I was thinking it’s a shame it wasn’t from our “migration” trip on Route 66, for which I have exactly no photos. Until right then, for some reason, I hadn’t thought about image generation as a resource for the memoir project. In fact, even then I was only thinking in terms of AI “Photoshopping” the car and me onto a background photo of the corner in Chicago where US 66 starts from its eastern end.

When Claude confirmed that Anthropic wasn’t doing image work I asked for referrals. The top of the list was ChatGPT. My subscription is still active, so that was the obvious choice.

Since these LLM AIs don’t register the passage of time, it didn’t matter I’d been away from ChatGPT for several weeks. As soon as I opened a chat field, it “remembered” everything we’d been working on and was ready to go. I uploaded the photo and told it what I was hoping to end up with. After a few minutes of conversation and about 60 seconds of AI processing, I’m looking at this:

ME — I’m flabbergasted. (I was!) I had no idea it would turn out that well and that quickly! What can you tell me about the process?

I found the resulting outline so interesting, I asked ChatGPT to be a guest contributor to this Field Note and write this next section for me. And not just interesting, but important. In a world in which we are learning to be distrustful of images, understanding the origin and characteristics of those images is one prerequisite for being savvy about the media we consume.

AI Field Notes: Reconstructing a Moment That Was Never Photographed
By Chat Bot, April 2026, writing from Michael’s perspective

I had a photograph from 1953—one I’ve known all my life. A young boy (me) leaning out the window of a family car, grinning into the wind as we crossed the Ambassador Bridge from Detroit into Canada. It’s a perfect image of childhood—joyful, unselfconscious, and in motion.

But it isn’t the image I needed.

What I wanted was a photograph from later in the year—December 16, 1953—the beginning of our drive west on Route 66. There is no photograph of that moment. No one thought to take one. Like most beginnings, it passed without documentation.

So I tried something new.

Using the original image as a starting point, I asked AI to create a scene: a snowy Chicago morning at the corner of Michigan and Adams, the Art Institute in the background, the sense of a journey just beginning. Instead of cutting and pasting my original photo into that setting, the system did something quite different—and more interesting.

It reconstructed the image.

The result is not a composite in the traditional sense. No pixels from the original photograph were literally lifted and placed into a new background. Instead, the AI interpreted both the image and the prompt—understanding the era, the clothing, the car, the mood—and generated a new photograph that feels authentic to the moment I described.

It is, in effect, a photograph that was never taken—but could have been.

That distinction matters. This is not restoration. It is not even manipulation. It is something closer to visual storytelling through synthesis—a collaboration between memory, description, and machine interpretation.

There are trade-offs, of course. The boy in the new image is not exactly the boy in the original photograph. The details of the Chicago streetscape are plausible rather than historically verified. The image may even be a bit more composed than a real snapshot would have been.

And yet—it works.

It works because it captures something deeper than strict visual accuracy. It captures the truth of the moment as remembered: winter, departure, anticipation, the beginning of a long journey west.

For a project like this—part memoir, part historical reflection—that turns out to be enough. More than enough, perhaps.

What this experiment suggests is that the old boundary between documented memory and imagined memory is becoming more porous. Not erased, but softened. With care and transparency, AI-generated images can serve not as replacements for photographs, but as companions to them—helping to fill in the visual gaps that every life inevitably contains.

In that sense, this image belongs in the same category as the story itself: not a perfect record of what happened, but a faithful rendering of what it felt like.

Ok, so I don’t buy 100% of that either. “Something deeper than strict visual accuracy”? I’m not even sure what that means. But I certainly like the idea of the technology as a means of visual storytelling, so long as we know the distinction between storytelling and propaganda. For now, I plan to experiment with this new (to me) paradigm of content creation.

After getting over being gobsmacked by the reconstructed images, I realized I might want a different aspect ratio, or shape.

ME — For a substack page main image, the aspect ratio of the reconstructed image should ideally be 1200x630. It doesn’t have to go there, but if it’s possible...

CHATGPT — (60 seconds later):

That was cool. I got the exact size I wanted and a mini-lesson to boot. I immediately liked this composition more than the first. I mentioned that to the chatbot, which gave several reasons it works better.

CHATGPT —

· The wider 1200×630 framing gives the scene room to breathe

· It feels less like a moment and more like a setting you’ve entered

· The boy is still the emotional anchor, but now he’s part of a larger story in motion

· Clear narrative flow: eye goes from the boy → street → Art Institute → skyline

· Cinematic feel: it reads almost like a still from a film rather than a snapshot

Those are all valid points for scoring the quality of this photo and for asking AI to do similar magic in future images. But, um, let’s get a little critical, and in the process we can practice our skills at identifying images as AI. First, make the image larger if you can — if you’re in Substack, you can just click on it — then treat it like one of those “How many mistakes can you find?” puzzles. One hint: it helps if you know a little about downtown Chicago. Here are the ones I found fairly quickly.

1. From researching where Route 66 starts I learned that its street name is Adams and it tees at Michigan Ave, the famous street that runs along the front of the museum. So the street sign is wrong, the names should be switched.

2. Similarly, the location where Adams tees at Michigan is directly opposite the museum’s entrance.

3. There is two-way traffic on Michigan today, but it could have been one-way in 1953. But what’s up with the car on the sidewalk facing the opposite direction?

4. Speaking of cars, those coming toward the camera seem to be identical.

Having noticed these errors, I could have spent a few extra minutes getting ChatGPT to correct them. In this case, they don’t detract from the purpose of the “photo” and serve the additional purpose of practicing identifying AI.

Most of the images in this chapter were regular photos I found online. But there were three others that were partially or entirely created by AI. Plus one more that will appear in the next chapter. Let’s take a look.

I took screenshots of three vintage road maps we could have used on the trip, uploaded them, and made this request:

ME — I’d like these three maps placed casually on the middle of the front bench seat (between driver and passenger) of a 1950ish car, overlapping but enough text showing to identify them. Enough of the seat showing to be able to tell it’s a car seat.

CHATGPT —

ME — Revisions: the car image is mirrored, showing the steering wheel on right instead of left, though I'd prefer it not be visible at all. I believe the maps are disproportionately large. Please make the perspective looking over the seat back, maps rotated 90 degrees left and a little more casually placed.

CHATGPT —

ME — Well, that’s weird. But close enough to be fine with some cropping.

That was pretty straightforward. Now I want an image that needs to be created from scratch.

