What Gets Skipped When AI Moves Too Fast

Most organizations are treating AI like an automatic upgrade. Faster answers, fewer tickets, more productive teams, lower costs. The promise is seductive, and the investment reflects it. Global AI spending surpassed $300 billion in 2024, and the tools are now everywhere.

Yet the experiences people are having with those tools are getting worse.

Forrester’s CX Index hit its lowest point in 2024, marking the third consecutive year of decline and the lowest level in the history of the index. By 2025, 63 percent of customers reported frustration with AI self-service tools. Those numbers are not a coincidence. They are a signal. 

Here is the uncomfortable truth underneath all the deployment momentum. AI does not fix broken experiences. It scales them.

If your journeys are confusing, your policies are inconsistent, or your internal workflows are already being held together by goodwill and workarounds, AI does not resolve any of that. It standardizes it, wraps it in a confident tone, and delivers it at speed and at volume. The people on the receiving end feel that difference long before the dashboards do.

The organizations racing to deploy fastest are not just discovering what AI can do. They are discovering what their real experience gaps look like, with nowhere to hide.

AI Is Already Designing Experiences, Whether You Meant It To Or Not

Here’s something I’ve noticed working across a lot of AI deployments. Most teams do not sit down and intentionally design the AI experience.

They buy a platform. They stand up a pilot. They configure some flows, write a few responses, and plug it into existing channels. Everyone is moving fast, usually under pressure. The assumptions are familiar. We will learn as we go. We will tune it later. This will at least take some pressure off.

What gets missed is a much more important question. Who actually designed the experience before it got developed?

Design for AI is not about colors and buttons. It is about decisions that shape the real experience people have. What tone do we use with a customer who is already frustrated? Where do we accept ambiguity, and where do we insist on a human? What is the escape hatch when the system is wrong, and how visible is it? How does this change the day-to-day reality of the people who now have to work alongside it?

Those choices exist in every AI deployment. If you do not make them deliberately, they still get made. They show up through vendor templates, technical constraints, default settings, and the path of least resistance inside the organization.

There is a meaningful difference between we configured options and we designed an experience. A lot of deployments are the former, presenting as the latter.

Air Canada learned that when its chatbot gave customers refund guidance that did not exist, and the company was still held responsible. Klarna learned it when it pushed hard on AI in customer service, saw quality concerns emerge, and later adjusted course by bringing human support back into the picture. Stanford’s 2025 AI Index also reported that documented AI incidents rose 56.4 percent year over year, reaching a record high in 2024. These are not edge cases. They are signs of what happens when deployment outruns design. 

CX And EX Are One System. We Are Still Treating Them As Two.

Most AI roadmaps still separate customer facing AI, chatbots, virtual agents, recommendation systems, from employee facing tools such as assistants, copilots, and workflow automation. In lived reality, they are the same system.

The most common failure is not AI that deflects too aggressively. It is AI that was never capable enough to be there in the first place.

A bot trained on two scenarios gets deployed across fifty. It does not have the right information. It cannot handle the actual question. So it serves up something adjacent, loops the customer through the same options, or hands back a response that sounds helpful but is not. The customer has now spent time getting nowhere, and reaches a human already irritated, already repeating themselves. The AI did not reduce friction. It added another layer of it before the friction that was already there.

The honest design question is not, can we deploy AI here? It is, is this AI actually capable enough that deploying it improves on doing nothing?

If it can only handle a narrow slice of what customers bring, it should not be the front door.

That frustration does not disappear when the interaction transfers. It lands on whoever picks up. That is where the employee experience becomes the customer experience problem, because when AI generates outputs employees do not understand or trust, it shows up as rework, shadow processes, and quiet resistance. When internal tools add oversight without adding clarity, employees feel surveilled, not supported. The two systems are connected. A breakdown in one amplifies the breakdown in the other.

The employee side of the data is especially telling. Upwork’s 2024 research found that 77 percent of employees using AI say it has added to their workload, not reduced it. Seventy eight percent resort to unauthorized tools because approved options do not meet their needs. One third say they are likely to consider leaving within six months. That is not just a workforce issue. It is an experience issue with downstream customer consequences. 

Every customer interaction that reaches a human, through escalation, complexity, or simple preference, flows through that employee’s working environment. When the environment is fragmented, you can hear it immediately. Let me pull that up in another system. Can you hold while I check something? I need to transfer you because I do not have access here.

Those pauses are employee experience debt, being paid by the customer.

What it looks like when you get this right is closer to the work we did with DreamWorks Animation. The challenge was not just modernizing tooling. It was redesigning the production workflows artists use every day, across systems managing hundreds of thousands of digital assets in highly complex pipelines. The real goal was to make powerful tools genuinely usable for the people doing the work. As DreamWorks CTO Bill Ballew put it, adopting new ways of working can be difficult because habits and processes are deeply ingrained. That distinction matters. The employee experience was treated as a design problem, not an adoption problem and not a training problem.

