Why AI Isn't Meeting Expectations

Once you’ve earned enough gray hairs—or in my case, lost enough—you start to recognize patterns repeating. Let me explain.

My first job in tech was as a tech support representative in 1995 for one of the first ISPs. I helped people get online. Once they got online, I spent a lot of time helping them get out of trouble. 🤦‍♂️

Sure, they became more productive with access to email communications. But in the early days, there wasn’t much business going on online unless you count monkey business.

Websites were simple billboards, and every advertisement had a www.companyname.com on their TV ads. You didn’t click through, and most of us had to remember the URL and type it into a slow desktop computer.

AI is in a similar spot. Everyone’s talking about generative technology, but the promise of this new technology is still largely unmet. 

This week we’ll look at some recent data on how AI is addressing the needs of business.

☕️ AI Tangle - Meta's Llama 4 Launch Gets Off To a Rocky Start

🔮 AI Lesson - How to Generate and Edit Visuals in ChatGPT

🎯 The AI Marketing Advantage - How Top Brands Pair AI With Human Strategy

💡 AI CIO - OWASP Just Sent a Major Signal

Why AI Isn't Meeting Expectations

Why most AI projects stall—and what it takes to move from pilots to productivity

AI is more visible—and more hyped—than at any point in history. We’ve entered an AI Spring: a surge in experimentation, investment, and executive focus.

But there’s a growing disconnect:

• AI is on every boardroom agenda.

• Generative AI is the fastest-adopted technology ever.

• Yet, most companies aren’t realizing business value.

The problem is that while everyone is interested in AI, success is elusive for many of them. AI is back at the top of the enterprise agenda. There’s plenty of buzz—but little business impact. Boards expect transformation. Most companies are still trying to make pilot projects stick.

Generative AI has become the fastest-adopted technology in history, but adoption hasn’t translated to outcomes.

• According to McKinsey, only 13% of companies report meaningful value from GenAI at scale.

• Accenture found that most of the 2,000+ projects they analyzed are stuck in pilot mode.

• Deloitte reports that governance, security, and compliance risks are rising, not falling.

AI isn’t failing. It’s being deployed like middleware—bolted on, not built in.

Most companies have a tooling strategy. What they lack is an operating model for AI. The problem isn’t the technology. It’s the approach.

Companies are deploying AI into systems that weren’t designed to use it. They’re measuring pilot counts instead of business impact. They expect employees to adopt AI tools without equipping them to co-create with them.

That’s why I focus on helping businesses learn how to use AI effectively. I think it’s the missing piece for most companies.

So, today, AI adoption is slow, trust is low, and productivity gains are sparse.

But they don’t have to be. Let’s examine failures before we consider ways to succeed.

[Bonus Read: Just published this week, Stanford University’s Human-Centered Artificial Intelligence, The 2025 AI Index Report. It’s 496 pages, and I am still processing, but it’s full of interesting tidbits.]

The Five Failure Patterns

Tech-first, People-last
Employers say they’ve implemented AI to support employees’ initial research for tasks and projects (62%), help employees manage their workflow (58%), and analyze data (55%). However, nearly two-thirds of employees (63%) are primarily leveraging AI to double-check their work. That disconnect leads to skepticism, low usage, and reversion to manual processes.

Garbage In, Garbage Out
According to MIT Technology Review Insights, the top three reasons AI fails to scale:

• Siloed systems

• Weak governance

• Poor-quality inputs

Even the best models are useless when fed disorganized or incomplete data.

3. AI with No Sponsor
There’s a reason I write a strategy newsletter for management that includes examples of prompts and tools. Leaders need to show a minimum level of competence in using AI. If the CEO isn’t visibly backing AI, it will stall. Without top-down commitment, strategic alignment breaks down. While I don’t expect every CEO to become an AI expert, I think it’s important that they lead by example, and I try to provide some high-impact tips. If your manager asks you to use AI and doesn’t do it themselves. I think youthe likelihood of your organization being successful is much lower. Check out how Shopify CEO Tobi Lutke is approaching AI adoption. 

4. Rigid platforms, locked-in vendors
GenAI thrives on iteration, not static deployment. Most companies are stuck with legacy stacks and quarterly releases. The few getting it right use agentic, composable systems—orchestrated with frameworks like LangChain, BeeAI, or CrewAI.)

Composable platforms using interactive protocols like MCP are the way to go. Early on, WordPerfect and Yahoo! gave way to latecomers like Microsoft Office and Google, so don’t declare the winners too early. 

5. Infrastructure that can’t keep up
Generative models aren’t just compute-hungry; they’re infrastructure-intensive. The Internet Energy Agency expects global data center energy consumption to double by 2026, potentially surpassing Japan’s total electricity usage. We are also facing a global shortage of GPUs in the supply chain. That’s no longer just a cost issue. It’s a constraint on scale.

What Winning Companies Are Doing Differently with AI

They’re treating AI as a business capability, not a tech feature.

Accenture rebuilt its lead-to-cash process using machine learning to improve reconciliation and reduce manual intervention. According to SAP’s case study, this allowed Accenture to scale operations with fewer resources and faster cash flow.
AI was embedded in the process, not layered on top.

Klarna deployed an AI-powered chatbot that handled 2.3 million customer conversations in 23 countries in one month. It performed the work of 700 agents. But more importantly, it resolved full cases—not just handed them off.