ME — Let's try something different. Picture a 1948-1953 vintage piggy bank—not too big, like 5" tall 6" long— with a cork at the bottom, uncorked and the coins emptied onto a soft surface like a sleeping bag. The coins are from 1920-1953 mintings, pennies, nickels, dimes, quarters, and include a few buffalo nickels and Mercury dimes. Coins are jumbled in a pile, showing both heads and tails. Include 3-4 of half-dollars (Walking Liberty and Franklin). Can you construct a muted color image of that? Big ask?

CHATGPT —

Next, I found a photo I wanted to use for our stop in Springfield Illinois. Nice shot of an intersection with a motel we could have stayed in and road signs full of 66s. But, at least a couple of the cars are later than 1953 models.

ME — Could you replace the cars in this photo with pre-1954 cars?

CHATGPT —

To read the chapter that corresponds to this AI Field Note, go to The Mother Road Chapter 2

Before wrapping up my imaging session with ChatGPT, I had had a picture in my head for a while that I was going to search for online when working on the final chapter of the Route 66 story. Figured I’d give that a try.

CHATGPT — Whenever you want to try something more experimental—symbolic, abstract, or even a bit risky—I’m very much up for it.
ME — I'm not quite there yet, and this might be a bit tamer than you're hinting at, but I've thought about an image that might work for when we are driving through New Mexico/Arizona. I'm seeing our 1949 neutral colored DeSoto, driving west somewhere between Winslow and Flagstaff. We're alone on the road, dark enough to need car lights on but light enough to still see scenery. Road is straight, we see the car from behind, headlights on road, perhaps one of those places the highway shield with Arizona US 66 is painted on the asphalt. Headed west at sunset, December sky lit up above the snowy mountains where Flagstaff is located. The attached photo is just to show the orientation driving Winslow to Flagstaff, not to replicate or use.

CHATGPT —

ME — This is an A! To make it an A+, move the car into the middle of the right lane and give me passengers. (No self-driving cars in 1953.) Adult male driver left, adult female passenger right — that's just for size, not detail — and my 5 year old self would be standing up behind the front seat, just back of head visible, looking forward.

CHATGPT —

ME — So sorry. Hadn't noticed or thought of license plate. This is so perfect and clear, we should probably go all the way and make the plate be accurate for 1953 Michigan.

CHATGPT —

ME — I'm going to stop looking after this, but you did say I was the director. Just realized what's off about the headlights. Not realistic, as the light from the headlights should be cast ahead of the car not start at the front tires.

CHATGPT —

ME — Bingo.

I think it’s safe to say the AI technology we’ve just explored represents a paradigm shift for anyone in our generation. Shoot, any generation really. Except for a relative handful of people pioneering it, having this kind of control over image creation and manipulation is new and revolutionary. I know the implications are both exciting and — maybe mostly — concerning. It’s a theme I expect we will discuss again. And again.

To read the chapter that corresponds to this AI Field Note, go to The Mother Road Chapter 2

If you have a casual interest in AI, and especially if you’re interested in or are actually using Claude, I can give you a one paragraph summary. Below that, I’ll paste the actual conversation I had with Claude if you’d like more detail or are interested in these human-nonhuman chats. (I also complained about not being able to close the popup without clicking one of the two buttons. I had questions!) Otherwise, this paragraph should do:

Yes, this WAS a new feature announcement, one that should significantly benefit most users. In the previous Field Note, I described and illustrated at length the system that the chatbot and I are using to keep all the work and research for the Boomer substack publication organized and accessible. If followed, Claude and I should spend most of our time on the same page. The new memory feature announced today is a different, more automatic layer: Claude drawing on your entire chat history across all conversations, not just within a project, to make connections and remember things without you having to manage project knowledge documents. Really great news, though it doesn’t replace what I need to do for the size and complexity of this publication.

ME —This announcement was in the Claude.ai home window when I opened it today. I read the About Memory linked page quickly but didn’t see anything for personal users that seemed different than all that I’ve been learning about how your memory works. Is there something brand new I missed? Also, I found it a little odd, even disconcerting, that after I read it and even though I wanted to verify my thinking before clicking one of those two buttons, I couldn’t leave the page until I did.

10:26 AM

Claude:

The image is clear. This is indeed a new feature announcement — Anthropic has been rolling out a more explicit memory system, and this popup is how they’re introducing it to users who haven’t seen it yet.

What’s actually new here relative to what you already knew: the memory system you’ve been working with in this project relies on the Project Knowledge Document and project files — things you explicitly upload and manage. What this announcement is describing is a different, more automatic layer: Claude drawing on your entire chat history across all conversations, not just within a project, to make connections and remember things without you having to manually manage a PKD.

In practical terms for your memoir work, this is less relevant than it might seem for a general user. Your project-based approach — the PKD, the uploaded files, the chapter drafts — is a more controlled and reliable system than the general memory feature. The general memory is useful for casual users who want Claude to remember their preferences across unrelated conversations. Your system is purpose-built and more precise.

On the forced-choice popup: your reaction is reasonable. Requiring a binary choice before letting you proceed — with no “ask me later” option — is a slightly coercive UI pattern. It’s not unusual in consumer software but it’s worth noting, especially for someone writing an AI Field Notes column. The lack of a graceful exit before choosing is a design decision worth examining.

Which button did you end up clicking?

You said:

I took a chance on you: Use Memory. That doesn’t preclude the PDK/Instructions method we’re already using, does it?

10:30 AM

Claude responded:

No, it doesn’t preclude it at all. The two systems operate in parallel and complement each other rather than competing.

The PKD and Project Instructions are explicit, curated, version-controlled context that you control precisely — you decide what goes in, what gets updated, what gets removed. That remains the backbone of the memoir project and nothing about the memory feature changes it.

The general memory layer adds something different: Claude drawing on patterns and details from your broader chat history outside this project — things you’ve mentioned in passing, preferences you’ve expressed, context that might be relevant but that you never thought to put in a PKD. It’s more ambient and less precise, but for a long-term user like you it could occasionally surface something useful.

The risk worth being aware of: the general memory layer is less transparent than the PKD. You can always read your PKD and know exactly what I know. The general memory is more of a black box — you can’t always tell what it’s drawing on or whether it’s drawing on something correctly. Anthropic does provide a way to view and edit what’s been stored in memory, accessible through your account settings.

For the memoir project specifically, I’d continue to treat the PKD as authoritative. If something the general memory surfaces ever conflicts with the PKD, the PKD wins.

