Breaking the 3% Barrier: Using LLMs Strategically
Initial ideas on how to get more done - and “go home earlier”

Over the last one and a half years, we have witnessed tremendous change with the launch of Large Language Models (or LLMs) like ChatGPT. Trying ChatGPT 3.5 for the first time in spring 2023 left me awestruck. I had two simple reactions: 'How did they do this?' followed quickly by 'How will this change our lives?'
Since then, we've seen a steady stream of developments: new LLMs like Google Gemini and Anthropic's Claude, regular updates to ChatGPT, and practical tools like the AI search engine Perplexity. AI capabilities are now built into Apple and Google smartphones and software. The pace of technical progress has been significant enough that even those following closely might struggle to keep track of what's available.
Yet, as impressive as these technological advances are, I find myself increasingly focused on a “simpler”, more personal challenge:
How can we use these AI tools now for our own benefits - to get more done and “go home earlier”? What are the practical steps to make these AI assistants work for us, at work or in side projects?
That’s what I’d like to explore over the next weeks - sharing my experience and some practical guidance on how to leverage ChatGPT, Gemini, and Claude. These three LLMs have the potential to transform your daily work or side projects. With them, you can break down problems, learn new content or skills faster, and sharpen your writing and thinking. Your costs in terms of time, energy, and money are minimal.
Probably for these reasons, adoption of LLMs is rather high. Almost half of U.S. working age adults have used them based on a recent study1. AI may be the technology with the fastest adoption curve yet, clearly demonstrating that LLMs are not just a passing trend.
However, there is an interesting gap between what LLMs can do - and what we have done with them so far. As I shared elsewhere2, the actual productivity benefits at work measured so far are rather small, often in the range of only 3-5% or less. If you do the simple maths, you’ll see that we effectively gain only 10-20 minutes per day. This is not nothing, but it’s far from life-changing. Also, while overall adoption rates for LLMs are high, most users find work for their LLMs only once per week or even less3. It seems that LLMs can be potentially life changing, but we haven't yet found ways to fully integrate them into our daily (work) lives, and get sufficient value from them.
Looking at how people work with LLMs, and drawing from my own experience using them, I notice a few recurring obstacles or frustrations that often come up. Let me share these to start a conversation:
We face strong organizational resistance. Organizations, particularly large corporations, have established processes and tools that resist new technologies like AI - and not all data may be available readily4. Regulatory requirements, IT policies, and data protection guidelines create additional barriers. While startups might adapt more quickly, implementing LLMs in any business environment requires significant process changes.
We don’t really know what an LLM can help us with. LLMs lack clear manuals or workflows, making it difficult to understand their potential applications. This is partly because the developers of the big LLMs focus purely on building the models - and leave building specific use cases to other startups, or us. At the same time, we're still developing “mental models” for what LLM can help us with. Consider established productivity apps for to-do lists or notes. Many of us find it much easier and more intuitive to use them because we have a clear mental model of what a to-do list or notebook does, and how it works to help us. We don’t ask ourselves - why would I write down my to-dos, why do I take notes? We are still building these mental models for LLMs, which is made harder given that these tools are so versatile and the potential tasks are nearly endless.
We use LLMs mainly for tasks that are not creating enough value. We may all have an intuitive understanding of what "value" means in our work life or for side projects: e.g., launching products that customers want, streamlining team workflows that cause delays, or fixing the root cause of customer complaints. Yet when it comes to using LLMs, we often focus on "low hanging fruit" tasks like writing emails faster or summarizing documents. While helpful, these tasks only save a few minutes per day - time savings that rarely translate into meaningful impact.
We are uncertain how to collaborate effectively with an LLM. LLMs don't come with a specific workflow that guides us. In practice, we therefore often find ourselves giving them vague instructions in hopes they'll figure out what we want. Or we simply use them like search engines - which they are not. The more effective approach is to treat LLMs like experienced thought partners who need context, clear direction, and interactive feedback. Moreover, LLMs can often miss important nuances or make outright errors, making it crucial to carefully verify their output.
We're not exactly sure what we need help with. This might be the most significant hurdle we face to making better use of LLMs. Whether we use LLMs or not, we often operate like on autopilot. We react impulsively to emails or messages, shoot from the hip with quick-fix solutions, and rush through tasks without much reflection. While that's somewhat inevitable in any fast-paced environment, there's huge value in occasionally hitting pause and trying to articulate (in your mind or on paper) your own thinking: Why are you doing what you're doing? What's your goal? How can you do it better? Once you gain clarity on these points, you'll have a much better idea of how to use your LLM. And you can then brief it more effectively, perhaps saying something like, "Hey LLM, I've been assigned this complex task, and I'm not entirely sure how to tackle it. Can we jointly devise the best approach?" Using an LLM to its full potential often requires us to resist the urge to dive in headfirst. Instead, we need to pause, reflect on the problem we're trying to solve, define the desired outcome, and then strategically decide how and where to deploy the LLM in our problem-solving process.
These ideas may not be the final word on what keeps us from doing more with LLMs - but they are a good starting point for developing improvements. As I always preach: the solution is contained in the problem. I will probably explore this more thoroughly in the next weeks, but here are a few initial ideas how to make better use of LLMs recognizing these challenges:
We face strong organizational resistance. Instead of exhausting yourself trying to reform company-wide processes, concentrate on enhancing your personal workflows or collaborating within your immediate team. Remember that LLMs excel at analyzing and improving processes - simply explain your situation and challenges to them.
We don’t really know what an LLM can help us with. Build your personal library of proven use cases and effective prompts, rather than starting from scratch each time. Learn to frame your requests effectively - often it's more productive to ask the LLM to guide your problem-solving process, such as suggesting methodologies, rather than seeking direct solutions.
We use LLMs mainly for tasks that are not creating enough value. Define what constitutes real value in your role or project, then focus your LLM usage where it can generate maximum impact. The greatest returns typically come from implementing structural changes, launching innovations, or supporting major decisions.
We are uncertain how to collaborate effectively with an LLM. Think of LLMs as thought partners - engage them in idea exchange and methodology discussions rather than just seeking answers. Establish clear roles and expectations, e.g., the LLM acting as advisor giving recommendations, vs. as an analyst who looks for patterns or data that sticks out.
We're not exactly sure what we need help with. This may seem basic but is crucial: Before diving into work, take time to articulate your thoughts clearly. Consider your broader objectives: What are you trying to achieve? What serves your goals? What generates real value? And then throw your LLM at the problem.
There is one overarching theme across these solutions: A more strategic approach to LLM work yields better results. In this context, strategic means following a deliberate plan, with a long-term goal in mind. In practice, this translates to: pausing before engaging with LLMs, developing a clear plan, implementing it methodically, adjusting course together with the LLM as needed, and following through to completion.
Making LLMs work for you strategically
To bring us full circle: LLMs can help us get more done, and go home earlier. Not just by saving us minutes of busywork, but by helping us tackle our most valuable work. The key is approaching them more deliberately and strategically.
Over the next weeks, I'll share more thoughts on how to put this more strategic work with LLMs into action. Sure, there are plenty of how-to-LLM guides already. However, most are rather technical and focus mainly on explaining which "buttons to press". I hope I can add value by taking a step back and looking at why to use LLMs, in addition to explaining how to use them.
If you know someone who could benefit from this post, please consider sharing it with them.
Also, as always, I'm interested in your thoughts: Feel free to message me. Or, if you prefer, you can share your feedback anonymously via this form.
All opinions are my own. Please be mindful of your company's rules for AI tools and use good judgment when dealing with sensitive data.