The Five Moments Most Organizations Are Skipping

In the deployments I have been close to, the AI stories that work and the ones that do not differ less in technology choice than in the steps that got skipped. Five moments keep showing up.

1. Choosing where AI shows up

Too many decisions about AI’s first touchpoints follow the technology, the vendor, or the loudest internal champion. Where does AI go gets answered before what experience problem are we solving. The better question is where is the human experience already fragile, and will AI make that better or just expose the cracks faster?

2. Defining how the experience should feel

Teams obsess over model performance and flows, but far fewer spend time on how the AI should feel — the tone in edge cases, the confidence it conveys, when it slows down instead of speeds up. And fewer still are honest about scope. A generic "customer service" AI that can field any question is a fundamentally different — and far harder — problem than one scoped to handle billing disputes for a specific product tier. The organizations that get this right start narrow and specific. The ones that struggle deploy broadly and optimistically, then discover the edges the hard way.

3. Designing for when AI is wrong — and for when it compounds. 

Most demos show AI at its best — quick, accurate, seamless. Real deployments are messier. A single-step bot that gets it wrong is usually recoverable. But as AI systems become more complex and agentic — chaining steps, calling tools, handing off between agents — errors multiply. A system running at 90% accuracy across three sequential steps doesn't deliver 90% accuracy. It delivers something closer to 60%. The organizations with real advantage design explicitly for failure at every link in the chain: clear handoffs, honest language about uncertainty, fast escalation paths (with sticky communications), and protections so employees aren't left owning machine mistakes they had no say in.

4. Understanding the employee story

Your employees already have an unofficial narrative about your AI. Is it the teammate that handles the tedious work, or the surveillance layer that adds oversight without adding control? Are they advocates, or are they quietly working around it? That story is an experience signal. It deserves to be treated as seriously as any customer metric.

5. Revisiting known compromises

Almost every deployment contains at least one decision everyone knows was a compromise. A pattern the model struggles with. A clumsy handoff. A policy edge case that gets handled badly. Those compromises tend to stay, not because they are unfixable, but because we will get to it later quietly becomes we have learned to live with it. As AI matures, those skipped moments do not stay small. They compound. What started as a workaround becomes the experience.

Experience Design Is Not The Cleanup. It Is The Strategy.

For years, experience design was treated as something you layered on after the technology decisions were made. Build the engine first, then bring in CX and UX to make it usable, clearer, more on brand.

In an AI first world, that sequence is backwards.

When AI is embedded in critical journeys, your experience decisions are operational decisions, risk decisions, and cultural decisions. They shape how your brand behaves, how your people experience their work, and how resilient your organization is when the system misfires.

That is also where a lot of organizations are getting stuck. Research from MIT points to a sharp divide between AI experimentation and actual business value, with only a small share of enterprise AI pilots delivering measurable P&L impact. The gap is not just about model performance. It is about integration, workflow redesign, and whether the surrounding experience was thoughtfully built in the first place. 

The competitive advantage in the agentic era will not come from infrastructure alone. The technology is converging. Every platform is moving toward agents. What does not converge is the quality of thinking behind what those agents should do, for whom, and what better actually means in terms customers and employees would recognize.

The companies that win will not be the ones who implemented the most tools the fastest. They will be the ones who treated experience design as strategy from the start.

Three Questions To Take Back To Your Team

At HumanX this year, I am facilitating a session for CX and EX leaders who already sense something important is being skipped. Not for people who need convincing that experience matters. They already know that. This is for people who need language, a framework, and a place to name what they are seeing before it gets harder to fix.

If you do one thing after reading this, make it this conversation.

Who actually owns the experience of our AI, not the technology, the experience? Name the actual people, not the functions. If no one can answer clearly, that is your first signal.

Where did we configure when we should have designed? Look for the places where defaults were accepted, flows, messages, thresholds, without asking what they mean for a real customer or employee in a real moment.

What is the one compromise we made that we need to go back and fix? Do not try to fix everything. Start with the compromise everyone knows about and no one has touched. Move it from we live with this to we are addressing this.

AI will keep getting more powerful. The question is whether your experiences will get more human, more intentional, and more trustworthy alongside it, or just faster in the direction they were already heading.

That is not a technology question. It is a design question.

AI is no longer just supporting experiences. It is participating in them.

Every response, recommendation, escalation path, and automated decision now shapes how customers interpret the brand and how employees experience the work behind it. In the agentic era, experience is not the layer that gets added after the system is built. It is the layer that determines whether the system creates clarity or confusion, trust or resistance, momentum or friction.

That is why this work matters more now, not less. As agents become more capable, the differentiator will not be who deployed them first. It will be who was clear about the role they should play in the experience, who designed for the human reality around them, and who understood that experience is not decoration around intelligence. It is the discipline that makes intelligence usable.