If you’re an engineer, developer, or even a tech enthusiast just trying to figure out where AI can actually fit into your day-to-day work, you’ve probably run into a wall of buzzwords: LLMs, RAG, agents, prompt engineering, fine-tuning, context windows, hallucinations. It’s easy to feel like you need a PhD or university degree just to use this new tool well.
Fortunately, you don’t. You simply need my handy-dandy guide to understanding AI jargon. Here’s a practical guide to the terms that appear most frequently, and — more importantly — tips on how you can actually use these tools to get better work done, faster.
First, the basics
LLM (Large Language Model) – The engine behind tools like Claude or ChatGPT. It’s a model trained on huge amounts of text that predicts what comes next in a conversation and focuses on natural human language and communication. Think of it as a well-read collaborator who can write, explain, and reason about almost anything — but who has never actually run your code or opened your project.
Prompt – This is the input you give the model. “Prompt engineering” sounds fancy, but it’s really just the skill of writing clear instructions that a computer – or AI – can successfully process in as few iterations as possible.
Context window – This one tripped me up when I first started using tools such as Open code. It means the amount of text the model can “see” at once — your conversation history, any documents you’ve shared, and its response, combined. Once a conversation gets long enough, older parts start falling out of view. This is why AI tools sometimes seem to “forget” something you said earlier.
Hallucination – When the model states something false with total confidence such as citing a library function that doesn’t exist or recalling a fact that is wrong. It doesn’t happen because the model is “lying”; it happens because the model is optimized to produce plausible-sounding text, not to verify facts against reality. Always treat generated facts, APIs, or citations as claims to check, not truths to trust.
Fine-tuning – Additional training on a specific dataset to make a model better at a more niche task. Most engineers never need to do this — general-purpose models and good prompting handle the vast majority of real work.
RAG (Retrieval-Augmented Generation) – A technique where the model looks up relevant information (from a database, your codebase, or documentation) before answering, instead of relying purely on what it learned during training. This is how tools “know” about your specific company’s data or a recent update. The answer is grounded in retrieved documents rather than memory alone which makes it really good for picking up on specific contexts.
Agent / agentic – Instead of replying to a promp once, an “agent” can take multiple steps on its own — running commands, reading output, deciding what to do next, and looping until a task is done. Tools like Claude Code or Cursor’s agent mode work this way: you give a goal, and the AI plans and executes a sequence of actions rather than just producing a single answer.
Token – The unit AI models process text in — roughly ¾ of a word. This matters mainly because pricing and context limits are measured in tokens, and long files or codebases can eat through that budget fast. Think of it a lot like a currency to use your model 🫠.
Where AI actually helps an engineer
Now that you have more than a vague idea of what key AI terms are, the real question is: how can AI fit into your workflow? You don’t need to incorporate AI into everything, but these are my top 5 low-effort ways of using AI in my day to day work as a Software Engineer, including at Canonical.
1. Reducing the cost of starting. Blank files and unfamiliar codebases are where a lot of time gets lost. AI is very good at scaffolding and helping you plan. This could be a first draft of a function, a boilerplate config or even a rough test suite that you then edit rather than write from scratch.
2. Explaining unfamiliar code. If you point an AI tool at a confusing function or an unfamiliar library and ask it to explain what’s happening line by line. This is often faster than digging through documentation, especially for legacy code with no comments.
3. Writing tests and edge cases. AI is strong at brainstorming edge cases that we humans may not think of such as empty inputs, race conditions and malformed data which is valuable even if you write the final test code yourself.
4. Code review and refactoring suggestions. Asking “what’s wrong with this function” or “how would you simplify this” often surfaces real issues.
5. Debugging. When you paste an error message and the surrounding code, AI tools can pattern-match common bugs and suggest where to look. That being said, they can still guess wrong, so verify its suggestions against what the code is actually doing, not just what the model says it’s doing. This goes for agents such as CoPilot too.
The bottom line
AI tools don’t require you to master new vocabulary before you start. Free versions of AI tools exist such as ChatGPT and Claude AI. For those who care, the jargon describes how the tools work under the hood; using them well is mostly about applying the same skills you already use with your human collaborators – I mean, colleagues. Start small: use AI to explain one confusing function or draft one test file and see where it saves you real time, before building from there!
I hope this article was insightful for you! Until next time~
Nkeiruka
#LifeAtCanonical #AIExplained #AIJargon #AIEducation
Leave a Reply