Working with Generative AI

Most AI guides start too high or too low. This one starts where you are.

Getting better at AI is less about the tools and more about knowing what to give them. Below is a path from your first conversation to designing production systems, plus the systems that make AI genuinely useful. Find your level and start there.

Find your level The context stack Delegate & automate

Where are you?

Not a quiz, just a signpost. Pick the one that sounds most like you and jump to it.

L0 I've heard about it but haven't really used it L1 I use it sometimes but results are hit or miss L2 I use it daily, mostly in chat, and I have opinions L3 I'm connecting it to real systems or building workflows L4 I'm designing AI products or advising teams on adoption

Thread 1

The skill ladder

Five levels, from your first conversation to designing systems. Each one has something to read, prompts to try, and a way to tell you're ready for the next.

L0

First contact

Understanding what generative AI actually is, and getting a genuinely useful answer out of your first conversation. The mental model: a fast, confident collaborator with broad knowledge, who can be wrong and only knows what you tell it.

Try this first 
I want to understand [topic]. Explain it to me clearly - assume I'm intelligent
but have no background here. Start with the core idea, give me one good analogy,
then tell me two things I'd need to know to go deeper.
Sanity-check an answer 
You just told me [paste what the AI said]. How confident are you that this is
accurate, and what should I verify independently before I act on it?

What to read

Resources

sebastianebg/ai-prompts Copy-ready prompts for your first conversations and beyond.
▸ You're ready for the next level when...

You've had at least one conversation where the output was genuinely useful, you know it can be wrong, and you've tried a follow-up to push it further.

L1

Prompting that works

Understanding why prompts fail and how to structure requests that work. Most prompts fail because you already know the context and forget to include it. Every good prompt has four parts: context, role, format, constraint.

The four-ingredient frame 
I need help with [specific task].

Context: [who you are, what you're working on, any relevant background]
Goal: [what a great output looks like]
Format: [email / bullet list / one paragraph / table / etc.]
Constraint: [length limit / tone / what to avoid / who the audience is]
The iteration move (when the output is off) 
This isn't quite right. The issue is [specific problem with the output].
What I actually need is [clarification of what was missing or wrong].
Try again with that in mind.

What to read

Resources

sebastianebg/ai-prompts Context-rich templates you can fill in and reuse.
▸ You're ready for the next level when...

Your prompts include context by default, you iterate on a bad answer instead of starting over, and you get usable output more often than not.

L2

Power user

Persistent context and systematic prompting. You stop re-explaining yourself every session by setting up project files and system instructions, and you use the AI as a thinking partner, not just an answer machine.

A reusable working relationship 
You are a [role] helping me with [area of work].

About my situation:
- [relevant fact about you or your context]
- [constraint or goal]
- [anything you should always keep in mind]

How to work with me:
- Default to [preferred tone / format]
- If you're unsure, say so explicitly rather than guessing
- Never [thing to avoid]

When I give you a task, confirm what you understood before you start.
The thinking partner 
I'm going to think through [problem or decision] out loud. Your job is to push
back on the assumptions I'm making, spot gaps in my reasoning, and ask one
sharp question at the end. Don't hand me the answer - help me find it.

Here's where I am: [your thinking so far]

What to read

Resources

sebastianebg/ai-context-templates System prompts and project-context docs so your AI arrives pre-briefed.
▸ You're ready for the next level when...

You have a project file or custom instructions you actually maintain, and you've noticed the AI needing less hand-holding as a result.

L3

Builder

Moving beyond chat. Connecting AI to your documents, systems, and actions, through retrieval, tool use, and no-code automation. This is where prompts matter less and architecture matters more.

Ground the AI in a document 
I'm going to share a document. Read it carefully before I ask anything. When you
answer, cite the specific section you're drawing from. If the answer isn't in
the document, say so plainly rather than filling the gap.

[paste document]

What to read

Resources

sebastianebg/github-as-kb A fork-ready repo structured as an AI-accessible knowledge base.
▸ You're ready for the next level when...

You've connected an AI tool to a real document store or workflow, and you know when an agent is overkill versus when it earns its place.

L4

Architect

Designing production systems and advising teams. Model selection, multi-agent patterns, evals, cost and reliability, and the organisational change that is most of the actual work. There is no single prompt for this level - it is too context-dependent.

What to read

▸ You're ready for the next level when...

You're making build-versus-buy calls, designing how systems fit together, and the bottleneck is adoption and trust, not the model.

Working at the top of the ladder?

If you're designing AI systems or leading a team through adoption and want a second opinion or someone to accelerate the build, that's most of what I do.

Start a conversation

Thread 2

The context stack

The other side of working with AI well is setting up the environment it works in: what it can see, what it remembers, and how it finds things.

What is Git, and Why Does My AI Use It?

The environment most AI coding tools work inside, explained without a terminal.

Using GitHub as a Knowledge Base

Structure a repository so your AI tools can actually find and use what is in it.

Memory, Project Files, and Retrieval

Three ways to get information into your AI tools, and how to pick the right one.
sebastianebg/github-as-kb Fork this template to stand up an AI-ready knowledge base in minutes. sebastianebg/ai-context-templates System prompts and project-context docs for any AI tool.

Thread 3

Delegate & automate

Once you're good at working with AI, the next move is to stop doing the work by hand. Package your prompts into skills, define agents you can reuse, and let the routine work run on a trigger or a schedule, with a human checkpoint where it counts.

What Are AI Skills

Package your best prompts into named, reusable capabilities you invoke instead of retyping.

Building Your Own AI Agents

When you delegate the same task twice, define an agent. The anatomy of a reusable template.

Automating Work with AI

From a chat window to a workflow that runs without you. Triggers, AI steps, and checkpoints.

Scheduling AI Work

AI that runs on a timer: recurring digests, watchers, and overnight batch work.

Templates to copy

ai-context-templates/skills Reusable skill definitions and MCP prompts-as-skills. ai-context-templates/agents AGENTS.md starters and subagent definition templates. ai-context-templates/automations No-code and code automation recipes with an AI step. ai-context-templates/schedules Recurring digests, watchers, and overnight batch patterns.