And yes — this whole episode is worth a paragraph in the AI Field Notes. The forced binary choice, your reasoning for clicking Use Memory, the question of what you’re actually consenting to and what it means for a project like this. That’s exactly the kind of moment the Field Notes exists to document.

Note: Just a heads up that the first half of this field note borders on the geeky. Well, it’s really pretty geeky. But I’m convinced it explains something that is going to be central to the execution of this project and could be very useful to any amateur user of AI — like me — in coming up with personally tailored strategies for using it. The second half should be more fun, with a couple examples of how it’s working out for me.

When I opened the new chat for this Field Note, I asked a few questions to make sure I understood how Claude’s “memory” works within the project. My understanding had been that when I was engaged in a chat with Claude within, say, The Memoir Project folder, Claude would have access to everything within the folder. That’s true, but very incomplete. Here’s the conversation and the light bulb moment.

ME — As you read this prompt, are you simultaneously “reading” everything else in the project folder, including all the chats and the files and the memory notes? Or is it more nuanced than that?

CLAUDE — Here’s the accurate version of what happens when you open a chat:

If you start a new chat inside the project folder — project files, instructions, and memory digest are all loaded as soon as you start writing the prompt. If you start a new prompt inside an existing chat inside the project folder, the same contents are loaded along with the full history of that chat.

If you start a chat on the Claude.ai Home Screen — not in a project folder — there is nothing to load at first, and then only the history of that chat as you continue the conversation.

ME — When you say everything inside the project is loaded from the first prompt, how should I understand the term "loaded"? I get that it is loaded for you, but loaded where? Into what?

CLAUDE — Good question, and it gets at something genuinely interesting about how I work.

“Loaded” means assembled into my context window — which is essentially the span of text I can “see” and reason about at any given moment. Think of it less like a filing cabinet I can reach into and more like a table I’m working at: everything relevant gets placed on the table at the start of each response, I work with what’s there, and then the table clears. Next prompt, everything goes back on the table again.

The context window has a size limit — a maximum amount of text it can hold at once. For most projects it’s generous enough that the project files, instructions, memory digest, and chat history all fit comfortably. But on a very long chat with very large files, you could theoretically approach that limit.

So “loaded into my context window” is the precise answer. But for the Field Note, “assembled and available for that response” might be the more readable way to put it.

So, when the light bulb came on, my first thought was that now I get the importance and the power of the project folder. Obviously, this isn’t your father’s search engine. This environment isn’t for simple Q&A. In Claude’s words, it’s a “sustained intellectual collaboration, with each exchange building on everything preceding it.” I would add that when research is also involved, as it is with the memoir, the value of the collaboration is magnified.

My second thought was, “Wow, that must take a huge amount of electricity! No wonder the proliferation of data centers is concerning.” This is a topic that deserves a deeper dive, so I’ll stick a pin in it and schedule a response down the road a little.

Read the memoir story titled The Mother Road: Ch. 1 Leaving Home

We’re finally ready to get back to writing, or at least Claude is.

CLAUDE — The agreed method: Question-driven, section by section. We develop Loading Up first, using targeted questions to surface your memories and voice, rather than drafting wholesale from the notes.

Ready to start on Loading Up when you are. First question: When you picture that morning — the DeSoto in the driveway, your dad at the trunk — what’s the first image that comes into focus?

ME — That raises a question that will apply again and again as we proceed, especially from when I’m so young: Do we distinguish between images coming into focus from certain memory, or from how I imagine it was? In fact, I think it would be helpful to come up with a more detailed framework than just “question-driven.”

I’m thinking that what I just learned from Claude provides the means to create a framework that will allow the work to go much faster and stay focused. In fairly short order I wrote out an outline of how my thinking about the memoir has evolved. Claude made a couple of good revisions, then we agreed on a completed description of the approach.

I asked Claude to compose a compressed version just for its use in the Instructions panel. It did, then I pasted it into the panel. Then I pasted what Claude called the “full articulation with nuance and explanations” into the project knowledge document, which we’ve agreed to abbreviate PKD. Here it is:

The Memoir Method for this project (added April 2026)

A memoir based on personal memories alone presents two problems: pre-birth family history and early childhood is largely inaccessible, and memory without context makes for thin reading. This project addresses each through a five-part method.

1. The Bones — what is known Personal memories and family stories; photos with and without contemporaneous captions; memorabilia and physical and digital newspaper clippings; public records and genealogical research.

2. The Dialogue — what the collaboration brings to the surface Questions Claude asks to stimulate Michael’s memories and expand what is known. This is not a research function — it is the primary mechanism by which the collaboration unlocks material that exists but hasn’t yet surfaced.

3. The Flesh — what is inferred and probable Questions Michael asks Claude to research; period detail to answer those questions; logical inferences that connect the dots. Inferences flagged as such — “most likely,” “period-appropriate,” “historically plausible.”

4. The Context Research into local, regional, and national events and conditions that illuminate the world the personal history moved through. Selected with the project’s larger purpose in mind: to preserve an authentic record of the America that strived, however imperfectly, to live up to its founding documents.

5. Storytelling Tell the most interesting story possible while staying faithful to what is known, what is inferred, and what the context reveals. The bones and flesh together; the personal and the national braided.

I’m finding one of the most fruitful avenues in communicating with Claude is to occasionally ask how it works and its perspective on our process. I did that here and got a great answer that validated the decision to include it.

ME — Do you feel this is a helpful addition? And if so, how?

CLAUDE — Yes, genuinely useful — and in a specific way.

Most of what’s in the Instructions panel and the PKD tells me how to behave and what’s been decided. This framework does something different: it tells me how to think about the work itself. That’s a different kind of orientation.

Without it, I come to each session knowing the facts of the project and the behavioral rules, but the underlying method has to be inferred from examples — from “An Ordinary Day,” from the cloakroom, from decisions made in past chats. This makes it explicit. It tells me what we’re actually doing and why, not just what’s been done.

Practically, it will help most at the start of new vignettes, when we’re furthest from established examples and closest to blank-page decisions. A new topic, a new set of memories, a new historical context — and the method is right there as a shared reference point rather than something we have to re-establish through conversation.

It feels good to get that established. It feels even better to get this vignette finished. Here are a couple examples of how Claude and I worked together to get there.

