10 · Project — Build a Personal AI Toolkit¶
This capstone pulls together every module in Level 1 into a single deliverable: a personal AI toolkit of 3-5 tools, each chosen deliberately and justified in writing, plus the verification habits you'll actually use with each one. This is not a theoretical exercise — the template below produces something you can keep using after this course.
Why a written toolkit, and why now¶
Most people accumulate AI tools accidentally — whatever they tried first, whatever a colleague mentioned. A deliberately built toolkit, chosen against your actual recurring tasks (Module 1) using a real decision process (Module 2), with matched verification habits (Module 9) already built in, outperforms an accidental collection because you already know why each tool is there and how much to trust each one's output.
The toolkit template¶
1. Task inventory (Module 1)¶
| Recurring task | Category | Frequency |
|---|---|---|
2. Tool selection (Module 2)¶
For each category represented above, name the tool (or type of tool if undecided) you'll use, and run the four-question framework:
| Category | Chosen tool | Stakes level | One-off or recurring? | Context needed |
|---|---|---|---|---|
3. Per-tool verification habit (Module 9)¶
For each tool in your toolkit, state the specific verification step you commit to using every time, matched to the stakes level from section 2.
| Tool | Verification habit | Trigger for extra scrutiny |
|---|---|---|
4. Prompting notes (Module 8)¶
For your two most-used tools, write one example prompt that worked well for you, and one lesson learned about what to include (context, constraints, examples) to get better results from that specific tool going forward.
5. Boundaries — what you will NOT use AI for (Module 2 & 9)¶
List at least two tasks or situations, specific to your own life or work, where you've decided AI tools are not appropriate — because the stakes are too high, verification isn't feasible, or a policy prohibits it.
Worked example (filled toolkit, abbreviated)¶
A small-business owner's filled excerpt:
- Task inventory: customer email replies (writing, daily), social media graphics (image, weekly), monthly bookkeeping summary (data analysis, monthly), website copy edits (writing, occasional).
- Tool selection: a general chat assistant for email replies (low stakes, recurring, minimal context needed beyond the customer's message); an image generation tool for social graphics (low-medium stakes, recurring, needs brand style context); a data-analysis-capable assistant for the bookkeeping summary (medium stakes since numbers feed real decisions, monthly, needs the actual spreadsheet as context).
- Verification habits: email replies get a 10-second read before sending (low stakes, fast check); the bookkeeping summary gets every number cross-checked against the source spreadsheet before she trusts a total (medium-high stakes, full check every time); social graphics get checked against the tool's commercial-use terms before public posting.
- Prompting notes: for email replies, including the customer's exact wording plus "match a warm but brief tone, 3 sentences max" consistently outperformed a generic "write a polite reply."
- Boundaries: she will not use AI to draft responses to formal complaints or anything mentioning a refund dispute — those get a fully human-written reply, since the stakes and nuance are too high for a quick-check verification habit to catch every problem; she also will not use an AI tool to generate final numbers for her tax filing without a human accountant's review.
How It Actually Works¶
Choosing tools by category rather than by brand — the approach this project asks you to formalize in writing — holds up well precisely because category maps to underlying mechanism, and mechanism is what actually determines a tool's strengths and blind spots. A "writing assistant" and a "research assistant" might be running the exact same base model under the hood, wrapped in different system prompts, different default temperature settings, and different amounts of retrieval — but knowing that is what tells you where each is trustworthy without re-testing every tool from scratch. Your toolkit document is, in effect, a personal map of which mechanism you're relying on for which task: pure generation from training data (treat with skepticism), generation grounded in something you supplied (trust much more, but still check), or generation plus a tool-use loop that checked its own work against real output (trust proportionally to how much of the work that loop actually verified).
Writing this down matters because these underlying mechanisms are far more stable than the branded products sitting on top of them. A product you chose can be discontinued, repriced, or have its system prompt changed overnight by its provider — but the reasoning "I trust this category of tool for grounded tasks and verify anything it recalls from memory alone" survives every one of those changes, because it's a statement about how the technology works, not about which vendor currently implements it best. That durability is the entire point of building the toolkit as a document with reasoning attached, rather than as a bare list of app names.
Exercise (the deliverable)¶
Fill out all five sections of the template above completely, using your own real tasks, tools, and habits — not hypothetical ones. Where you genuinely haven't picked a tool yet, write "Undecided — trying [option] and [option] next" rather than leaving it blank. Keep this filled-out toolkit; Level 2's capstone (an AI-augmented workflow design) builds directly on it.