If you’re being asked about your strategy for agent-based capabilities, this is what you need to know: there’s a good bit of work your company needs to do to get ready. The good news is, whether the agentic web becomes the next big thing in six months or six years, you need to do this work anyway. Much of it is just foundational.
The companies that will succeed on the agentic web are those that focus on the underlying capabilities rather than chasing the latest AI agent platform.
From clicks to outcomes: what changes when agents take over
The agentic web refers to AI systems that can act autonomously on behalf of your customers to complete transactions and tasks. We’re not talking about internal automation or chatbots within your business processes here. We’re talking about your customers’ agents transacting directly with your company.
Think of an agent that has been tasked by an individual with balancing their investment portfolio across multiple retirement accounts – three 401ks, post-tax accounts in different places – maintaining a specific allocation like 40% domestic stock, 20% foreign stock, with the rest split between bonds and cash. The agent handles the rebalancing automatically while minimizing tax impact and maximizing returns. All without any human interaction with any of the parties involved.
Whatever the equivalent scenario is for your industry, your company needs to be ready to serve that agent seamlessly, just as you would serve the customer directly. And if it’s not? Well, goodbye agent and goodbye customer.
Of course, this changes the entire user experience paradigm. We’re no longer talking about clicks, we’re talking about outcomes. And we’re talking about a whole new layer for the internet that’s going to be laid in: one that’s for AI-first agent-based systems.
We’re moving beyond human-driven actions and beyond simple automation like “if this then that” type stuff, which has been around for a very long time. Now we’re talking about independent actions where you have to define your intent so that it can be fulfilled by automated systems.
Getting ready for agents
The big thing that companies need to understand is that agents don’t click on ads or browse sites – they complete tasks.
There’s a difference between somebody who’s browsing, somebody who’s looking with intent, and somebody who’s arriving to perform a transaction. Agents represent that person who wants to perform that transaction.
At the moment, as customers, pretty much all we can do is browse: “go find me all the products that look like this.” What’s not supported is “go buy me a used pair of shoes that are at excellent or better condition in this size and color, wherever they come from: eBay, Amazon, Facebook marketplace, everything.”
Browsing, in the traditional demand funnel, gets replaced by algorithms and SEO starts to become less relevant. Agent optimization becomes more important because if it’s not easy for an agent to complete a transaction with your company, it’s not going to.
This changes the relationship with the customer fundamentally: if autonomous agents are acting on behalf of your users, you may not meet your customer directly. You’re fungible as a provider. If there’s a better option for the service, the agent is going to pursue that.
All of a sudden, you’re in ‘real-time’ mode.
When every transaction becomes an auction
Your KPIs are now completely different. You have to look at where the bottlenecks are when it comes to task completion – where do transactions fall off? Your digital success metrics move away from PPC, SEO and page views, and go to an automated pipeline. How far through the pipeline do you get? And why do people fall off? The same problem we have today, just better instrumented.
Customer retention becomes challenging when agents can switch providers constantly. Every single time window where you’re allowed to make changes becomes an auction. And it’s a multi-dimensional auction: price, risk, amenities. You’re trying to optimize along each of those vectors, and that’s something people are not very good at (but agents are).
This shift is already happening in adjacent areas. Look at how price comparison tools and automated switching services have commoditized industries like insurance and utilities. The companies that survived those transitions were the ones that had already built superior underlying capabilities – better pricing engines, faster onboarding, more reliable service delivery. Agent-driven commerce will accelerate this same dynamic across every industry.
If you’re a commodity-based business, efficiencies are going to take you down. There’s no arbitrage between what you can charge and what you have to pay. You’re going to have to deliver real value. The competitive advantage narrows over time because everyone that you’re competing against is going through the same decisions.
Data, APIs, and governance: the three pillars you need (regardless)
Getting ready for this shift is a matter of building machine readable capability. Agents have to be able to transact with your business on your customers’ behalf without going through interfaces that were designed for humans.
Specifically, we’re talking about data and APIs (not GUIs). Three things are mission critical: your data architecture, your API infrastructure, and your governance model. Don’t think about this as a tech upgrade but as a significant operational transformation for the business.
Clean, accessible data
Your data needs to be in good shape. If you have data silos, and inconsistent formats, and legacy systems that are bottlenecks, that’s going to be a real big problem. Now, you should be modernizing the enterprise to begin with, but you may need to create a clean operational data store that has all of your current information in it – one place for agents and other transactions to go to.
This data problem shows up everywhere. I’ve seen very large financial services companies where they had multiple data stores and they didn’t agree. Depending on which service you’re asking, you get a different answer to your question. There was no way they could say “this is our unified vision of the customer” because they have 17 operating units and contradictory stuff about their customers spread all over.
Without a holistic view, agents can’t make optimal decisions but these are existential problems whether you’re transacting with agents or not.
Well-documented APIs
You need robust, well-documented and semantically-rich APIs for your services. You need to adhere to standards like OpenAPI, schema.org and JSON-LD so agents can understand the interfaces and interact with them meaningfully.
You have to be responsive enough for the demands of an agent. You start looking at what amounts to real-time access. If you’re lagging, that’s a problem because agents will make decisions based on old data and could commit to a transaction without having updated information.