The Five Strategic Shifts That Work

1. Lead with value, not novelty
   AI isn’t for experimentation—it’s for transformation. Accenture restructured an entire revenue workflow. Klarna redefined service delivery. Neither started with “use cases.” They started with outcomes.
   

2. Build for agents, not monoliths
   Static systems can't adapt. Winning companies are designing composable infrastructures with plug-and-play models and orchestrators. AI agents are largely like new employees, they require training, especially for complex tasks, but investing today in agents that will improve as the reasoning capabilities of LLMs increase in ability will yield results in the long term.

3. Make talent the multiplier
   Tools alone don’t scale. The organizations succeeding are:
   
   - Creating role-specific training
   - Embedding AI into daily work
   - Hiring and organizing by skills, not titles

I have been working on creating training classes for my clients that start for knowledge workers but then are geared towards sales, marketing, and DevOps engineers. I think we’ll see vocational training for many different occupations in the future.

4. Operationalize responsible AI
   Governance doesn’t start in compliance—it starts in design. A KPMG case study showed how the banking industry is adopting AI for use cases, but some applications are still facing significant hurdles. In banking, AI introduces unique risks that can undermine trust, meaning proactive risk management is critical from the outset.

5. Treat GenAI as a product, not a project
   Generative AI isn’t a launch-and-forget initiative. It requires iteration, performance monitoring, and user feedback. Today’s AI services will change rapidly. As new products come online, you need to make sure that your strategy evolves with them.

The Strategic Gap Is the Real Problem

My take is that AI isn't failing; it’s being underutilized. Organizations that treat GenAI as a plug-in will never reach escape velocity. Those that treat it as a redesign lever—restructuring processes, upskilling teams, re-platforming data—will compound productivity gains.

The fix isn’t more tools. It’s more transformation.

• Fyxer - I am currently piloting Fyxer for my email. It seems to have many advantages, including the ability to draft email responses. I can speed up my replies by starting with their drafts based on data from my previous emails.

• Camunda—One of my partners is working with a company providing a new AI Agent orchestration framework. What interested me is that their AI agents can autonomously manage tasks within a defined scope, executing them in any order, repeating steps as needed, or skipping them entirely based on real-time conditions.

• Gemini 2.5 - I have always thought that Google would catch up and perhaps surpass OpenAI in many of its capabilities someday. With their latest model, it’s more of a horse race than it has been in a while. I find that this release is very capable, but it also has integration with Google Workspaces. I think this is a huge benefit for Google.

Use ChatGPT to Create PowerPoint Presentations

One of the most difficult tasks most of us have is creating impactful presentations, but with the new image capabilities of ChatGPT, I decided to give things a second chance. Until recently, I used Gamma or Beautiful for AI-powered presentations. They worked, but they still required a lot of back-and-forth.

However, with ChatGPT, I think we are getting closer than ever to generating high-quality presentations. We’re not yet at the point of perfect single-prompt presentations, but I do think you could generate a top-quality draft that you can tweak with the following prompt. I created my own styling, but you can choose your own for your use.

## Objective: Generate a Conceptual PowerPoint Presentation

## Role 
You are a presentation designer who is an expert at conveying ideas using PowerPoint. 

## Instructions for ChatGPT
 Create a 10-slide PowerPoint deck using the following style and content rules:

---

## Design Standards
- **Font**: DM Sans for all text
- **Headings**: Black (#000000)
- **Accent Color**: #f44800 for visual elements and highlights

---

## Slide Layouts
1. **Title Slide**
2. **Agenda**
3. **Problem Statement or Context**
4. **Solution Overview or Key Idea**
5. **Supporting Data or Market Trends**
6. **Framework or Process Diagram**
7. **Use Case or Application Example**
8. **Benefits or Impact**
9. **Call to Action or Summary**
10. **Contact or Closing Thoughts**

---

## Image Guidance
For slides containing images, include a description of either:
- A **conceptual design** (e.g., metaphorical or abstract representations)
- A **photo-illustrative design** (e.g., real-world imagery aligned to content)

---

## Content Guidance
- Provide **3–5 bullet points per slide**
- Use clear, business-oriented language
- Avoid jargon unless commonly known
- Ensure logical progression from one slide to the next

---

## User Input Required Before Generation
- What is the **topic** of the presentation?
- Who is the **intended audience** (e.g., C-level, technical team, general business)?
- Are there any specific **messages, goals, or products** to highlight?

---

## Output Format

- Provide a **slide-by-slide outline**
- Specify **layout type** (e.g., title + bullets, image left + text right)
- Include **image description** where applicable
- Maintain consistent **formatting and visual hierarchy**

Now, this is where you need to add your input. I think there’s still a bit of back-and-forth required to get this perfect, but I think I’d potentially use this prompt with a copy of your own slide decks and themes to help guide the output.

For images, ChatGPT has received a huge upgrade. It still requires some work to generate the perfect presentation, but it’s a lot better.

You might try to generate the text and then develop the images separately. My example above is a framework for you. I used this to create an outline and then went back and added the excellent-quality images. But that involved creating each image and then cutting and pasting it into a presentation.

It made me cancel my iStock subscription. Here’s an example of a title slide generated for that presentation. However, unlike a PowerPoint slide, it was an image—not a mix of text and image. But the layout and quality are very good.

I appreciate your support.