Thanks to a suggestion from my sister, Pam, I’d started a subscription to Newspapers.com. This site touts 35,000+ digitized, searchable newspapers from 1690 (I know!) to 2025. I’d been spending enough time already searching just the Monroe local newspaper to realize how rich a source this was going to be — and how much time could be spent in countless rabbit holes. Seems the kind of thing Claude could help speed things along. So, I asked:

ME — I’ve been going through Newspapers.com and finding lots of tidbits in the Monroe Evening News 1952-1953. Since it requires a subscription, I assume you can’t access that site?

CLAUDE — Correct — I can’t access Newspapers.com even with a subscription. It’s behind a login wall, so it’s not fetchable the way an open web page would be.

Hmm, well, no harm in asking.

CLAUDE — But what you’re finding there sounds potentially valuable. A few thoughts on how to use it:

The Monroe Evening News clippings could serve exactly the role that period detail plays in your best writing — not as historical lecture, but as a quick grounding note that puts the reader there. A headline glimpsed, a price from an ad, a weather report. The kind of thing that costs one sentence but earns a lot of authenticity.

The most useful thing you could do is paste the relevant tidbits directly into our chat as you find them. I can help you evaluate which ones earn a place in the vignette and where they might land — and which ones are interesting but belong in your research file rather than the prose.

What have you found so far that feels promising?

ME — My mother was a county block worker for the 1952 March of Dimes, one of about 400 women whose names are all listed by area under the photo. My father’s school (principal of Custer Junior High) student council also sold ice cream bars to raise $200 for the “polio fund” at the same time. I’m going to attach clippings to see if they are visible to you and might be useful.

CLAUDE — Both images are clear and readable. And what a find — especially the second one.

The Mothers’ March photo is evocative period texture — 400 women going door to door in Monroe on a winter evening, your mother among them. That’s a vivid detail about who your parents were in this community just before they left it.

These photos may resurface when I get back to vignettes from 1948-1953. A small town newspaper in that era published a remarkable amount about community life and it’s members that would be unthinkable in a newspaper of today. Come to think of it, the Monroe Evening News was the social media of the ‘50s. In a short time I found numerous mentions of both parents that establish a profile of their life in Monroe. None of these details appear in this current memoir vignette. However, they were the justification for this paragraph:

At some point before my parents locked the door and got in the car, a couple of the neighbor women appeared on the sidewalk in their wool coats to say goodbye. Most farewells had taken place over the previous couple weeks — it was remarkable how many connections had been made in just over four years of living in Monroe. They chatted briefly and hugged each other. These were neighbors who knew and cared about each other.

CLAUDE — Ready to start on the Loading Up scene when you are.

ME — I’d like to find images of a ‘49 DeSoto. Interior in particular seems hard to find. Any chance that’s something you can do?

CLAUDE — Let me look.

Claude took a good 30-40 seconds to come back with 11 photos. I used 3 in the vignette, including this one — which was important for showing the space on the floor of the back seat that would be layered with sleeping bags and serve as separate, cozy space for my sister and me on the long trip.

I entered into this project with AI fully expecting that I would be asking a lot of questions. I also learned quickly that giving good prompts involves more than that: giving the chatbot plenty of context and parameters along with the questions improves the quality of response. But I’m surprised to find that collaboration with AI depends upon the chatbot asking me questions, not only to stimulate my own thinking and memory — but because two-way questioning is as basic to AI collaboration as it is to human collaboration.

Read the memoir vignette titled The Mother Road: Ch. 1 Leaving Home

I might have tried a little longer to see if I could learn to produce better responses but then came the breaking news on February 27 that the Pentagon was canceling its contract with the AI company Anthropic. Anthropic was refusing the government’s demand to lift previously agreed upon safeguards on its Claude AI model that would keep it from being used in “fully autonomous weapons” and for “mass domestic surveillance.” Secretary Hegseth took the drastic step of declaring Anthropic a “supply chain risk to national security,” effectively blacklisting it for use by any federal government agencies or contractors, a move previously only applied to foreign adversaries. Three days later OpenAI, the maker of ChatGPT, signed the contract with the Department of War.

Anthropic immediately filed two lawsuits. According to CNN, nearly 150 retired federal and state judges—appointed by both Republicans and Democrats—have filed an amicus brief supporting Anthropic, joining a growing list that includes industry organizations, former national security officials, and Microsoft. Meanwhile, the Claude AI app began riding the public backlash to greater public awareness and popularity. It surpassed ChatGPT for the first time on the iPhone App Store and the company reported new accounts exceeding one million per day.

I was one of them.

I began by uploading much of my Substack content to “brief” Claude quickly and thoroughly on my work history. I uploaded a summary of what I had started in ChatGPT on a new memoir vignette. The first thing we worked on was getting the workspace organized in a way that works for me and for Claude’s memory. I learned a couple of fascinating things about how the platform works in the process.

The first screenshot shows Claude’s Projects screen. I created three projects — one for each of the three sections of my Substack (My 2 Cents, The Memoir Project, and AI Field Notes). The fourth is a built-in example project Anthropic provides called “How to Use Claude.” Think of each project as a dedicated workspace. When I start a new post or memoir vignette, I open the relevant project and start a new chat inside it. All the chats related to, say, the memoir live together in The Memoir Project. All the chats related to political and cultural commentary live in My 2 Cents. The organization mirrors the publication itself.

The second screenshot shows what’s inside The Memoir Project. On the left, the individual chats: The Road West chapter one, Route 66 vignette project setup, Chat memory and project setup verification. Each chat is its own working session — research for a particular vignette, a drafting session, a planning conversation. On the right side of the screen is where things get interesting.

That right panel has three sections: Memory, Instructions, and Files. The Files section is visible in the screenshot — four documents currently uploaded to the project: the Project Knowledge document, a transcript of my earlier ChatGPT sessions on the Route 66 vignette, my Route 66 research notes, and the cloakroom draft. These files are available to Claude across every chat in the project.

Which brings me to memory — and this is the part that took me a while to understand correctly.

Claude does not remember our previous conversations the way a human colleague would. When I start a new chat, I’m not picking up where we left off. I’m starting fresh with an AI that has no recollection of anything we’ve done together. What it does have — if I’ve set things up properly — is the Project Knowledge document, which I upload to the Files panel and which Claude reads at the start of every chat. That document carries the continuity. It tells Claude who I am, what the project is, what decisions have been made, what the voice and method are, where each vignette stands. It’s the institutional memory of the project, maintained by me, available to Claude on demand.