Trust and permission frameworks
You have to understand everything that an agent is permitted to do for a person. Agents can discover, retrieve, modify, transact. All these things are on the gamut: you have to decide whether or not it’s permitted, given the intersection of what the agent’s trying to do and who they represent.
If an AI agent’s autonomous, you have to grant autonomy to your agent and that authority has to somehow be reflected. And if the agent is misbehaving or compromised? This is a whole other nightmare.
Imagine your bank accounts drained, paying for things that you didn’t want to pay for, agreeing to contracts you didn’t agree to, starting automated collections processes, driving a tax auction for your property. Every bad pattern you can think of can be automated.
So guardrails are extraordinarily important, especially in regulated industries. Underlying frameworks around authentication, authorization, and audit trails are essential. You have to be able to kill an agent when it’s out there running. You have to have very good transactional semantics and contracts around what the agent can do.
The ability to explicitly define and grant decision-making authority to agents is where the new competitive advantage lives.
Why this work pays off today
While these guardrails become strategic differentiators when the agentic web becomes reality, the truth is you need to do this work anyway.
If you’ve addressed your tech debt using something like theory of constraints, if you’ve identified your critical challenges, and if you’re addressing them around your key areas of data, API, and governance, you’re moving in the right direction.
The know-your-customer aspects of getting your data in order are urgent today. Companies need to be able to say “this is our unified view of the customer” if they want to make any progress in improving customer experience, marketing or operational efficiency. You have to solve those problems no matter what. The fact that solving these existential problems makes you ready for the agentic web (and everything else) is, for most businesses today, just an added bonus.
40% will fail: managing the risk of early agent adoption
Internal agents have a lower risk profile because you know who all the people involved are. The transactions you run internally are different from the ones you run externally. The things that you do that you don’t sell are the internal work that you can automate.
Gartner predicts over 40% of agentic AI projects will be canceled by the end of 2027. They recommend these projects only be pursued where they deliver clear value or ROI. That’s a great idea. I agree. You have to understand what your intent is and what your outcomes are, and you have to be very specific about that.
There’s both cost and complexity in deploying AI agents, and a lot of work you have to do up front. The current ecology is very young, just emergent. CB Insights notes that there are over 170 AI agent startups in a lot of different industries, but when you start slicing and dicing those across all the different categories, the notion of agents working together is really very early. There’s a lot of agent washing where existing products have just been rebranded as ‘agents’.
Consolidation is going to happen. You’re seeing plays for talent because talent’s really tight. You’re seeing plays for market share and you’re seeing plays for features (between writing the first draft of this article and publishing, we’ve seen the acquisition of agentic IDE Windsurf and Amazon launching its own agentic IDE, Kiro). Building features takes time and building market share takes time. Big companies like Meta are throwing around hundred-million dollar first year salaries to more than 10 people, according to Zuckerberg.
There are going to be orphan technologies out there – really good ideas that get left behind because they were more challenging to adopt or weren’t as well described and defined.
The key thing to remember is that while an experiment either succeeds or fails, you get information either way. Only do it if you’re going to realize real business value. If you want to seize competitive advantage, you have to make bets. You can’t just say “let’s choose one of the top three vendors.”
Getting started without overcommitting
We’ll know more about how this is going to pan out in a couple quarters. We’ll know a lot more in six quarters, but ultimately the pragmatic solutions are going to win out over the elaborate solutions. You see companies like PwC with their agent OS trying to be all things to all people. Much more focused approaches are coming from the 170 agent startups because they’re essentially hand tooling agents.
The competitive dynamics are already shifting. As a mid-market company, you have about an 18-month window where you’re more nimble than enterprises and more resourced than startups. Large corporations are paralyzed by procurement cycles and compliance reviews – there are companies spending six months just deciding which AI vendor to pilot with. Startups are burning through runway chasing every new framework. You can move on the foundational work now while your competitors are stuck in analysis paralysis. But once the big players sort out their bureaucracy and the startup space consolidates, that advantage disappears.
Do the housekeeping work now, or you’ll be stuck playing catch-up later.
You have to have workflow orchestration and event driven architectures to support the agent orchestration. Your data needs to be in good shape. Your APIs need to be in good shape. You have to have the trust and permission frameworks (governance) in place.
Removing friction, not chasing trends
Once again, it all comes back to first principles: what are we trying to do? We’re trying to remove friction between transactions between customers’ clients (agents) and the company. This is a question of moving agency, where the person determines the intent but they delegate authority to the agent to achieve the outcome.
Defining intent and defining outcome are extremely important and that’s not something you can outsource. It’s not something that a large language model can do or that an agent’s going to solve for you.
The outcome includes things like: how good of an answer do I need? “I need with a 95% certainty that this decision is the right decision. I don’t want to lose more than 20% of this investment and I want to make at least 15% per year.” These things are notoriously difficult for people to define. Determining intent and outcome is the part that’s not going to get automated. That’s where you need to focus on developing competitive advantage.
But we’re getting ahead of ourselves: you’ve got to do all your housekeeping before you get new toys. You may not know if competitors are already stealing a march on you, but you have the table-stakes stuff you have to address whether you adopt the agentic web or not.
That’s the pragmatic view. You have to address your data, your APIs, and your governance – the key capabilities that create value whether the agentic web becomes the next big thing or doesn’t.