How to get more out of your LLM

Most people use AI in a very simple way: they type something in, get something back, and move on. That already works pretty well for a lot of things. But it is also where most of the value gets left behind.

The issue is not the tool. It is how we use it. Instead of asking “what can this model do?”, a more useful question is: “where in my work could this actually save me time?”

This post looks at that through a simple lens: start with a problem, understand what the tool can actually do, and then build a habit of improving from there. Not prompt tricks or complex setups, three simple habits that help you get more value from the AI tools you are already using.

This is not for the people setting up 14 workflows and a small digital monarchy of sub-agents. This is for non-technical people, the ones opening ChatGPT, typing something in, and getting a response.

1. Start with annoyances, not features

The best starting point is not a feature list, it is the small, repetitive task that keeps slowing down your day.

Most people do not care whether something is called an app, connector, integration, project, workspace, or agent. They care about annoyances. Things like:

• I waste time digging through files
• I keep rewriting the same updates
• I have to move information between tools manually
• I lose context halfway through longer pieces of work
• I still do meeting prep like a medieval clerk

That is a much better starting point than reading a feature list and pretending to feel something when you see the phrase “cross-surface context orchestration.”

Start with friction. Then work backwards. Pick one thing you do every week that is repetitive, annoying, and just complicated enough to waste more time than it should.

Then test one change. Not five. One. The point is not to build an AI command center in your basement. The point is to remove one redundant manual task and then repeat when useful.

2. Check what the tool can actually do

A lot of people decide AI cannot help before they have checked what it can actually do.

A lot of people still treat these tools like smarter autocomplete, something that responds, but does not really participate. That model is outdated. So before anything else, open the settings. Actually look at the product you are using. Check what is enabled, what can be connected, and what your plan includes, because these tools are increasingly designed to work with you, not just answer you. There is a meaningful difference between:

• Summarising a document
• Searching across your files
• Keeping context across a longer project
• Creating or updating something on your behalf

Those are not minor feature differences. They are different categories of usefulness.

A concrete example: in Claude, skills let you package a brief, past decks, a style guide, and more, into a reusable instruction set, and every response you get after that already knows your context. No re-explaining every time. That is not something you stumble across in the chat window. You set it up through your account settings or the skills page. For a deeper look at how this works in practice, see Kata’s post on skills.

This is where a lot of people leave value on the table. Not because the product is incapable, but because they never checked what was turned on or what was connected. Before deciding “AI cannot do this,” check what it can access, what it can remember, and what it can do.

3. Use the model to audit how you work

The highest-leverage habit is not asking the model to do the task, it is asking whether you are doing the task in the most manual way possible.

If you keep doing something by hand, ask whether a simpler workflow now exists. If you mainly use AI for writing, admin, or research, ask which features you are underusing. If you have not checked in for a while, ask it to look at the latest official updates from your provider and tell you what actually matters for your workflow, not what looked impressive in a launch demo.

This only gets good when you give real context. The more specific you are, the more useful the audit becomes:

Instead of this	Try this
“I work in marketing.”	“Every week I summarise five customer calls, draft one internal update, rewrite the same follow-up email three times, and dig through two folders to find the latest deck.”

That second version is where the audit becomes useful instead of sounding like workplace astrology. Once you have that description, combine it directly with an audit prompt:

RECURRING AUDIT PROMPT

I do this task regularly: [describe it in one or two sentences, the more specific the better].
Here is how I currently use AI for it: [your specific description].
Based on the latest official features from [ChatGPT / Claude / Gemini], what am I still doing
manually that I no longer need to do? Tell me what to try first, what depends on plan or
permissions, and what is still more hype than help.

Run this every few months. Most feature updates are only interesting in theory. A feature becomes useful when it maps to work you actually do.

That is the whole idea

Find an annoyance. Check what is possible. Audit regularly. Not by becoming AI hobbyists, but by becoming slightly less busy for very boring reasons.

At Zypp, this is exactly how we work with teams, not “here are all the AI features,” but “here are the three that will actually make your Tuesday less annoying.” If that sounds useful, we would love to chat.