What you'll learn
You've gone from "what is a prompt?" to building grounded, reliable, agentic systems. In this final module we look outward: prompting beyond text, how the field is changing, and — most importantly — how to keep your skills sharp as the models keep improving.
By the end of this module you'll be able to:
- Prompt effectively with images and other modalities
- Reason about how better models change (and don't change) prompting
- Choose between prompting, fine-tuning, and building
- Chart your own path to keep growing
Prompting with images
Many models are now multimodal — they accept images alongside text. You can ask questions about a photo, extract data from a screenshot or receipt, describe a chart, or debug a UI from a mockup. The prompting principles are identical: be specific about what you want from the image, specify the output format, and ground the answer in what's actually visible.
Tip
Audio, video & beyond
The same expansion is happening across modalities: audio (transcription, understanding tone), video (summarizing footage), and generation of images, audio, and video from text prompts. Each adds its own craft, but the core skill transfers directly — a clear, specific, well-grounded request is still the heart of it.
As models get better, what changes?
Models improve constantly, which raises a fair question: will prompt engineering become unnecessary? Some of it will — better models need less hand-holding, follow instructions more faithfully, and reason more on their own (as reasoning models already show, Module 10). But the core doesn't go away:
- You still have to say what you want. A model can't read your mind, however capable it is.
- Grounding still matters. Even great models don't know your private, current data.
- Reliability, evaluation, and safety only grow in importance as models take on more.
- Context engineering becomes more central as systems get more autonomous.
Key idea
Prompt, fine-tune, or build
A closing decision framework. When a plain prompt isn't enough, you have escalating options:
- Prompt engineering — start here; it's the fastest, cheapest, most flexible lever, and it's usually enough.
- Add context (RAG / tools) — when the model needs knowledge or abilities it doesn't have.
- Fine-tuning — for a consistent style or specialized behavior that prompting can't reach (Module 19).
- Build a system — chains, agents, and evaluation around the model for real applications.
Reach for the lightest option that solves the problem — and remember every one of them still relies on good prompting underneath.
Your path from here
You've covered the whole arc — foundations, the toolkit, reasoning, structured output, grounding, tools and agents, production, and applied practice.
Phase 1 · Foundations
How models read your words
Phase 2 · The toolkit
Instructions, roles, examples, structure
Phase 3 · Reasoning
Chain-of-thought & beyond
Phase 4 · Structured output
JSON, constraints, sampling
Phase 5 · Grounding
Hallucination, RAG, context
Phase 6 · Tools & agents
The API, tools, agents
Phase 7 · Production
Evals, security, safety
Phase 8 · Applied
Patterns, playbooks, what's next
The way to keep growing is to practice deliberately and stay curious: build something real, keep an eval set, save your best prompts, and read others' work as the field moves. Most of all, keep experimenting — the fastest way to get better at prompting is to prompt, inspect, and refine, every single day.
Key idea
Recap & final check
Key takeaways
- Multimodal models take images, audio, and video — the same specificity, format, and grounding principles apply.
- As models improve, low-level tricks fade but the core discipline — communicate intent, provide context, verify — endures.
- Escalate deliberately: prompt first, then add context (RAG/tools), then fine-tune, then build a full system.
- Keep an eval set, a template library, and a habit of measured iteration to keep improving.
- Prompt engineering is a way of thinking about communicating with AI — you now have the full foundation.
Quick check
1. How do prompting principles apply to images and other modalities?
2. As models get better, what happens to prompt engineering?
3. What's the recommended order when a plain prompt isn't enough?
4. What's the best way to keep improving at prompt engineering?
That's the whole course. You started not knowing what a prompt was; you can now design, ground, secure, evaluate, and ship prompt-powered systems. Congratulations — and happy prompting.