What are the 4Ds?
Delegation
Description
Discernment
Diligence
Delegation (4D)
When should humans do the work, and when should AI?
Success here is:
1. Understand your goal/problem you’re solving
2. Know what AI can do well / not well
3. Decide how to divide the work
Description (4D)
How do we communicate clearly with AI systems?
Sets up context!
Success here is:
1. What do you want the final output to be?
2. How do you want AI to approach it?
3. How do you want AI to behave?
Discernment (4D)
How do we evaluate what AI gives us?
Success here is:
1. Is the output useful?
2. Is AI taking the right approach?
3. Is the AI behaving correctly?
Diligence (4D)
How do we ensure our interaction with AI is responsible, transparent, and accountable?
Success here is:
1. Ensure accuracy
2. Be transparent about invovlemnet
3. Are you willing to be accountable for the work it helped with? Being willing to stand behind this.
What are three main ways that people engage with AI?
- Automation: The AI completes specific tasks based on your instructions.
- Augmentation: You and AI collaborate as creative thinking and task execution partners.
- Agency: You configure AI to work independently on your behalf, establishing its knowledge and behavior patterns rather than just giving it specific tasks.
What are examples of Automation?
The AI completes specific tasks based on your instructions.
- Summarize a document
- Draft an email
- Create an image
You define what needs to be done, and AI executes it.
What are examples of Augmentation?
You and AI collaborate as creative thinking and task execution partners.
- Developing a character for a story and getting stuck
- Difficult architectural issue with an app you’re building
Creative-thinking a problem-solving partner. Works best when solutions aren’t straight-forward and you need space to explore.
What are examples of Agency?
You configure AI to work independently on your behalf, establishing its knowledge and behavior patterns rather than just giving it specific tasks.
- AI assistant that categorizes emails by topic/urgency
- Power a dynamic website based on the type of visitor
Rather than establishing a script you’re describing behaviors you want it to take.
How do Description and Discernment work together in AI?
Most of our time is actually spent giving the AI Description, then Discerning the output, then refining more Description, getting and Discerning further.
Which 4D is critical for safe AI collaboration?
Diligence, because you take responsibility for standing behind your AI output.
What three elements make Generative AI possible?
- Algorithms
- Data (particularly websites, articles)
- Computation (specialized hardware)
What is a context window?
It’s a practical limit for how much information an LLM can consider at once
This includes: Prompts, it’s responses, and any other information you’ve shared in conversation
What are some limitations of LLMs?
- Knowledge cutoff date on when they were trained
- Incorrect information
- Hallucination
- Context window
- Lack access to tools, database, etc. even if it’s so smart it might not be able to answer without the access
What are the three Claude models?
- Opus (complex work)
- Sonnet (daily work)
- Haiku (simple tasks)
What is Model Context Protocol (MCP)?
MCP is a communication layer that provides Claude with context and tools without requiring you to write a bunch of tedious integration code.
You don’t need to author your own integration to another tool. A lot of times, products (AWS, Asana) offer their own MCP servers.
What’s the top criticism of MCP?
A lot of times people think MCP servers and tool use are the same thing.
They’re not, they’re complimentary.
MCP - You don’t need to author the tool function/schema.
Instead, someone is doing it for you.
What does an MCP expose?
- Tools
- Resources (Data, Files, Context)
- Prompts
