AI Learning Hub › Learn AI › Self-Paced Courses › Introduction to AI › Getting Started › Choosing Tools and Building Your AI Stack

Level 1 of 4 · Getting Started · Lesson 6 of 6

Choosing Tools and Building Your AI Stack

Starting from the task rather than the product when choosing AI tools and assembling your own set of them.

About 15 minutes

Skip to lesson

By the end of this lesson you should be able to

  • Work through a short selection path from a task to a tool, including an access and review step
  • Recognize when ordinary software, a template, or a person is the better choice
  • Sketch a small starting set of tools for your own work, beginning with what Stanford provides

A small, sensible starting set

An AI stack is simply the set of tools a person uses together. It does not need to be large. A practical starting point is one general assistant, reached through Stanford’s access routes, adding a specialized tool only when a specific task calls for it. Adding more subscriptions, or running every task through several models, adds cost and complexity without necessarily adding quality.

From task to tool in seven steps

  1. Define the task. What do you need done, for whom, and by when?
  2. Identify the inputs. What sources or material does it depend on: a public page, your notes, licensed legal databases, nothing at all?
  3. Identify the output. An explanation, a draft, an editable file, a table, practice feedback?
  4. Separate content from action. Do you need words or files you will review, or do you need something sent, scheduled, or changed? Actions raise different questions; keep them manual while you learn.
  5. Check permission, access, and information. Is the use allowed (course or office rules), is there a Stanford access route, and is the service approved for the information involved?
  6. Decide how you will review it. Against which source, by whom, and how long that will take.
  7. Compare with a non-AI option. Would a search, a template, a spreadsheet formula, or asking a person be faster or more reliable?

Three illustrative starting sets

Examples of choices, not required bundles. Start with one assistant and add a tool when a task calls for it. Check what your own account provides and what your work permits.

A student
Start with one Stanford-provided general assistant for explaining readings and self-quizzing, within each course’s AI policy. Add Gemini Notebook when you want questions answered from a set of readings, or CICERO, requested through the library, when you want to rehearse an oral argument. For legal authority, the library’s research databases remain the main resource; they are not an extra AI subscription. Better without AI: pulling up a case you already have the citation for, or asking a librarian where to start.
A faculty member
One general assistant for drafting teaching materials from public sources. When a research question calls for it, a research platform’s AI features for a first pass at authority, followed by reading it; and, for guided practice, the Hub’s AI Skills for course-related practice. Better without AI: a saved template for a recurring syllabus section.
A staff member
One general assistant for drafting public announcements and summaries of public pages; ordinary office software for the rest. Better without AI: a spreadsheet formula to total monthly registration counts. It suits a routine total; still check that the range covers every row, the values are right, and the result is plausible.

For wider discovery, the Hub’s Your AI Stack directory describes many more tools. A tool’s listing there is not an institutional approval; check Stanford’s Approved AI Services (opens in a new tab) and the Hub’s AI Resources for what the community can actually use.

Level check: choose for a task

A light self-check across this level, not a graded test.

A fictional task and fictional capability cards.

The task. The library’s events team wants a one-page handout for students about three upcoming public workshops. The inputs are the three public event pages. The output must be an editable document the team can revise, with each date, time, and room matching the event pages.

  • Card 1, general assistant: web search; file upload; side-by-side editing pane; can export a document. Offered through the university.
  • Card 2, source notebook: answers from sources you add; creates study guides and audio overviews; no document export described.
  • Card 3, legal research service: searches licensed case law and statutes; drafting tools for legal documents.
  1. Which tool category and which features fit the task?
  2. Name one access or permission question to settle first.
  3. What review step would you require?
  4. What is a reasonable non-AI alternative?
A worked answer
  1. Card 1 is a sensible candidate: web search or pasting the three pages, the editing pane, and document export. Confirm that its export produces an editable format the team can revise, such as a Word document; the card says only that it can export a document. Card 2 plus a word processor can also work: ask it about the pages, then transfer and revise the text in the team’s own document. The lack of an export claim does not make that workflow impossible. Card 3 is described as covering licensed legal sources and legal drafting, which this task does not need.
  2. For example: is the team’s account the university-provided one, and does the office have guidance on AI-assisted communications? The information is public, which keeps the data question simple.
  3. Compare every date, time, room, and title in the handout with the three event pages before it is printed. For a handout like this, a second person’s proofread is a good addition.
  4. Copying the three event details into the team’s existing handout template. For three events, that may be just as fast.

Other answers can be justified, as long as they name the features, an access question, and a review step.

Where to go next

You now know what the main kinds of AI tools are for, what their features mean, and where to find Stanford’s access routes. The Beginner level explains how this technology works, why fluent answers can still be wrong, and why human review remains necessary. Start there before moving on.

Start small and start with Stanford. Begin with the tools the institution provides, add one only for a specific need, and decide how you will check the result before you rely on it.

Resources