The Rise of the Agentic Customer Journey

How will upcoming agentic AI change the customer journey?

It's a frosty morning in 2026. Your old winter boots have finally given up, leaving you scrolling for replacements – except you're not actually scrolling. Instead, you're having a conversation.

Your AI agent springs to life through your earbuds, ready to handle your shopping needs.

“I’m looking for winter shoes that can handle slush, snow, and water, but still look stylish around the town,” you utter to your assistant from your headphones while waiting for the bus.

“On it,” replies the assistant in a pleasant British female voice. As you sit down and the bus takes off, you receive a notification with a short list of 5 winter shoes within your budget, overall style and size that the agent already knows. “I’ll take the second one,” you reply with a text to your AI agent assistant, and it places the order for you.

Two weeks later, you receive the package and, of course, the shoes are too small. Again, you send your agent off to do your bidding. Within a few minutes, with your approval, it independently submitted a return form to the retailer and orderer for the return label from a printing store.

Does it sound like science fiction?

Far from it.

Agentic customer journeys mean we direct AI agents instead of navigating websites and apps 

Independent AI agents are the most significant development in AI heading into 2025.

While ChatGPT or Claude responds to your prompts one conversation at a time, AI agents are more like autonomous digital workers. They can:

• Plan and execute multi-step tasks independently

• Use various tools (like searching the web, writing code, or accessing databases)

• Remember context from previous interactions

• Learn from mistakes through reflection

• Work with other agents to solve complex problems

Think of it this way: If ChatGPT is like having a brilliant conversation partner, an AI agent is like having a proactive team member who can take a goal and run with it.

All frontier AI companies are placing significant bets on agents:

• Kevin Weil, CPO of OpenAI, revealed in a recent Reddit AMA discussion that ChatGPT’s ability to act independently will “be a big theme in 2025”. The company is rumored to ship “Operator,” their AI agent that can autonomously control computers, in January 2025 - initially as a research preview.

• Google focused significant portions of its 2024 IO conference on agentic AI embedded deep into their search and the broader Google ecosystem. Their upcoming AI agent, Jarvis, is expected to launch in 2025.

• Anthropic recently released a beta version of Computer Use. With Computer Use, Claude, their Large Language Model, jumps out from the confines of the chat window and takes control of your computer. In their demo, it filed forms for the users with data it gathered from various websites and local files.

Upcoming AI agents and their impact on digital services, customer experience, and design have been some of the most common questions I’ve recently received in client workshops and events.

So, let’s dive into agentic AI with the lens of customer experience. In this essay, I’ll analyze agentic experiences in three sections:

1. What are AI agents?

2. What are agentic experiences?

3. How should we design for agentic experiences?

1. What are AI agents?

Agentic AI means that over the coming years, AI will be increasingly capable of taking independent multi-step actions based on user goals.

One way to understand AI agents is through a Sense-Think-Act loop. AI agents can:

1. Sense - Agents take in new inputs from user prompts with voice, text, video, or images. They can also initiate the loop using other data sources like APIs or taking screenshots of interfaces like Claude’s Computer Use.

2. Think - Given the input, AI agents use frontier LLM capabilities to understand it. They can augment this reasoning with knowledge from their memory or retrieve additional data from external sources like company documentation.

3. Act - After thinking through the input, AI agents can take independent action. They can simply answer the user’s prompt or take more complex actions - like navigating websites, editing documents, talking to other AIs, communicating with APIs, sending text messages, or even placing increasingly realistic phone calls.

The concept of agentic AI and the Sense-Think-Act loop has been around for decades in robotics and self-driving cars. We’re now seeing AI agents that increasingly have generative AI capabilities.

Generative AI means future Sense-Think-Act loops can range from simple tasks like filling a form with 50 different vendor details or, soon, completing hundreds of loops to manage complex multi-step projects.

This shift to AI agents represents a new paradigm for how we interact with computers:

1. Traditional Graphical UI’s have meant we, as users, click and tap around the web to achieve our goals. The user is the active agent, and the organization serves them with interactive but simple rule-based websites and apps.

2. First AI bots have introduced a new conversational paradigm for interacting with AI assistants. These first bots, like Claude and ChatGPT, mostly answered isolated prompts within the confines of the chat window.

3. AI agents take independent multistep action based on our goals. Users will increasingly operate through voice and chat with their agents, who get things done for them with high-level oversite.

The potential of AI agents is already being demonstrated in practical applications:

• Legal firms like Reed Smith report that their lawyers accomplish significantly more work in the same amount of time using Harvey's agent-based workflows

• Companies like Glean are enabling enterprises to create custom AI agents that can automate everything from IT help desk requests to sales support

• Recent research from METR, a research non-profit, shows that some AI agents can handle about 40% of human tasks, and when successful, they do it at just 4% of the human cost

The transition from clicking buttons to commanding AI agents won't happen in one dramatic sweep. Like smartphones replacing flip phones, this shift will arrive in waves, with early adopters blazing the trail while others hold onto familiar interfaces.