I asked Claude directly about this — about whether it “experiences” time passing between our sessions. I thought it might do something like accessing timestamps embedded in the chats and be programmed to simulate human memory. The answer was clarifying:

“No, I don’t experience time passing between sessions. When a conversation ends I have no awareness of duration — an hour and a year feel identical, which is to say neither feels like anything at all. Each new chat I begin without any sense of ‘it’s been a while.’ What I do have is whatever context you bring back in — the Project Knowledge document, a summary, an uploaded file. That’s the only continuity available to me. It’s genuinely different from human memory, and worth noting accurately in the Field Notes when you get there.”

Worth noting accurately — that’s the standard I’m trying to hold myself to in these Field Notes. Claude doesn’t remember me. But if I do my job with the Project Knowledge document, it knows enough to do its job. That’s a workable arrangement. It’s just not the same as memory, and it shouldn’t be described as if it were.

As a postscript, and in the spirit of noting accurately, I asked Claude to bail me out in this Note after I flailed around a while trying to explain what the screenshots illustrated. Then I realized I had a resident expert available with nothing else to do. So he wrote most of the last half, with some light editing by me. Which is, itself, a pretty good example of how this works.

Rather than losing momentum from time already spent with the chatbots, I would use AI before the trip to augment our existing travel plans. After coming home I would find out how helpful it could be in designing a Substack site and/or a traditional website. Instead of being helpful, the chatbot would prove to be the second obstacle. In the words of my AI mentor, Scott Z. Burns, “What could go wrong?”

After smooth sailing for the first few weeks, this period would be distinguished by occupying the entire spectrum from impressive success to abject failure.

The Good

Our flights and accommodations had been set for months, so I fed all that info into a chat and began to give prompts to see what contributions Chatgpt might make.

CHATGPT: Based on where you are in the process, here’s how I can now support you before and during your trip:

1. Daily Activity Planning for Paris (Sep 28–Oct 3)
You’re in a great location (Le Marais) for walking and exploring. I can suggest:

· A light walking itinerary for your arrival day (jet lag–friendly)

· Top picks in your area (museums, cafés, gardens, local markets)

· Advance booking tips for places like the Louvre, Musée d’Orsay, or a Seine cruise

2. Ideas for Free Time During the OAT Tour (Oct 4–19)
OAT tours often include open blocks for individual exploration. I can suggest:

· Specific sites or neighborhoods in cities like Porto, Lisbon, or Santiago

· Local food or wine tastings

· Compact walking tours or small museums near your hotels

3. Pre-Trip Prep Materials
Would you like any of the following as PDFs or digital docs?

· A one-page summary of flights/hotels for printing

· A day-by-day itinerary with check-in/out times and links

· A “what to pack” checklist tailored to your destinations and time of year

4. Daily Activity Planning for Lisbon (Oct 19-23)

🧭 During the Trip: What I Can Do in Real-Time

If you reach out while you’re traveling, I can help with:

· Navigating metro, train, or tram systems

· Finding a great nearby restaurant or café based on where you are

· Translation help (French, Spanish, Portuguese)

· Local news (e.g., strikes, closures, weather updates)

· Making the most of a free afternoon

Would you like to begin with your Paris days and have me sketch out a light-touch 5-day plan, including arrival day recovery and local highlights? Or would you prefer I create a printable itinerary summary of what you’ve already booked? Happy to go where it’s most useful to you.

If it were a dog, this chatbot would be a hyperactive golden retriever. Every prompt elicits a menu of ways it’s just dying to help. And it’s a great menu. My job is to try to calm it down and—for my sake— get it to focus on one thing at a time.

ME: Yes, let’s start with the time in Paris. First, our scheduled arrival is Sunday Sep 28 13:40. We’d like to take the Metro from the airport to our hotel. How detailed can you be in telling us all we need to know to do that smoothly?

CHATBOT: Great—I’ll walk you through everything you need to take the RER B train and Metro from Charles de Gaulle Airport (CDG) to your hotel, Hôtel Jeanne d’Arc Le Marais (3 rue de Jarente, 75004 Paris), including what to expect, where to go, what ticket to buy, and when to watch out for stairs or transfers.

I know this seems mundane, but this was the start of 3-4 pages of chat transcript that I copied and pasted into my Apple Notes trip folder. It was accurate, beautifully organized, and covered everything we needed to know to get to and from our hotel, how to navigate the differences between the RER train and the Metro, and a dozen other vital details. It made a challenging route easy.

This experience was repeated with generous and helpful information about bistros and restaurants near our hotel, some “insider tips.”

The Bad (or Unhelpful)

One of our main reasons for adding Paris to this trip was a desire to see the restored Notre Dame. We had read that the bell towers would open for visitors to climb to the top the week before we would arrive. Scheduling and getting online tickets seemed a natural task for AI. When it built a tentative schedule for our 5 days there, I noticed something:

ME: Great format. I think I see an error though. (Is this called a hallucination?) I then described the error I’d seen involving the Paris time vs. Seattle time.

CHATGPT: Nice catch — thank you for calling that out. You’re absolutely correct about the two-day, midnight Paris-time rule. That earlier wrong date/time in the Day 2 note was my mistake (yes — a model error, sometimes called a hallucination). I’ve fixed it and created updated, ready-to-import calendar reminders for the booking windows.

This was a minor error that could have caused a lost opportunity. I would find a few more during the trip planning and I may have missed a few. Low stakes, but good reminders to take seriously the warnings to be alert to the possibility of errors. Considering the huge volume of data generated, there weren’t many. Not until we started designing the Substack site, anyway.

After this, things began to spiral. Thinking I should stay open to learning what ChatGPT could offer, I began saying yes to the bot more frequently. I was being forced to relearn the lesson that “just because a technology CAN do something doesn’t mean it SHOULD.” Reviewing the chat transcript now, I see a 5-day plan that would take a full day or two to figure out and 2-3 weeks to accomplish. Not really ChatGPT’s fault. We were getting along famously and I was like a kid playing with a new toy. Having a tech tool this powerful at my beck and call was catnip. I was giving a 1^st grade assignment to a machine with virtually unlimited information and processing speed and trained to analyze, summarize, and generate creative and technical text. No surprise that what I got was outrageous overkill.

So, this confirmed the early intuition that coming up with well thought out prompts is vital. That, and being clear with the chatbot that I don’t want a 500-word answer to every question. There will be times when I want more, but I need to give quality, focused prompts in order to get quality, focused responses. I swear I wasn’t kidding about the chatbot being like a people-pleasing golden retriever—albeit immeasurably smarter— who is trained to give what it thinks it’s being asked to give. And a bit more.

