Diving into Agentic AI

First published April 2026

Over the last two years, I’ve built or co-developed a lot of custom GPTs (custom chatbots, essentially) in ChatGPT and repeatedly run into some of the same limits: they could provide good content, but they were often detached from current information, prone to filling in gaps ('hallucinating'), and not able to manage ongoing or evolving work (so they outdated very quickly).

I’ve been trying out agentic AI workflows as an alternative, along with so many others, and finding them really useful.

Agentic AI refers to systems that can take actions on your behalf: working through multi-step tasks, interacting with tools or data sources, and iterating toward a goal without generating as many mistakes.

In the right setup, agents can read from current files, sort through large amounts of information, and surface the most relevant quotes or examples, then automate repeatable tasks.

Because they’re working from data and defined workflows, the output tends to be more consistent and reliable than asking a chatbot to generate something, even with a knowledge base and training (there are still mistakes and human review is pretty critical).

For example, I’ve been experimenting with building (primarily through ChatGPT and Claude):

• A content agent that pulls from the most recent files and analysis (e.g., Google Analytics, email data) to create content aligned to specific outcomes and audience goals

• An email agent that runs each morning, summarizes high-priority messages in my inbox, and drafts initial responses where it makes sense

• A reporting/data agent that can read through content databases and pull a redux of recent content or data for specific uses (e.g., donor engagement or reporting)

• An agenda agent that reads through meeting notes and crafts follow-up emails/info with action items and weekly agendas for the upcoming week,, then automatically puts the drafts in your email or cowork space for review/send/publish

  
  

Where agentic AI shows up in systems: ChatGPT is making agents a visible product layer in the sidebar, Claude enables them more implicitly through context and integrations (connectors), and Copilot is pushing them into day-to-day operational systems.

That said, I do not think the answer is to use AI everywhere and in every case. There are still real concerns to weigh, including environmental impact, data security, privacy, and whether a workflow genuinely benefits from AI at all.

If you have wisdom/experience to share about any of this, I’d love to hear it. If you want to read more, here’s a recent article for nonprofit leaders on agentic AI: https://nonprofitsdecoded.com/agentic-ai-nonprofits-explained/