Can AI solve all your problems? A practical look beyond the hype

We can’t ignore Artificial Intelligence in the IT world today. And from what we hear, it will solve all our problems. Automate our processes fully. Deliver better customer experiences. Let us have a three-day work week. Build better relationships with our children. I don’t know about you, but there is a lot to say for the Alex Hirsch way of looking at the world. 

These hypes are expected. When the internet came on in a big way at the turn of the millennium, we experienced a similar wave. We also felt a similar desire to embrace this new technology in everything we did. Notable successes from this period, chiefly things like internet shopping, have stood the test of time, for good reason. Whereas companies with arguably innovative approaches, including digital currency, video streaming and more, were consigned to the dustbin of history. The desire to solve all of life’s problems using something like the internet was not as successful as many people would have liked.

When AI is not the right answer

Today, history seems to be repeating itself. AI is touted as a band-aid for all our current woes, particularly from a business perspective. Every product or service seems to have “AI” bolted on in some way. You take a short taxi ride from San Francisco airport, and all you see are billboard after billboard advertising the latest AI company or offering. It seems that 25 years is a long time indeed for people to forget the folly of over-hyping things. 

People also seem to forget that certain business processes don’t align well with a non-deterministic solution, which is what AI essentially is. A great example from the UK is any process requiring Financial Conduct Authority (FCA) oversight. Strict rules ensure financial products aren’t mis-sold and that consumers’ rights are protected. If your AI agent forgets to communicate these details, the scrutiny and fines your company will face will almost certainly make you regret it. 

So, I think it’s time to take a step back. Breath. Exhale. And think about the right scenarios where we can start incorporating AI into our workflows to deliver practical value. Not just as something we can brag about with peers at our next industry conference. 

AI use cases that work

I see AI’s key strengths today as empowering established creative and technical professionals and delivering incremental gains. As I type this, I benefit from AI assistance via Grammarly. It can catch mistakes, guide my tone, and even rewrite portions of my text to deliver greater impact. All while helping me retain the essence of my “voice”. The benefits for me? I can get a much faster, polished version of this blog post, and you won’t get surprised by pesky typos or peculiarities in my text. 

From idea to prototype in a few steps 

Turning to a more technical example, with my developer hat now firmly fixed on. GitHub Copilot is making a huge impact across the technical community as a code-assistance and orchestration tool, letting us turn an idea into a fully working prototype in minutes.  

A recent, perhaps not so business-relevant example: I was trying to get to grips with the new code apps capability in the Power Platform, which lets you run any custom app directly in the platform. So, I asked: what is the one thing every business user wants to do in their core business applications? Play DOOM.  

With GitHub Copilot and a few iterative turns, I built a fully working application in less than two hours and deployed it successfully within a model-driven Power App. I didn’t need to write a single line of code. And if I did, I knew I would have code completion at my fingertips to help me write code faster, incrementally. These small “gains” may sound insignificant in isolation, but they add up over time to the point where you couldn’t imagine being without these tools. 

What’s in it for you? 

Introducing AI, or at least testing it, within your daily workflow can deliver surprising results. This is a key recommendation from Ethan Mollick in his rather excellent book, Co-Intelligence: The Definitive, Bestselling Guide to Living and Working with AI. He also posits another key idea I keep in mind: today’s version of whatever Large Language Model (LLM) you work with is the worst version you will ever use. The pace of progress is staggering, with new models released every fortnight. We all have a lot to look forward to.

So, when we’re trying to assess whether AI can address a business need or challenge, it’s important to consider a few things:.  

1. What is the degree of human oversight required in the process?  
2. How deterministic do your processes need to be?
3. Will long-term consequences outweigh the potential short-term gains if the solution is poorly implemented?
4. And crucially, does the math add up?

What I mean is: does the cost of token consumption and implementation clearly demonstrate a return on investment (ROI) and business value? If your processes require frequent, often detailed, human analysis, AI won’t help. If you need to be absolutely, 100% sure that a customer receives the same message each time, then AI again can’t help here. If your AI solution could cause embarrassing headlines, like selling your expensive product for a dollar or offering an unconventional way for customers to generate code themselves, then AI isn’t a good fit here. Cost also plays a huge role, particularly for many hosted LLMs on the market today. One innocent mistake, and your resulting token bill could be enough to cancel your Christmas party plans.

Conclusion

Understandably, many organisations want to innovate, especially when they worry competitors may be nipping them in the heels. Becoming irrelevant in today’s market of fickle, highly expectant customers is all too possible. But innovation should always be grounded in a sensible business argument, a clear benefit (either to employers or end customers), and with all costs fully rationalised, considered and planned out. A softly, softly approach in the maze of AI will deliver the best impact, in the short to medium term. 

If you want to learn more about practical ways we can use AI today or hear about the work we are doing with our clients in this area, reach out to me anytime. Even if it’s just to laugh over the craziest way you’ve seen AI shoehorned into a product or initiative elsewhere today.

FAQ

Does AI replace traditional automation approaches?

No, there is still a time and place for deterministic automation, and typically, combining traditional and AI-infused automation delivers the best outcome.

What is the best way to assess the current capabilities of AI?

Find good excuses to use it when you are given a business problem or challenge. Even if you don’t get the result you want, you can better understand the technology’s limits and more easily identify scenarios where it can deliver practical benefit.

Is AI just for “vibe coding”?

Although this is definitely one of its more effective use cases, AI can be used for a myriad of other tasks as well – including summarisation, process automation, image & video generation and more.

If my business process requires detailed auditing and accountability, is AI really suitable for it?

In certain scenarios, an AI-based automation can be more transparent than a manual process carried out by a human. With tools like Copilot Studio, each session can be logged individually for full traceability, and the platform supports bringing humans into the loop based on the scenarios we define.

Is it a truly a “one-size-fits-all” when it comes to working with LLM’s?

Each vendor’s LLMs are trained on different datasets and, therefore, are better optimised for certain workloads or requirements. For example, OpenAI’s Codex models are best used when generating code. Researching and aligning towards the best available model is therefore crucial to ensure you get the best results. Certain tools, like Microsoft Foundry, support capabilities like model router, to help you take the guesswork out of this process.