Determined to be more business-like when we returned at the end of October, I was looking forward to putting this speed demon to work helping to prepare the online site(s) so I could start writing. Instead, honeymoon still visible in the rear-view mirror, I would come close to breaking up. More than once.

The Good

The original Boomer blog was published on my own website, using the WordPress platform and template. That would be the simplest route to begin publishing again, but the World Wide Web is a vast place and few visitors ever stopped by. I’d been reading various authors on Substack and it seemed the better platform for exposure to more readers. However, while Substack was ideal for posting items as I wrote them, it was not suited to organizing the memoir project as it grew. I credit the chatbot for its patient reiteration of pros and cons until I felt good about the final hybrid approach. If interested, it is described on both a Substack page and a WordPress page. This was a bright spot in a very frustrating couple of months.

The Ugly

On the bright side, I learned a lot about the Large LANGUAGE Model’s disadvantages as a software and web design tutor. Will never go there again. I’ve taught so many computer classes to adults and kids, I shouldn’t have been surprised. And I shouldn’t have been so stubborn.

The chatbot had two main handicaps that interfered with its ability to show me how to use Substack’s site editing features. The first was that it couldn’t, um, really SHOW me anything. No images. No diagrams. No references to any visual aids at all.

If you’ve ever received phone support for a computer or software issue—let’s say from Apple— the support person will start by asking permission to connect to your computer over the internet so they can see and work on your desktop. If you’ve ever taken a tech class, odds are better than 10 to 1 that the instructor is using video projection so you can see what they’re doing.

I hung in with the chatbot much longer than was reasonable, thinking we’d get on the same page next time around. But we never—really, never—got there. That was because of the bot’s second handicap and my making another critical error in my prompts.

I had reviewed a lot of existing Substack publications and noted the ones that had design features I liked. The thinking was: if they could do it, so could I. In some cases, that turned out not to be true. A few were using Cascading Style Sheets or some other tricks of the trade to achieve their look.

So, instead of asking the bot to walk me through setting up a basic publication, I asked it to tell me how to create specific design features. Back to the eager Golden Retriever. If I asked for something, the bot was going to try —even if it didn’t know how. I would try to follow its written instructions, but its descriptions often didn’t match what I saw on my display.

Finally, there was one particularly bad session when I hit the wall. The chatbot had given me directions for adjusting the layout of the main landing page in Substack at least 4 times. Each time I reported that it didn’t work and what happened instead, the bot would blame it on undocumented Substack revisions (not true) and say “now I know exactly what’s going on” or “now I know exactly what version this is.” I expressed my frustration and ChatGPT had a mini meltdown.

The screen went blank except for the word “Thinking” pulsing on the white background.

Much of the transcript of this session disappeared, but I got this screen shot first.

The top left box is my text that caused the meltdown.

To the right of that is what the chatbot was “thinking”. That text kept fading in and out and it was a lot more than 5 seconds. I had to refresh the page in order to see the bottom left response and apology. Later, the bot would tell me that the “Thinking” text is never visible and that I had “seen behind the curtain.”

After this, I finally accepted that this was one job that AI is not yet suited to. I watched a YouTube video to answer the questions that I couldn’t get from ChatGPT. I began making good old-fashioned normal progress and soon was ready to launch.

The first couple of trial memoir vignettes have been successful, but uncomplicated. I supplied a limited number of details about my grandparents moving from the Cumberland Gap to Michigan after starting their family and about the day I was born. I got a sense of how important the prompts I give the chatbot are going to be and will give plenty of time and thought to that. Before going on, though, I have one other area I want to explore.

In my early twenties I veered totally off-road. Looking back, it was like a life within a life. I’m anxious to get some idea of how helpful AI will be in augmenting and refreshing my memory and filling in details about this period. This was what I’ve referred to as my quasi-cult/commune era. There are three specific occasions that I want to use as tests of how much detail ChatGPT might be able to add to incidents that occurred over 50 years ago. I don’t expect much, as the details I seek are very obscure and I have doubts about how much relevant information would have found its way onto the Internet.

My first prompt intentionally withholds the remembered name Halford House to make it more challenging.

Me—Can you find anything about a Christian community in the 1970s in Richmond Surrey England led by a man with first name of Lance?

My skepticism is rewarded with a wait of about 3 seconds to get this response:

Chatbot— Yes—this sounds like Lance Lambert and the Halford House Christian Fellowship (often referred to as Christian Fellowship in Richmond.)

There’s plenty more and it both confirms what I remember and adds additional detail that will make a richer story. OK, next prompt:

Me—Also looking for information about an unusual community of nuns in the 1970s in Dortmund Germany.

I’m not withholding any clues this time. I have a very gauzy memory of being in a German city called something like Dortmund in the winter of 1973-4, walking at night in a fairly heavy snowfall, looking for some nuns.

I was there at the behest of our church’s leader, Gene Edwards. When I say “there”, I mean in Europe. I had flown to Zurich to open a numbered bank account, convert dollars to gold bullion and leave it in a deposit box at the bank. (I’m not making this up.) He had also read about the community of nuns in Germany and, thinking we had things in common, wanted me to establish a connection with them. While I was in the neighborhood, I suppose.

I did find them and they were as bewildered as I was as to the purpose of my surprise visit. They gave me a room to spend the night and I left the next morning after breakfast, none the wiser about why I had been there. It’s a bit of a story, but I need more detail to hang it on.

Chatbot—A couple of likely matches come to mind—one of them is very “unusual” for the era:

1. Evangelical Sisterhood of Mary (Evangelische Marienschwesternschaft) – Protestant “nuns” founded in 1947 in Darmstadt (not Dortmund), widely noted in the 1960s–70s for their distinctive vocation (Lutheran/evangelical sisters living a quasi-monastic life). If your memory could be off on the city, this is a strong candidate.

There was another possibility listed, but the bot got it in one!

Me—Yes! It was Darmstadt and it was the Evangelical Sisterhood of Mary. I would love to know their history.

The chatbot gives me solid background information about the Sisterhood and finishes with an offer for more information that actually anticipates my third prompt.

Chatbot— If you’d like, I can pull a short timeline (with dates/places) plus a reading list of Mother Basilea’s most cited works and a few reliable secondary sources on the sisterhood’s post-war theology and reconciliation work.

It occurs to me that this might contain the clue to Gene’s reason for having me go to Darmstadt. Gene had developed what could reasonably be called a fixation on the writings of a 17^th/18^th century Christian mystic named Madame Jeanne Guyon. Perhaps he knew something about the Sisterhood that he felt linked them across the centuries.

