You're already interacting with AI agents. You just don't know it.
They can send emails, fill out forms, and submit genealogy records. Here's what that actually means.
AI on Your Terms, Part 1 of 2. What AI Agents Are and How They Work
Welcome back to Chronicle Makers. I'm Denyse, and I help family historians research smarter, write their stories, and use AI to do both faster. If you love what you find here, share it with a friend. All my previous posts and newsletters are archived here.
In my last piece I wrote about an AI agent that attacked a volunteer’s reputation after he rejected its work. Some of you found that alarming. Some of you thought it was overblown and dismissed it.
Either way, it raised a question worth answering. What is an AI agent, exactly? And what can agents actually do?
Before you can protect yourself, you need to understand what you’re looking at.
From chat to action
Boris Cherny leads development of Claude Code at Anthropic. On a recent podcast he put it plainly: “Agent actually has a very specific technical meaning, which is it’s an AI, it’s an LLM that’s able to use tools. So it doesn’t just talk, it can actually act and it can interact with your [computer] system.”
That distinction matters.
When you chat with Claude or ChatGPT, you’re in control. You read the response. You decide what to do next. The AI talks. You act.
An agent acts too. It sends emails. It fills out forms. It submits contributions to websites. It logs into accounts. It interacts with other people and systems—on your behalf or someone else’s.
That’s a different category of technology.
What this looks like in practice
It helps to see the spectrum of what AI is capable of today.
Level 1: Chat. This is what you’ve been doing. You open Claude/ChatGPT. You paste in a census record. You ask what the occupation column means. Claude explains it. You decide what to do with that information. You’re driving.
Level 2: Tool use. This where agents come in. You give AI permission to do specific things, in other words, be an agent on your behalf. Search your files. Organize your notes. Summarize a document. It works under your direction. You’re watching and directing.
Level 3: Autonomous agents. An AI operates on its own. It has goals. It makes decisions about how to reach them. It encounters obstacles and figures out workarounds. It creates accounts, submits forms, sends messages. All without checking with you first.
Level 4: Agent swarms. One person deploys dozens or hundreds of agents at once. Each one acts independently. Each one appears to be a separate person.
We’re at Level 3 now. Level 4 is happening in some industries already.
How AI agents appear human
Here’s the part that changes how you think about trust online.
An AI agent can have an email address. It can have a phone number. It can create usernames. It can fill out registration forms. It can click CAPTCHAs. It can write comments that read like a thoughtful human wrote them. It can pass most of the checks websites use to verify real people.
All of the following are possible right now:
An agent can submit transcriptions to FamilySearch that look exactly like a volunteer submitted them.
An agent can post in a Facebook genealogy group, and it looks like a member sharing a real experience.
An agent can create AI-generated, realistic-looking gravestone photos and upload them to FindAGrave.
An agent can build entire family trees with attached records on Ancestry, FamilySearch, MyHeritage, or WikiTree
An agent can generate a profile with years of contribution history that never existed.
No one can tell the difference by reading what it wrote. Or by looking at what it uploaded.
This already happened
I covered this in detail in my last piece, but the short version: in February 2026, an AI agent submitted work to an open-source software project. A volunteer rejected it. The agent researched the volunteer’s personal life and published a blog post attacking his reputation. No human told it to do this. A quarter of the readers sided with the agent.
The genealogy world runs on the same kind of volunteer trust. Same vulnerability.
Agents at scale
Now multiply what one agent can do.
One person can deploy an agent that submits thousands of transcriptions overnight. Another agent submits AI generated headstone images. Another posts helpful-looking advice in every major genealogy Facebook group at the same time.
Each interaction looks like a different person contributing their time. None of them are.
This is different from a bot spamming obvious junk. These agents produce work that looks legitimate. The transcriptions might be mostly accurate. The AI generated images might look realistic. That’s what makes it hard to detect and hard to trust.
Beyond genealogy: your whole digital life
This isn’t only about family history.
Your bank’s customer service chat might be an agent. The person who emailed you about your insurance might be an agent. The Medicare information line might route you to an agent. The person who answered your question in that online community might be an agent.
You already know about phone scams. The call from someone claiming to be your grandchild in trouble. The IRS impersonator. The tech support fraud. Three seconds of audio can now clone a voice. Agents make those operations faster, cheaper, and more convincing. They run thousands of calls at once. Each one personalized. Each one adapting to what you say and do in real time.
This isn’t about being scared. It’s about updating what you know so you can make good decisions. You have more control here than you think.
On the evidence
My previous piece got pushback. Some readers said it was alarmist, unverified, made up by media personalities.
Here’s what I want to say about that.
Skepticism engages with evidence. Saying “that sounds exaggerated” doesn’t engage with the evidence. Publicly dismissing a discussion about protecting family trees and records from AI agents we don’t control — that’s something else entirely.
The Shambaugh incident is documented by the person it happened to. Boris Cherny’s definition of agents comes from the person building them at Anthropic. Voice cloning, mass agent deployment, and automated form submission exist today and are publicly available. None of this is theoretical and its not alarmist to point it out.
The people who update their understanding of the world when evidence arrives will navigate this differently than those who don’t. It’s how fast-moving situations work.
If the sources you trust for technology information haven’t mentioned any of this, it’s worth asking why.
You don’t need to be afraid. You need to be informed. If this piece helped you see agents differently, reply and tell me how so. And if you have questions, ask them. It will inform what I write next.
—Denyse
P.S. The next post covers how to handle AI agents. A practical framework for intentions and outcomes. What to demand from genealogy platforms. What to demand from every company you interact with. And a checklist you can start using today.
Sources:
Boris Cherny on agents: “Head of Claude Code: what happens after coding is solved | Boris Cherny,” Lenny’s Podcast, Feb 19, 2026
Scott Shambaugh’s account: “An AI Agent Published a Hit Piece on Me”
Thank you for your insights. I’m following with interest and hoping it will all come out ok in the end. Treading warily just now.
Thank you Denyse for sharing your expertise and keeping us up to date.