I’ll admit something. Most weeks I open my feed and there’s another model, another feature, another post shouting that “everything has changed.” I read it, I nod, and then I go right back to working the same way I did last month. If that sounds familiar, you’re going to like this one.
I just came across a refreshingly honest post from an AI professional who writes about AI twice a week. Even this expert admits they can’t keep up. September alone brought 25 major releases. Claude beats ChatGPT. ChatGPT beats Claude. “X is dead.” Repeat every Tuesday.
The author’s sharpest point is a simple question: how much has your work actually changed? If the honest answer is “not much,” the problem isn’t that you’re missing the latest model. You just haven’t turned any of them into a real habit yet. So the creator laid out a five-step plan to fix that, and I think it’s one of the most practical starting points I’ve seen in a while.
Why chasing new models doesn’t work
New releases are fun to read about. But reading about AI and using AI are two very different things. Every hour you spend comparing benchmarks is an hour you’re not saving on your actual work.
The original poster flips the approach. Instead of starting with the tool, start with your job. Pick something you already do, make AI good at it, and lock in the result. That’s it. Here’s how the plan works, step by step.
✦ Step 1: Pick one task you do every week
The author suggests choosing a recurring task, such as:
- A client update
- A sales report
- A presentation
Why it works: You need something you can judge. You already know what a good result looks like, so you’ll spot bad output instantly. That makes testing fast and honest.
Then comes the clever part. Claude or ChatGPT? Don’t argue about it online. Try both on that same task and keep the one that needs less fixing. I love this because it replaces opinions with evidence from your own work. The “best” model is simply the one that saves you the most editing time.
✦ Step 2: Give it the right context
This is where most people fall short. AI can’t read your mind, so feed it what it needs to do the job well:
- Your brief
- Your spreadsheet
- An example you like
The expert also recommends connecting the apps where your work lives and asking the AI to read the relevant files before it starts. That one instruction alone can change the quality of the output.
Why it works: Generic input gets generic output. When the model sees your real data and a sample of what “good” looks like, it stops guessing.
A quick safety note from the post: check your privacy settings before uploading work, and use your company’s approved workspace for client information.
✦ Step 3: Give it a finish line
“Help me with my report” is too vague. The AI doesn’t know where to start or when it’s done. The creator shared a much stronger alternative.
Prompt of the Day:
Read the attached sales spreadsheet. Find the 3 biggest drops. For each: show the evidence, a possible cause, and one next action. Return a one-page report. Flag anything you can’t verify.
Look at what that prompt does. It names the task, the source, and the output format. It even builds in honesty by asking the AI to flag what it can’t verify. The author’s formula is easy to remember:
- Name the task: what exactly needs doing
- Name the sources: which files or data to use
- Name the output: length, format, structure
- Set approval points: tell it when it needs your sign-off
Why it works: A clear finish line turns a fuzzy conversation into a deliverable. You can reuse this structure for almost anything, from a weekly client recap (“Read last week’s notes, list 3 wins, 2 risks, and next steps in under 200 words”) to a meeting summary.
✦ Step 4: Make checking part of the task
Once you have a draft, don’t just trust it. The post’s author suggests asking the AI to compare the report against the spreadsheet and check:
- Dates
- Calculations
- Sources
Then review the result yourself. As the creator puts it, a confident answer still needs checking.
Why it works: AI models sound sure of themselves even when they’re wrong. Building a self-check into the workflow catches a lot of errors early, and your own final review catches the rest. I think this step is what separates people who trust AI at work from people who got burned once and gave up.
✦ Step 5: Save the process once it works
This is the step that turns a one-off win into a lasting upgrade. When the workflow works, the expert recommends turning your prompts into a reusable skill. Include:
- Your preferred format
- A good example
- The checks it should run
Next week, you just give it the new material. When you spot a mistake, improve the skill. Over time, it gets sharper and you do less.
Why it works: You stop rewriting the same instructions every week. The process lives in one place and keeps getting better.
That’s one recurring task you’ve made easier. Repeat it with the next task, and the gains stack up.
What I’m taking from this
Honestly, this post was a relief to read. The author ends with a line I felt in my bones: “It’s a battle. And we’re getting new models every Tuesday. I’m tired, boss.” Same here.
But the plan itself is the antidote to that fatigue. You don’t need to master every release. You need one task, good context, a clear finish line, a check, and a saved process. New models will keep arriving, and when they do, you can test them on a workflow you already trust instead of starting from zero.
My suggestion: block 30 minutes this week, pick your most boring recurring task, and run it through these five steps. You’ll know pretty quickly whether it sticks.
Check out the full LinkedIn post for the creator’s complete breakdown and their exact wording on each step.