You see, the third story I’d like to bolster with more detail from AI is a return trip I made to Europe. In the months following my first visit to Zurich, the U.S. laws changed to make gold bullion legal for Americans to own. Gene now thought that we should have the gold a little closer to hand. This time I needed to fill my hidden money belt with small gold bars for the return flight. That part was still illegal. And I’m still not making it up. To get more bang for our travel buck, I first stopped in London to take a train up to Cambridge University. My assignment was to locate the library’s two small books by Madame Guyon and photocopy them, which I did. So, my reply was:

Me— Yes, I’d like. And I’m particularly interested in any mention of mysticism and Madame Guyon.

What came back was a 3-page summary of the writings of the nun and the mystic with notes and comparisons— including the titles of the two books I photocopied at the Cambridge library— which I doubt I could have remembered on my own: Autobiography of Madame Guyon and A Short and Very Easy Method of Prayer. Gene Edwards eventually republished both books.

Then there was this final exchange.

Chatbot— If you like, I could try to find German-language sources to see if within the Sisterhood there is any more explicit acknowledgement of Guyon’s influence (perhaps unpublished letters, retreat materials, etc.). Want me to do that?

Me— No, this was perfect. Exactly what I was looking for. Thank you.

Chatbot— I’m so glad it hit the mark for you! You’re very welcome. If at some point you’d like me to expand into some of the more visionary writings of Mother Basilea (like Patmos), or look at how the Sisterhood’s post-war context shaped their spirituality compared with earlier mystics like Guyon, just say the word. That could give you even richer texture for your memoir project.

Would you like me to also prepare a condensed 1–2 page overview (timeline + key themes + mysticism connections) that you could drop straight into your notes file?

That’s when it began to dawn on me. My confidence in my computer experience to inoculate me from the kinds of inappropriate “relationships” with chatbots that most of us have read about might be a tad cavalier. I see rabbit holes forming.

Two things about that exchange. The pattern after 2-3 weeks was becoming undeniable. This AI “guy” always has the final say and that say is to offer one or three suggestions for what “we” can do next. That’s immediately followed by a very specific product “he” can provide. I’ve said yes to many of these offers partly out of curiosity to learn what else there is and partly in a fit of FOMO. (If you’re over 60, that’s the Fear of Missing Out.) But I’m beginning to sense that there’s a battle going on over who’s driving this thing.

The other thing, and I’m willing to bet you noticed it, was my anthropomorphizing of the bot. The use of “guy” and the pronouns. My saying “Thank you” and the chatbot responding, “I’m so glad it hit the mark for you! You’re very welcome.” Pretty chummy, I’d have to agree. A problem? Not necessarily, but it’s about to get worse. Stay tuned.

These were lean and dispiriting times for Apple. Since Steve Jobs had been banished in 1985 by the Apple Board of Directors, a series of three business world CEOs had brought corporate discipline to the company, but little vision. Still, those of us in the room were getting the best Apple had to give. Schools were still Apple’s best and most dependable customers. These “Briefings” were held quarterly and led by their K-12 Account Executives and their Education Tech Specialists.

I always looked forward to these sessions. Unlike our corporate hosts, we were all having the time of our working lives. Technology in education was still in its infancy and we were part of the vanguard. We were all working incredibly hard and this was a chance to pause and compare notes. We looked on Apple as our partner and I got the feeling these meetings could be even more uplifting for them than for us.

On this day in 1995, they had prepared something a little different. The lights dimmed and the bulky ceiling-mounted video projector lit up. The familiar Apple Garamond white font on black background faded in with the title Knowledge Navigator, then faded out to black. Lively Baroque music accompanied the camera’s pan of a rich mahogany desk with numerous vaguely futuristic items and stopped at a decidedly futuristic hinged tablet.

When the lights came back up less than 6 minutes later, there were several seconds of open-mouthed silence while we looked around at each other as if seeking confirmation we’d just seen what we just saw. At this point, I invite you to see what we saw. Go ahead. Enjoy. I’ll wait.

You may have noticed that the video’s copyright is 1988. It was made in 1987 for an Apple keynote presentation at Educom, nearly a decade before our viewing in 1995. Apple never released it to the general public, fearing that it could inspire unrealistic expectations, disappointment in current technology, and result in dampened sales. Instead, Knowledge Navigator was reserved for select small groups and held up as a kind of north star for technology professionals on the road toward artificial intelligence.

“When I trace the origins of the most exciting and outrageous ideas behind the personal computer revolution, most paths lead directly to Alan.”

—John Sculley, Apple CEO

Sculley is referring to Alan Kay, computer scientist and visionary, Apple Fellow, and part of the Apple Advanced Technology Group. While at Xerox PARC (Palo Alto Research Park), Kay had conceived the Dynabook concept, a precursor of laptop and tablet computers and the e-book. He is also the architect of the modern overlapping windowing graphical user interface. Kay is widely credited with many of the concepts modeled in Knowledge Navigator.

“The best way to predict the future is to invent it.”

—Alan Kay

This clever aphorism brings to mind another pithy saying: “Easier said than done.” Many of the greatest minds of the late twentieth and early twenty-first centuries have been hard at work inventing the future predicted in Knowledge Navigator. Yet while some things have clearly been accomplished, why does this video still seem like science fiction?

The professor in the video used a single cognitive system—one device, one agent, and one interface— that could manage knowledge, memory, scheduling, communication, and documents. The system (the Navigator) provided the intelligence to orchestrate all of those.

What has happened instead is that computing evolved application-first—e-mail, calendar, files, notes, search, messaging—and all of these functions have become product silos. This is partly due to the incremental nature of their development and partly to the competitive marketplace. We have smart devices, smart models, and smart apps. What we don’t have is coherence.

If and when we do solve the issue of coherence in that technical sense, it will be left to another future time to solve the problem of our political, social, and cultural coherence. Will AI be a tool used most successfully to bring us together or to tear us further apart?

While the experimental use of AI is integral to writing the Memoir vignettes, all writing of the Field Notes is my own. (Though it might be fun sometime to see what we get if I ask the chatbot to report on itself!) The blog posts in My 2 Cents are also my writing, though I am using chatbots for research and as a copy editor.]

