Everything you need, nothing you don't.
The reference handbook. Pop in mid-level when stuck. Each section is anchored by what it changes about your prompt.
Why learn prompt engineering?
AI tools are everywhere, but most professionals use them like they're DM-ing a junior colleague. Vague, contextless, then surprised when the output is mush. The shift: execution cost has dropped to near-zero. The value of your work now lives in the decisions you make: what to build, what to exclude, what to challenge.
Direction over execution
You're not a typist anymore, you're a director. The prompt IS the work.
It pays back
Each session makes the next faster. Better prompts today → better outputs forever.
It's a craft, not a hack
Plan, direct, refine, encode, shifting roles. Not a single skill, a practice.
Sustainable prompting
Sustainable prompting isn't about writing shorter prompts, it's about writing smarter ones. More context upfront = fewer correction cycles = less total token spend. A 200-token prompt that produces a usable 300-token response is more sustainable than a 20-token prompt that triggers 5 rounds of 500-token corrections.
High token spend, low signal, output isn't reusable.
Lower token spend, high signal, prompt becomes a reusable template.
Why use AI responsibly?
AI use should be effective and efficient, but also ethical and safe. Three diligences to anchor your practice:
Be intentional
Critically aware of which AI systems you use, how you work with them, and the impacts.
Be open
Open about AI interactions with relevant stakeholders. Who needs to know? At what level of detail?
Take responsibility
Verify facts, check for biases, ensure accuracy, and check usage rights before deploying.
PE 101 · The 4D Framework
Anthropic's core model: four interconnected competencies that work together, not in sequence.
Delegation
Decide which tasks to assign to AI, and how. Domain expertise is the foundation.
Description
Clearly define what you want. Context, format, audience, style, constraints.
Discernment
Critically evaluate what AI produces. A continuous loop of instruction and evaluation.
Diligence
Take responsibility for AI interactions and their impacts. Be rigorous and accountable.
Description has three flavours
What you want
Output characteristics: context, format, audience, style, constraints.
How to approach it
Should AI think step by step? Debate itself? Explore multiple paths before concluding?
How to behave
The behavioral aspects of the interaction. AI behaves differently in different contexts.
Six techniques that compound
Each addresses a specific failure mode in how AI processes instructions.
Define role, style, tone
Not "act like a designer" but "act as a senior product designer who has shipped AI features in regulated industries."
Ask AI to think first
"Before answering, think through the problem carefully. Consider constraints and approaches before recommending."
Break into steps
For complex tasks, number the steps. "1) do X, 2) compare Y, 3) highlight Z, 4) suggest reasons."
Negative constraints
Tell the model what NOT to do. "Do not jump to solutions, first help me align on the problem." The most underused.
Define output format
"Give me: a) key user needs, b) business goals, c) P0 features with rationale," not "a comprehensive response."
Ask for variations
"Give me 3 versions: conservative, balanced, aggressive, so I can compare directions."
Evaluating AI output
Discernment ensures AI collaboration remains guided by human judgment. Three lenses:
Did it produce good work?
Factually accurate? Appropriate to audience? Coherent? Adds value?
Did it reason well?
Watch for: logical inconsistency, lapses in attention, getting trapped in circular reasoning.
Did it behave right?
Communication style appropriate? Right level of detail? Response to feedback efficient?
The senior-level unlock
Chain-of-thought prompting. Add "Walk me through your reasoning before presenting the answer." For any ambiguous or high-stakes prompt, this dramatically improves output quality. The model is forced to show its work before concluding.