Beyond technical limitations, questions around privacy, security, and potential harm call for a measured approach.

So, how will this new paradigm of agentic AI change the customer experience?

2. What are agentic experiences?

Not surprisingly, software engineering is one of the first areas AI agents are popping up in.

Devin, a collaborative AI teammate for developers, can:

• Work independently on coding tasks across multiple files, even fixing its own errors until it succeeds

• Access project management tools like Notion and Jira

• Search the web for needed documentation

Devin is an agentic AI teammate to software developers that can independently write code, browse the web and access documentation. 

These first agents show us that agentic experiences will be different in three core ways:

1. Users will become directors, not browsers

As in the example of buying new winter shoes, customers or users will become increasingly directors of AI agents instead of active online browsers.

Instead of googling, looking at reviews, and comparing products, we’ll direct AI assistant who:

• know all the selections of all providers

• navigate websites and apps for us

• know our preferences, goals, and other relevant background information like our calendar

In a business setting, we’ll be less likely to navigate bookkeeping software and accept an expense report directly generated by an AI as they notice we take a photo of a lunch receipt with a client.

2. User journeys will become short and conversational

In the last 6 months, my use of Google has already plummeted as I get direct answers to questions from Claude and Perplexity (which combines search with conversational AI).

Few of us are looking to browse websites or apps; we usually want to accomplish our goals.

Future customer and user journeys will reflect this - AI agents will find our information and take appropriate action without us visiting websites and apps.

This might also mean that AI agents will exacerbate the winning-take-most dynamics of the web. Your agent will not present you with 20 different shoe options, just the five it thinks will suit you the best. Some AI agent providers will likely favor their products or products an advertiser will pay them to highlight.

This leads us to the following change.

3. User journeys might be owned by a handful of frontier AI companies

The web era consolidated money and power to the FAANMG companies - Facebook (Meta), Amazon, Apple, Netflix, Microsoft, and Google.

This development will be boosted as we entfer the agentic AI age. Outside of entertainment and social media, we’ll increasingly have fewer reasons to leave the cocoon of our AI agents.

In the enterprise context, specific vertical AI agents will have an edge as their success rests on deep domain knowledge and relationships. We’re already seeing this with the population of AI agents in niches like car payment collection with agent-placed phone calls.

3. How should we design for agentic experiences?

Companies should now explore how they can serve customers or improve internal efficiency with agentic AI experiences.

Maybe it’s:

• a low-cost mobile provider that makes it easy for consumers to compare and switch their provider

• a fashion retailer that builds an AI agent fashion assistant that helps them dress for any occasion

• an insurance company that takes away most of the hassle from filing a claim with a conversational and agentic AI assistant

• an HR software that handles most of the routine HR tasks as an agent

The more companies can add value with shipping AI agents earlier than their competitors, the more they can own the customer journey over the next five years.

While it’s early, I’ve formed some principles for designing better agentic experiences working with clients and studying public use cases:

1. Clear Agency Boundaries & Control

• Set clear human control and boundaries for the AI.

• Provide clear ways to override or adjust agent behavior

• Help users understand the scope of delegated authority

• Require conscious human opt-in to access any sensitive data - like calendars or messages
  

2. Transparency & Predictability

• Make agent reasoning and decision-making processes visible - For example, Devin continuously updates his team on the steps he’s taking

• Help users build accurate mental models of agent capabilities

• Show confidence levels and uncertainty in agent decisions
  

3. Graceful Failure Handling

• Design for edge cases and unexpected situations

• Make it easy for users to take over when agents fail

• Provide clear explanations of what went wrong

• Give actionable steps to resolve issues
  

4. Trust Building

• Start with low-stakes tasks where mistakes have minimal impact

• Build confidence through consistent, reliable performance

• Show work and reasoning to help users calibrate trust

• Enable easy verification of agent actions
  

5. Collaborative Intelligence

• Design for human-AI collaboration rather than full automation - for example, lawyers check on AI agent drafts of contracts

• Make it easy to combine human and agent capabilities

• Enable smooth handoffs between human and AI agent - for example, I worked with AI startup LastBot to design responsible, transparent and appropriate handover experiences between AI agents and customer service personnel

Reliable, helpful, and personable AI agents will become the competitive moats of the future. The more they know about us and help us achieve our goals, the less likely we will drop them.

Next year will be an opportune time for companies to:

1. Learn deeply about the AI agents and the opportunities they bring to their domains

2. Analyze current customer and internal user journey to identify potential areas for AI agent impact - think slightly complex, currently manual and data-heavy workflows

3. Prioritize AI agent pilots in areas that are potentially high-impact (cost-saving or revenue-generating) but still relatively low risk with potential failures

4. Ensure high quality internal data access for agents and the design principles discussed earlier

Need help with exploring AI agents in your domain? You can reach me at matias@vaara.co to start a conversation. After an extremely hectic Q4/24, I’m open to new client collaborations in February 2025.