Asama Akhtar

I just reduced Grok Bot's token usage by 40.6% on the same task by changing how I instructed it. Instead of telling it what to do, I taught it how to do it. I wanted to experiment with two different ways of getting Grok Bot to complete the exact same job. The task was simple: Go online and find an iPhone Duo in Night Sky with 256GB storage. Check whether it's available for delivery to Sydney, postcode 2000, and report the delivery date or whether it's out of stock. TEST 1 - TELL GROK BOT WHAT TO DO I gave Grok Bot the instructions and let it go online, navigate the website, find the product and check delivery availability. Total usage: 1,316,200 tokens. TEST 2 - TEACH GROK BOT HOW TO DO IT This time, I took a different approach. I used Grok Bot's computer and demonstrated the process myself while recording the screen. I showed it where to navigate, what to click, how to find the product and where to check the delivery information. Then I had Grok Bot watch the recording and create a reusable skill based on what I'd demonstrated. Once the skill was ready, I gave Grok Bot the same task again, this time using the skill. Total usage for that execution: 782,200 tokens. THE RESULT 1 - Instructions only: 1,316,200 tokens. 2 - Using the taught skill: 782,200 tokens. That's 534,000 fewer tokens, a 40.6% reduction on the skill-based execution. This was one experiment, not a guarantee that every skill will deliver the same saving. Creating the skill also involved upfront work and usage. But it demonstrates something worth investigating. When you simply tell Grok Bot what to do, it may need to figure out how to navigate and complete the process. When you demonstrate the process and turn it into a skill, you're giving it a reusable method to follow. HERE'S HOW TO EXPERIMENT WITH THIS YOURSELF 1 - Choose a repeatable task Grok Bot normally completes online. 2 - Run it using ordinary instructions and record the token usage. 3 - Open Grok Bot's computer and record yourself demonstrating the process. 4 - Ask Grok Bot to review the recording and create a reusable skill that follows the demonstrated workflow. 5 - Run the same task using the skill. Compare the result, accuracy and token usage. For a fair comparison, use the same inputs and check that both approaches produce the correct result. MY BIGGEST TAKEAWAY We're used to thinking of AI agents as something we give instructions to. But for repeatable jobs, it may be worth teaching them the process once and letting them reuse it. The real opportunity is finding which tasks benefit enough from that approach to justify the upfront effort.