Key Takeaways
- Prompting is not the main bottleneck in working with AI — the real challenge is learning to lead and manage AI systems effectively.
- AI should be treated like a new intelligent colleague: you need to brief it, train it, give feedback, and choose the right system for the task.
- Most people get poor AI results because they give bad commands to the wrong AI system and skip configuration steps like custom instructions.
- Providing feedback — such as thumbs up/down or written corrections — is essential to improving AI output over time.
- Tools like Power Automate can help address labor shortages by automating repetitive tasks, yet too few companies use them.
- Microsoft Copilot results can be significantly improved by using the Research Agent or configuring Copilot Studio with specific data sources.
- Microsoft Purview can be used as a compliance engine to configure which data AI is allowed to access.
- For organizations not on Microsoft 365, alternative solutions exist that support data pseudonymization and anonymization for AI use.
- Enthusiasm and practical implementation matter as much as technology — the goal is to make participants eager for digitalization, not afraid of it.
- Thorsten Jekel has been connecting digitalization, sales, and productivity since 1988 and is available as a keynote speaker or workshop leader.
Introduction
Nice to be back for another episode of TJ's Technology Tuesday. This episode features an excerpt from a keynote I recently gave on the topic of artificial intelligence. I hope you enjoy it.
Management of AI systems
Let's have a look. So everybody says, Prompting is the bottleneck. I say, no. Prompting is not the bottleneck. Because when you look, for example, into ChatGPT 5, it already optimizes your prompt. And you can use AI number one to optimize your prompts for AI number two. No problem. The bottleneck is leading the system. Because when I look into AI, augmented intelligence, it's like a new intelligent colleague. And when you hire somebody, what do you have to do? You have to brief him or her, you have to skill him or her, you have to train, you have to get feedback. And now we see that people are just giving a bad command into the wrong AI system and wondering why they're not getting the right results. And then they say, This is crap. Now we said, Okay, how was your prompt? Which system did you use? Did you enter custom instructions, which you can do in every system? Did you give a thumbs up or a thumbs down to your result? Did you give a feedback in terms of this was wrong, that was right? So now I assume all of you or a lot of you also see a shortage of labor.
Applications in companies
So if I see too few companies that are looking into Power Automate, for example. And in the workshop, I will show you some possibilities, and I'm very interested in those of you. Anybody who's already using Power Automate in his or her company? So I see some hands. So if some of you and I also see the tech pros, so that's cool. By the way, applause for the tech crew. They're doing a great job. So super. Applause. So also when I see, okay, yeah, Copilot is giving me bad results. Yeah. Have you looked into the latest Research Agent version? Have you configured a Copilot Studio where you can exactly say which sources it shall use? And when you talk about compliance, have you already used Purview as the compliance engine where you can configure it? And if you don't have Microsoft 365, there are solutions in the market, like a Spanish company, for example, working on European processing plans, where you can pseudonymize data, where you can anonymize data, which you're using for artificial intelligence.
Conclusion - increasing enthusiasm for AI together
Yes, I hope you were able to take away one or two ideas from this excerpt of my presentation, especially with regard to implementation. If you are looking for an informative, inspiring, moving speaker for your next event who will enthuse and inspire your team, your sales partners and your service partners and, above all, get them to put things into practice, then I would be delighted to hear from you. I am also happy to speak as a keynote speaker, workshop leader or in other formats for and with your customers.
Of course, I don't just talk to your customers for an hour when I'm giving a presentation, but I'm at the events all day so that your participants say afterwards: "Wow, that was a great service." That has been my idea since I started at Nixdorf in 1988, to bring together the topics of digitalization, sales and productivity, but also to do it in a fun and motivating way, so that new technology doesn't create fear, but that the participants have an appetite for digitalization and AI.
I would be delighted if I could enrich your next event accordingly.
Yours, Thorsten Jekel.
Frequently Asked Questions
Is prompting the main bottleneck when working with AI?
Prompting is not the main bottleneck when working with AI. Modern systems like ChatGPT 5 already optimize prompts automatically, and you can even use one AI to optimize prompts for another. The real bottleneck is learning to lead and manage the AI system properly.
How should you think about AI in the context of a business team?
AI should be thought of as a new intelligent colleague joining your team. Just like onboarding a new employee, you need to brief it, skill it, train it, and provide ongoing feedback to get the best results.
Why do so many people get poor results from AI tools?
Many people get poor AI results because they give vague or bad commands to the wrong AI system and do not take the time to configure it properly. Skipping steps like entering custom instructions or providing feedback on outputs leads to disappointing results.
What role does feedback play in improving AI output?
Feedback is essential for improving AI output over time. Giving a thumbs up or thumbs down, or explicitly telling the system what was right or wrong, helps the AI learn and deliver better results in future interactions.
How can Power Automate help companies dealing with a labor shortage?
Power Automate can help companies address labor shortages by automating repetitive and time-consuming tasks. Despite this potential, too few companies are currently exploring or using Power Automate to its full advantage.
How can organizations improve results from Microsoft Copilot?
Organizations can improve Microsoft Copilot results by exploring the latest Research Agent version or by setting up Copilot Studio, which allows you to specify exactly which data sources the AI should draw from. Proper configuration makes a significant difference in output quality.
What is Microsoft Purview and how does it relate to AI compliance?
Microsoft Purview is a compliance engine that allows organizations to configure which data sources their AI tools are permitted to access. It helps ensure that AI usage within a company meets data governance and compliance requirements.
Are there AI compliance solutions for companies not using Microsoft 365?
Yes, there are solutions on the market for companies not using Microsoft 365. One example mentioned is a Spanish company offering European data processing plans that include capabilities for pseudonymizing and anonymizing data used in AI systems.
What is Thorsten Jekel's philosophy when presenting AI topics at events?
Thorsten Jekel's philosophy is to make new technology approachable and motivating rather than intimidating. His goal is to inspire participants to take action and embrace digitalization and AI with enthusiasm, not fear.
In what formats is Thorsten Jekel available for company events?
Thorsten Jekel is available as a keynote speaker, workshop leader, or in other formats for company events. He also stays present throughout the entire event — not just during his presentation slot — to maximize value for participants.
Tools & Resources Mentioned
- ChatGPT 5 – AI system that automatically optimizes user prompts
- Microsoft Power Automate – Workflow automation tool for reducing manual tasks and addressing labor shortages
- Microsoft Copilot & Copilot Studio – AI assistant for Microsoft 365 with configurable data sources via Copilot Studio
- Microsoft Purview – Compliance engine for configuring data access rules for AI tools
