have the enterprise infrastructure to do all this already. Risk can be mitigated by opening it up to known, limited agents first before going broad.
2. Preferences Management for Desirable Actions
Product owners worry that an agent will bypass the entire storefront and head straight for the cheapest checkout option while ignoring lucrative upsells, sponsored listings, and premium add-ons. For example, a bank would promote signing up for e-delivery or account balance alerts, and an insurer would want the user to sign up for payment alerts and auto-pay.
An agent is task focused so it doesn’t look at these offers. To overcome this challenge, preferences and permissions can be published as structured options for the end user to select as a way to enable the access for an agent. Incoming agents will adopt their user’s preset guidelines during the actions. That way an agent will sign up for preferred airline seats if within a budget threshold, and sign up for e-consents, account alerts, and eco-friendly shipping options. Enterprise applications will help enforce these preferences.
Travelport recently rolled out agent-ready APIs designed so automated assistants can programmatically view and select bundled travel options. While not yet a full preferences system, these kind of APIs will probably evolve to handle more granular preferences.
Most enterprises have these built already as part of their digital CX strategy. We now need to expose them for agentic use.
3. Implement Value-Recovery and Validation Campaigns
Agent preferences and settings outlined in the previous section should address apprehensions that agents will bypass offers and go right for the core transaction.
But there is more that can be done to make it even more engaging and productive which marketing and CX folks would love. If an agent executes a standard transaction and bypasses a limited-time upgrade, we can consider that event as an immediate marketing notification hook rather than a lost sale!
These hooks would be in the form of automated, post-transaction marketing loops aimed directly at the human buyer. When an agent finalizes a purchase, the merchant’s backend can instantly follow up with an email or app notification highlighting what was left on the table.
Imagine an agent books a standard business hotel room. The hotel’s CRM immediately pings the guest: “Your assistant just reserved your stay at our standard rate. It missed signing up for discounted breakfast. Do you want to sign up for that?”.
As another example, consider Amazon’s eCommerce innovation where it prompts buyers to “Add to upcoming delivery?”.
This preserves your brand’s direct consumer relationship, personalizes the upgrade process even more, and protects premium margins without annoying users during the initial booking.
4. Enable Two-Way Negotiation Loops
Agentic interactions should not be limited to single-shot agentic actions. The future of commerce is a dynamic dialogue between the enterprise, the customer’s agent, and the human buyer.
We can allow negotiation of counter-proposals in real time. If an incoming agent requests a sold-out item or is missing a discount that needs an additional purchase, the system can return an alternative bundle. If thresholds are not already configured, the agent can then route an asynchronous validation check to the human, e.g. via a lock-screen push notification or SMS.
For example, an agent attempts to order a laptop from an electronics retailer. The retailer responds: “Discount is 20% if you bundle with a 3-year protection plan.” The agent surfaces a quick prompt to the user’s mobile screen, the user taps to approve, and the merchant securely processes the expanded sale. Agents have gotten (or will soon get) smarter to surface these scenarios themselves, but enterprises should be ready to engage in a dialog with agents.
The Regulatory Push & Technology Readiness
This shift isn’t just about digital business strategy to keep up with agentic AI but is quickly becoming a legal requirement.
Global regulators are rapidly shifting their focus to how we are protecting our customers and auditing what autonomous agents do.
In the US, the Consumer Financial Protection Bureau (CFPB) finalized its open banking rule in October 2024. This rule prohibits screen scraping and mandates that banks provide fee-free, secure, structured data access to authorized third parties through developer APIs. However this rule ran up against legal challenges claiming that it exposed banks to security liabilities and forced them to absorb massive infrastructure costs. So the April 2026 enforcement deadline was not met. Also in the US, courts routinely enforce “user agreements” that prohibit automated queries, screen scraping, and proxy logins. This opens up agentic AI companies to major liabilities. In the EU, the AI Act is moving to make it illegal to deploy fully autonomous agents in high-risk environments without strict traceability and a human kill switch.
It follows that while agents are getting around these issues today through various tactics, businesses cannot satisfy robust CX and commerce without proactive action. They also cannot fight consumer demand by blocking external agents forever.
Implementing scoped authorizations and interactive confirmation loops will strike an optimal balance between customer experience, marketing strategies, and regulatory compliance.
The technical transition is already underway.
Open frameworks like Model Context Protocol (MCP) are rapidly standardizing how LLMs interface with various APIs, enterprise applications, and internet access. The OpenID authentication framework for agentic AI is already being widely adopted by agent developers.
Embracing Open Ecosystems
By designing our digital commerce and customer engagement strategy around secure authorization, structured preferences, and two-way negotiation loops, we can transform agentic AI from a disruptive threat into our most efficient sales force.
Where does your organization stand on this transition? Read my companion guide on the Enterprise AI Maturity Model for Autonomous Agents to evaluate your organization’s readiness across governance, infrastructure, and agent federation.