How your lack of training is making your Copilot adventure more expensive than not investing in AI

In my day-to-day work I help a lot of clients implement Microsoft 365, and get a grasp of their Information Architecture, both using just OOTB (Out Of The Box) SharePoint and Teams, but also loads and loads of custom extensibility, all with the purpose of doing more, in less time and I’ve spent almost a decade doing so, but these past few years, something has changed in the world of IA.

The age of AI

Now I’m sure none of you’ve missed it, but, back in November 2022 OpenAI launched ChatGPT, a revolution in the way we thought about computers, suddenly I could chat with my computer, and get very detailed, long, almost human-like responses, the potential was VERY promising.

And it didn’t take long for big tech to catch on, Google with Gemini, Microsoft with Copilot, Meta with Llama and so on, the promise was big! - AI with access to all the knowledge in your organization, what could go wrong?

Fast forward almost three years, it’s now late 2025, and everyone has access to some sort of AI, and we’ve all learned to spot AI generated text, if not by the sometimes a bit odd formatting, em dashes, or emoji usage, then by the just waaay too long/detailed answers, or the fact that your insurance advisor who you’re chatting with, and that is definitely a human being, happily provides you a pancake recipe if you ask for it.

So have we become more efficient?

Well, that’s a tough one, I know loads of people who feel more efficient, and as a matter of fact, I would dare to say, that Github Copilot, has probably made me 10-20% more efficient, and I know I’m not alone.

But perceived efficiency does not mean actual efficiency, if you’re shipping things off faster than before, but your colleagues are now spending more time having to correct your work, have we really become more efficient, or have we just gotten faster at producing more?

The dark side of AI

While AI helps us produce more, faster, something that I see more and more, is that people confuse more with better, combine that with a common misconception that an AI cannot “lie” about the information it’s grounded in, is costing organizations around the world tons of money!

The problem isn’t the AI, and it’s not the people, it’s a lack of understanding of how the technology works.

And I’m not saying that “Henry from accounting” or “Simone from engineering” needs to understand AI, and what an orchestrator is, or the difference between grounding, and training a model is, HECK, they don’t even need to know what an LLM is.

But they do need to understand that the chatbot is not human, and is not actually capable of reasoning, even if you press the reasoning button, YES it does provide a lot of value, if, AND ONLY IF, you make sure to review the outputs, and actually view them through a critical lens, understand what it says, don’t just copy-paste it.

I’ve seen an uptick in obviously AI generated things that are costing organizations a ton of money, from project requirements that include obscure things, to sales proposals that don’t match what your client is requesting, I’ve seen requirements ranging from “should work, and be translated to any language” for an organization of just a few Danish folks, to “Any extensibility in SharePoint must be built using PCF” (Power Apps component framework) - these are the kind of things that will throw people on the other end off, now I need to call you, and tell you that what you’re requesting isn’t feasible.

Heck I’ve even had clients send me a bit of code, telling me to “run this against our environment” - however upon inspection, said code would slowly delete every file in their tenant.

Now most of the above examples are code related, but the same thing goes for your organization’s communication, I recently got added to a Project Plan to help out with a marketing initiative, where the first task was to build out a SWOT report, and the second step was to identify the global target audience … this was for a small local shop, this was just because someone had found the Project Manager agent and thought “huh, this looks professional” and then forwarded a list of 100+ tasks to everyone in the team, that’s now wasting everyone’s time.

In this case, my time is billable, which becomes an expense because the employee didn’t bother reviewing what the agent had provided before just blindly forwarding it.

And that’s my biggest point in all of this AI Without human in the loop, will cost you, not just money, but also business

So what should we do?

ADOPTION, ADOPTION, ADOPTION, ADOPTION, ADOPTION, ADOPTION, ADOPTION, ADOPTION, ADOPTION, ADOPTION, ADOPTION, ADOPTION.

Adoption is the only solution, you really need to make sure that your end users understand the importance of being critical of what an AI outputs, and learn to not blame the AI, an AI is a text generator, it cannot be faulted, only the human that forwards its outputs can be blamed.

The excuse “But Copilot told me” is not an excuse for anything short of laziness - and if you’re technical, it’s not an excuse at all.

TL;DR

AI is just math and text - we all need to understand that human in the loop and agentification is where the actual value of AI is, without it, you just have a random text generator.

In essence, AI used without any guardrails, or consideration just improves the quantity of work you do, not the quality, I’m not saying AI can’t help us produce better quality, but to get it to do so takes work!

My key takeaways are:

• AI does not necessarily improve the quality, but rather the quantity.
• “The old saying “Garbage in, garbage out” still applies.
• AI with the proper training, and understanding can help improve quality, but it takes work.
• AI without propper training, grounding/agentification and human in the loop will just cost you money.

AI Can and will revolutionize and improve the way we work, but right now it can be more of a productivity drain than a productivity gain without the right guardrails, setup and user education.