AI4RA Tip of the Week #19

From tasks to tools: Three Levels of AI Use 

Not all AI use is created equal. This week’s tip introduces a simple framework for thinking about how deeply AI is integrated into your work, and what the tradeoffs are at each level. Think of AI use on a spectrum. At one end, you ask AI to do something once. At the other, AI is embedded in a tool that does it automatically, every time. Each level has advantages and limitations. 

Level 1: Ask AI to Do the Task

This is the most common entry point. You bring your data or document to an AI tool and ask it to do something directly, like summarize a report, analyze a spreadsheet, or draft a paragraph.

  • Fast and requires no technical skill 
  • Great for one-off tasks or exploring what’s possible
  • Not repeatable — the next person who needs the same table has to start over 
  • Low transparency — it can be difficult to explain exactly how the AI produced the output or reconstruct it later 
  • Results may vary depending on how the prompt is phrased  

Example: You upload a budget spreadsheet and ask the AI to reformat the data into a table summarizing expenses by category and specific time periods.

Prompt: “I am a research administrator, and I have been tasked with assisting with a post award summary of expenses, to help a PI with their project management. To prepare for this, I’ve attached a budget expense spreadsheet. Please create a summary table showing total expenditures by cost category (personnel, fringe, travel, supplies, indirect) for each quarter, beginning with the first quarter that expenses appear in. Format this in an excel output as a table.” 

Level 2: Ask AI to Build a Formula or Script

Instead of doing the task itself, you ask the AI to write a formula, macro, or script that you (or anyone on your team) can run in an existing tool like Excel, Python, or SQL.

  • Repeatable — the formula or script works the same way every time
  • Transparent — the logic (formula) is visible and reviewable by anyone who opens the file 
  • Portable — colleagues can use it without needing access to an AI tool 
  • Requires some comfort reading and applying code or formulas 
  • Breaks if your data structure changes (column names, added fields, etc.) 
  • Still requires a person to run it manually 

Example: You ask the AI to write an Excel formula using your specific column headers and row names that automatically calculates and organizes the table — you do not need to use AI to run it again. 

Prompt: “I am a research administrator and I have been tasked with assisting with a post award summary of expenses, to help a PI with their project management. This will be an ongoing task, where I will need to update this output monthly.  Using the excel file I upload, write an Excel SUMIF formula that totals expenses from column C where the cost category in column B matches ‘Personnel’. My data starts on row 2 and runs through row 200. The output should be an excel formula I can use each month in my workbook.” 

Level 3: Ask AI to Build a Tool

Using agentic coding, you ask AI to build a standalone application that automates the task entirely, so there are no manual steps required once it’s set up.

  • Fully repeatable and scalable across your whole office or institution 
  • Frees staff from repetitive manual work entirely 
  • Can be shared, version-controlled, and maintained as an institutional resource 
  • Requires the most upfront investment — time, technical skill, or a developer 
  • Transparency requires intentional design: if the tool is a black box, auditing outputs is harder 
  • Maintenance burden — when data structures or requirements change, someone has to update the tool 

Example: An app that watches for new data files, automatically generates the pivot table, and outputs a formatted report — without anyone having to open Excel or write a formula. 

Prompt: “Build a Python app that monitors a folder for new Excel budget files. When a new file arrives, read the data, generate a pivot table summarizing expenditures by cost category and quarter, and save a formatted summary report as a new Excel file in an output folder. The input files will always have the same column structure: Date, Cost Category, Amount, Award Number.” 

Why This Matters for Research Administrators

The right level depends on what you’re trying to accomplish. A one-time task? Level 1 is probably fine. A calculation your office runs every month across dozens of awards? Level 2 gives you something repeatable and auditable. A compliance check your institution needs to run automatically at scale? That’s a Level 3 need. 

Transparency is worth calling out specifically. Level 1 outputs may be difficult to explain if an auditor or supervisor asks how a number was derived, “the AI gave it to me” is not a satisfying answer. You can combat this with documentation and evaluation (learn more about how to document and evaluate AI using the TaMPER framework here). Level 2 formulas and Level 3 tools make the logic visible and reviewable, which matters in a compliance-heavy environment like research administration. 

Most RAs perform tasks that could benefit from use of all three levels. The goal isn’t to always reach Level 3, rather it’s to recognize which level fits the task, and to move up the ladder deliberately when repeatability and transparency justify the investment of time and effort related to Levels 2 and 3. 

How can I learn more about this topic?  
AI4RA Office Hours – Occurs every fourth Tuesday at 11 AM PST and includes live Q&A and demos (https://ai4ra.uidaho.edu/#events).  

Share your questions or success stories  
Post them to the AI4RA community forum (https://groups.ai4ra.uidaho.edu/g/main/topics). Your tips help the whole research‑admin community thrive. 

Leave a Reply

Your email address will not be published. Required fields are marked *