One question comes up in almost every class I teach, whether it is a room of nonprofit staff or a workshop of teachers: "I get how to ask a chatbot a question. But how do people actually use AI for real work?" It is the right question, and it comes up often enough that it is worth a full answer instead of a quick one.
A Quick Recap
A few months ago I wrote about the three tiers of AI tools: the web-based chat window most people start with, the AI now built into apps like Word and Excel, and a third tier of agentic tools such as Claude Code that work directly on your computer instead of just responding in a browser tab. If you have not read that post, Beyond the Chat Box is a good place to start. This one picks up where that left off, with concrete examples instead of the general framework.
What "Agentic" Actually Means
In a normal chat, you ask something and the AI answers. You read the answer, decide what to do with it, and if you want something different, you ask again. Every step runs through you.
An agentic tool works differently. You describe a goal, not a single question, and the AI breaks that goal into steps, carries them out on its own, checks its own work, and only stops to check in with you when something genuinely needs a decision. It can read files, search through folders, write new files, and run programs. Instead of handing you a draft to act on, it goes ahead and does the thing.
None of this requires the organization using it to know how to write code. It requires someone who knows how to set it up, point it at the right information, and build in the checkpoints where a human reviews the result. That is usually where a consultant comes in, not where every staff member needs to become a programmer.
Four Real Examples
1. Finding the Right Grant Before You Write the Proposal
Most grant research is a slog: scrolling through a database or a foundation's list of past awards, trying to guess which ones are worth the hours it takes to write a full proposal. An agentic tool can be pointed at a grant database export and an organization's mission statement, and asked to flag the genuine matches, the ones where the funder's stated priorities actually line up with the work the organization does. It can run once against a list you already have, or on a regular schedule to catch new postings as they appear. Once you have found the right grant, our grant writing guide covers the next step: actually writing the proposal.
2. Turning a Filing Cabinet Into a Searchable Archive
A lot of nonprofits and historical organizations are sitting on decades of scanned newsletters, meeting minutes, or old records that nobody can search because nobody has time to read through them page by page. An agentic tool can work through hundreds of scanned pages in a single sustained session, transcribing each one, flagging anything too faded or handwritten for it to read reliably so a person can double check it, and organizing the result into something you can actually search. I do this kind of work myself, on a much larger scale, for a historical research project of my own. The same approach scales down to a much smaller archive just as well.
3. A Report That Writes and Sends Itself
Some information is worth checking regularly but not worth a person's time to gather by hand every single day. Program numbers, a calendar of upcoming deadlines, attendance figures, whatever matters to your organization. An agentic tool can be set up to gather that information on a schedule, put it into a clean, readable format, and send it out automatically as an email, without anyone opening a chat window and asking for it each time. I run several small tools exactly like this for my own daily routine, and the setup is a one-time task, not a recurring chore.
4. One Prompt, One Finished Piece
A typical AI video or social media session is back and forth: draft a script, revise it, generate a visual, revise that, and so on. A single, carefully written prompt can instead drive an agentic tool through the entire production in one pass: script, visuals, and narration together, with the back and forth happening once, up front, in how the prompt is written rather than across a dozen separate exchanges afterward.
Why This Still Needs a Human in the Loop
Every one of these examples works best with a person checking the output before it goes anywhere that matters, whether that is an email inbox, a donor list, or a published report. An agentic tool that can act on its own is also a tool that can act on a mistake just as readily as it acts on the right thing. The organizations that get real value from this tier of AI are the ones that build in a review step, not the ones that hand over the keys and walk away.
Where This Fits
This is exactly the kind of work that falls under our automation and workflow service at Cochise AI. You do not need to learn to use a command line or become a programmer. You need someone who can look at a repetitive or time-consuming task your organization already does and figure out whether an agentic tool can take it off your plate, and then actually build it, with the right checkpoints in place.
If any of these examples sound like something your organization deals with, or you are curious what else this tier of AI could handle, the contact form is the best place to start.
