The 2025 AI Playbook

This week, we hosted a live session on The 2025 AI Playbook—a roadmap to unlock generative AI’s real potential for your team or organization. It wasn’t just another webinar about ChatGPT; we pulled back the curtain to show exactly how AI works, what it can (and can’t) do, and how you can stay ahead in 2025’s AI-driven world.

You can catch our supercut of the recording here:

Missed the live session? Here’s the quick recap:

We started with a question that resonates with every team leader:
"Surely AI can do more than draft emails or answer simple questions… right?"

The answer? Absolutely.

Generative AI has enormous untapped potential, from automating compliance checks to managing knowledge systems, but only if you know how to use it. It doesn’t work out of the box. That’s why we introduced The 2025 AI Playbook—a four-step approach to make AI work for your high-value use cases.

Here’s what we covered:

1. How Generative AI actually works.
   We simplified the “magic” behind neural networks so you can understand how to use them effectively. (Spoiler: They’re just input-output machines. Garbage in, garbage out.)

2. AI’s biggest limitations.
   Today’s AI has no memory, no awareness, and operates in a turn-based system. Yet with the right inputs and systems, it can transform your workflows. (Imagine a Michelin-star kitchen. Without recipes and skilled chefs, even the best tools won’t cook the meal.)

3. The Playbook: Identify, Investigate, Instruct, Introduce.

   • Identify: Find recurring, time-consuming “high value use cases” that are worth automating.

   • Investigate: Map out how these tasks are currently done.

   • Instruct AI: Engineer effective prompts and data flows.

   • Introduce: Pilot AI solutions, refine them, and track ROI.

4. A sneak peek into the future of AI: Skynet.
   From bigger context windows to memory and real-time, continuous AI, we explored what’s next—and what risks we should prepare for.

Want more?

Catch the full recording to see:

• Animated explanations of how AI works, from first principles

• A live demo of Auxee, the tool we built to overcome AI’s limitations.

• The honest truth about “agentic AI” - this year’s hype.

The questions we didn’t get to in the webinar—answered!

How would you go about developing a new or Future view of what a strategy should be for a specific department in 5-10 years, i.e. Future of Packaging strategy in 2035?

The question is a little vague, but I would develop a strategy on the basis of the capabilities that are now possible, and the impact of those.

I always compare the impact of Gen AI to the impact of Excel-like technology on ledgers and numerical work. It automated the mundane part of the work. Humans still need to operate Excel, improve their formulas etc. In the case of Gen AI, the Excel formulas are the prompts. While we don't have bean counters anymore, we do have more expert humans that operate Excel.

With each new AI model, while they do get smarter -- in their ability to "connect the dots" with the information given to them -- surprisingly little other progress has been made. They don't have a memory as such, and they don't learn from the work they do. These and other limitations of AI models suggest there is still a role for humans in acting as the experts who can engineer the AI to produce the desired output - much like there are expert Excel users.

Perhaps, though, in 10 years we will see more progress on AI's limitations than the last two years... in which case perhaps we'll all be on a beach somewhere, leaving the hard work to the AIs.

Are you finding that transfer learning & fine-tuning is typically needed for the majority of the repetitive-task (low-hanging fruit) business application AI projects? Or are well thought out templates sufficient?

In our experience, transfer learning and fine-tuning is not needed for most use cases - at least initially. It can be taken as a step for reducing costs, later, but the models continue to fall in price such that it usually is not worthwhile. Deepseek, for example, is probably now lower cost than most fine-tuned models.

Thanks for reading!

Know someone who needs an AI boost?

📨 Share this email with them and tell them to subscribe.

🗓️ Let us know if there are topics, pain-points or other things you’d like to see us cover.

– Dino Gane-Palmer
CEO, PreScouter

P.S. What’s one task you’d love to offload to AI? Hit reply—I’d love to hear your thoughts!