**By:** webdevel@sparkhausmedia.com (Lori White)
© Lead Image © pitinan, 123RF.com
AI is here to stay, and understanding how it works under the hood can mean the difference between frustration and genuinely useful results.
This article covers LLM fundamentals, effective prompting, and a structured agentic workflow that puts you in control.
Whether you’re an AI advocate or skeptic, there’s no doubt that it is here to stay, and it is already having a massive impact on our personal and professional lives.
The speed at which AI tools and models are being developed is staggering, in part a result of AI companies reaching the point where their own AI tools write the code for ongoing evolution and refinement of the AI tools! Development is no longer restrained by the speed of a human.
This is simultaneously impressive and concerning.
There’s a near certainty that by the time this article goes to print, there will already be some major changes in AI.
However, the aim of this article is to talk about the fundamentals of working with AI in order to better understand the core concepts, which, and I hesitate to write this, are unlikely to change significantly.
I’ll describe the steps for building an AI-generated applications – first from a more intuitive approach that some might think would fall into the category of what is considered vibe coding.
Along the way, I’ll point out some of the problems that come with adopting this method.
Then I’ll show you a more rigorous approach that will produce a result that is perhaps closer to professional software development.
But first, I’ll begin with some key concepts.
Tokens are the currency of LLMs.
They’re used to measure usage, limits, and pricing.
An LLM doesn’t see text the way you do, so any text you send it is first broken down into tokens.
For example, “cat” might be one token, and “cats” might also be one token.
Here is another example: “Unbelievably I can’t stop.” is four words but seven tokens.
The two culprits are:
[…]
We’ll show you some best practices for introducing Claude Code (or another LLM-based coding assistant) while maintaining knowledge and control of the code.
The Warp AI agent takes the guesswork out of working at the command line.
We show you how to build a simple website with one prompt.
Affecting millions of systems, a kernel flaw discovered by Qualys could allow users to gain root privileges.
If you’ve ever wondered if your laptop or PC is officially certified to run Ubuntu, that curiosity will soon be met.
The lastest version of IPFire features a fundamental change to how the system handles DNS.
It’s now possible to test experimental features on the Gnome desktop without worrying that you’ll break things.
Hiding out for nearly 15 years, the Ghostlock vulnerability allows a standard logged-in user to gain root privileges.
A 16-year-old vulnerability allows an attacker to escape a virtual machine, gain access to the host, and execute malicious code.
Developer Noah Cagle decided the world needed the once obscure but beloved Linux distribution and gave it a decidedly pink refresh.
If you’re looking for a laptop with tons of power and battery, look no further than the latest iteration of the System76 Lemur Pro.
Using the cloc utility, Michael Larabel of Phoronix discovered that Linux kernel 7.2 has over 43 million lines of code.
The Kubuntu Focus team has upped the performance ante of its M2 and Zr laptops with the latest, greatest CPUs from Intel.
—
**📚 Original Source:**
[AI and Agentic Workflows](http://www.linux-magazine.com/Issues/2026/308/Claude-Code-Workflow)
About The Author
