Hello fellow AI explorers and a warm welcome to new subscribers!

Let’s dive in:

#6: The AI Adoption Paradox & Using Task Decomposition

💡 One Idea

The AI Adoption Paradox

I read a lot of reports and surveys on AI these days. Well, I don't read them end-to-end; I upload them to ChatGPT and then talk to them (I'll share this research tactic in an upcoming issue of Inside Track).

When you're deep inside a bubble like AI, it's easy to forget how far ahead you are. Using reports like the Deloitte survey below helps gain perspective from outside that bubble.

Just being subscribed to this newsletter puts you ahead of 99% of people in the world when it comes to AI adoption.

My favorite thing to dig for in reports is contradictions. Contradictions signal that something has shifted while other things haven't adjusted. Many great business opportunities have historically been found at these intersections.

Today I'm sharing the contradiction in "Deloitte's State of Generative AI in the Enterprise" report from March 2025 for the Nordic region.

(In a later issue we'll explore interesting regional differences that may signal specific opportunities.)

Check these two graphs:

(Source: Deloitte's State of Generative AI in the Enterprise, March 2025)

• Leaders expect generative AI to transform both their organizations (65%) and their industry (80%) within 3 years.

• Executive and board interest in AI have dropped quarter over quarter. -27% for the former, -51%(!) for the latter.

That's a clear contradiction. It's like saying: "This will rewrite the rules of the game soon... but we're taking a backseat for now."

Why this contradiction? My guess is a combination of:

• Expecting people "in the org to figure it out" (below the C-suite)

• Initial excitement followed by mean reversion to old ways

• Uncertainty about how to approach adoption

In issue #2 we discussed adoption challenges for established companies:

• Path Dependence

• Talent Gap

• The Time Lock Trap

• Principal-agent problem

This creates a time-based goldilocks zone:

Whether you're a founder, freelancer, or work inside a large company, leveraging this goldilocks zone is a huge opportunity.

If you're building your own thing, you'll gain competitive advantages. For example:

Imagine running a consulting practice. You adopt AI for proposals and back-office functions, shifting your billable/non-billable hour ratio favorably. You leverage this bandwidth through increased marketing and slightly lower hourly rates (because your unit costs improved). Lower cost, higher revenue. Pulling ahead.

In a corporate environment, there's going to be a moment when everyone wakes up. Executives will scramble to find AI talent. And there you are.

Here's the curriculum I recommend:

• Explore AI's higher-level impact beyond workflow automations. The commoditization of knowledge will shift entire business models.

• Learn how LLMs work, at least at a basic level.

• Experiment with AI for everything — from family trip planning to research and financial analysis. Find what works and make it stick.

(On the last point: My two most used apps are ChatGPT and ManusAI. Not because AI does everything for me, but because I have so much context there that they become my natural places to ideate, analyze ad reports, set reminders, etc. I even have a separate project with all the context for training our puppy, Louie!)

Coincidentally, these are exactly the things we'll explore together over the coming months.

Share Inside Track with a friend. It helps them get ahead, and it helps me grow the newsletter to bring even more value to you.

🛠️ One Tactic

Better Outputs with Task Decomposition

ChatGPT and similar assistants function best as collaborators, not autonomous workers. How you structure this interaction directly affects your results.

❌ Simple Delegation

The "Do this for me" approach typically produces mediocre outputs. One-shot delegation attempts to outsource complex tasks with minimal direction.

Example: "Create a marketing report to increase our sales" → results in generic, surface-level content.

Without sufficient context and feedback, AI tools cannot deliver sophisticated work through simple delegation.

This is what I call “lazy prompting”. A lot of people do this, get low-quality outputs, and mistakingly assume these AI assistants are incapable.

That’s…wrong.

✅ Task Decomposition

A more effective approach follows this structured workflow:

1. Task Overview – Provide specific context about your objective*

2. Plan – Have the AI propose a breakdown of the task into components

3. Feedback/Alignment – Review the plan, refine it, and confirm next steps

4. Task Execution – Address individual components (Task 1, Task 2, etc.) sequentially

5. Final Output – Direct the AI to integrate all components into a cohesive deliverable

This approach combines your expertise and direction with the AI's processing capabilities.

The Quality Difference

Task decomposition requires more engagement from you but significantly improves output quality. You guide the process while using AI to accelerate work.

Instead of expecting the AI to understand your full context from a single prompt, you develop the solution through defined, manageable steps.

• Tools like ChatGPT's memory feature allow you to establish context once for future sessions, reducing repetitive explanations. I'll cover advanced memory-optimization techniques in an upcoming newsletter.

That’s it for today. Thanks!

– Martin