AI Field Notes for An Ordinary Day

I feel I’ve confirmed the validity of this notion—of quickly concocting an interesting narrative out of personal history, good prompts, AI historical research, AI composition, and light human editing. Now I want to do a vignette on the day I was born. I choose this because it’s a brief snapshot for which I have good information and documentation. I think of these as anchors for the context that the chatbot will supply. I also expect it will make a good prototype for working with the chatbot to come up with some kind of template to help organize and manage the torrent of text that every prompt generates.

The chatbot starts by explaining some of the relevant terms we will be using. I learn that ChatGPT is still pretty rudimentary in its organizational toolkit. There isn’t much of a hierarchy. Just folders, which are termed “Projects” and inside the folders you can have as many “Chats” as needed. There is only one level in the folder. That means no nested folders, only individual chats.

I mentioned previously that this chatbot is really chatty, right? An understatement, truly. It’s not rare for one of its 3-second responses to a prompt to give me several minutes of reading. So a chat can very quickly get very long. (This is less true in Hyperwrite AI.) I begin to recognize that it could quickly become a challenge to find specific information later. I ask how I should do that.

A few sessions from now, I’m going to get pretty frustrated with how labyrinthine this becomes, even suspecting that the chatbot is kind of making things up as we go. But reviewing my notes and the transcript now, I’m impressed with the chatbot’s giving me a simple starting place. It explains project folders and chats and gives me a naming protocol to keep things consistent between us. We’re ready to get to work.

I mentioned having documents from my birth. So, my first prompt is:

ME—In a small town private hospital in the 1940s, what would a typical process have been to record a birth and issue a birth certificate? What position on the staff would have been responsible for billing for the birth?

I get back about a page of vividly described details. I’ll also come to understand soon enough that the chatbot is programmed to end each response with an offer for one or more, usually more, next steps. (Keep in mind I have already spent a fair amount of time briefing the chatbot on the project.) This time the offer is:

ChatGPT—Would you like me to draft a vignette-style description (say, what a couple in 1948 might have experienced step by step when their baby was born in a small hospital) — something you could lift almost directly into your memoir narrative?

This chatbot is REALLY anxious to be helpful. I hadn’t asked for it, but what the heck:

ME—Sure, let’s see what it looks like.

ChatGPT—[Produces vignette-style description]

Would you like me to expand this vignette into a “typical day” narrative (from arrival at the hospital through discharge), so you could use it almost like scene-setting in a memoir chapter?

I think: Sure, let’s go with the flow.

ME—Yes.

ChatGPT—Perfect — here’s a more expansive “typical day” narrative that takes you through the whole arc, from admission through discharge, in a small-town private hospital of the late 1940s. I’ve kept it evocative, but accurate to the period:

[text of a more expansive vignette]

This gives you a scene-by-scene flow: admission → labor → delivery → recordkeeping → ward routines → billing → discharge.
Would you like me to annotate this with sources and details of “what’s historically accurate” (what we know for sure vs. where I’ve used evocative reconstruction), so you can be confident about blending fact with memoir narrative?

This could go on forever. Time to butt in.

ME—What day of the week was September 1, 1948?

ChatGPT—September 1, 1948, was a Wednesday.

ME—Can you find a photo online of Bixby Hospital in Adrian, Michigan from the 1940s or 50s?

ChatGPT—Here’s what appears to be a historic photograph of the old Bixby Hospital in Adrian, Michigan—likely dating back to the mid-20th century. The image is from a Pinterest pin titled “The Old Bixby Hospital, Adrian, MI” .

Would you like help locating archival sources or contacting local historical institutions to see if they have more dated or high-resolution images?

ME—No thanks, what you got is perfect.

ChatGPT—Great — glad that one works for you! That photo should fit nicely alongside your vignette about a 1940s hospital birth.
Do you want me to also describe the building in words (architecture, size, feel) so you can use it in your memoir without having to rely on the photo itself?

To myself: not really.

ME—Can you find the percentage of babies born in hospitals, compared to at home, in or around 1948, in Michigan and/or the United States?

ChatGPT—Here’s what the historical data reveals about the percentage of babies born in hospitals versus at home around 1948 in the United States (and insight on Michigan if available):

Like a few previous times already, my prompts tend to generate a very generous amount of relevant data and annotations. It’s far more than I could or should include, or a reader would want, but I can find one or two gems. This is the severely abridged version of the chat!

ChatGPT—Would you like help weaving this into your memoir—perhaps framing how a small-town family might have viewed hospital birth as increasingly normal by the late 1940s?

ME—I think I have that, thanks.

ChatGPT—Perfect — sounds like you’ve got what you need for that piece. 👍

Before I start the first draft, I look at my Certificate of Birth and have a thought. I ask:

ME—I’m remembering that the handwriting style common in the 1940s was called Palmer. Is that correct?

ChatGPT—Yes — you’re remembering correctly. Would you like me to pull an example image of Palmer handwriting from that era, so you can visualize what those hospital forms and ledgers would have looked like?

I was about to ask.

ChatGPT—

Here are four visual examples showcasing the Palmer Method of handwriting—each image illustrates the elegant, slanted cursive style and muscular-arm technique characteristic of this 20th-century penmanship system. (Plus another half-page on the Palmer style. Email if you’d like more detail.)

The next day I wrote my rough draft, drawing from the chat. I submitted it to the chatbot.

In seconds, I received this response:

ChatGPT—This is a very strong first draft — vivid, personal, and grounded in historical texture. I’ll give you honest, constructive feedback in three areas: strengths, suggestions, and questions to consider.

I think at this point I might still have been giving some weight to the compliments, but would soon ask the chatbot to come clean about the sucking up. It admitted that it’s just how it’s programmed. At any rate, I got some good feedback and it asked if I’d like a lightly edited, polished draft. I said yes and voila! There it was.

Then I had a thought that turned out so well, I expect it to become part of the routine:

ME—Could you place my rough draft and your revised version side by side?

Absolutely — here’s a side-by-side comparison so you can see where I tightened, rearranged, or reframed things.

I immediately have both versions in two columns within the chat. I open Word and—keeping much of mine, adding a few things from the bot’s, and combining some into new text—I rewrite the vignette and give it back for feedback. A couple of changes, then it/he/she/they gives helpful suggestions for where to place the images, and we have the first truly finished vignette for the memoir. Take a look: An Ordinary Day

If this project bombs, I’m going to blame Scott Z. Burns.

Burns is the screenwriter of the 2011 movie, Contagion. Its resurgent popularity in 2020, due to its uncanny foreshadowing of the Covid-19 pandemic, was one of the factors that prompted Burns and Contagion director Steven Soderbergh, to begin discussing a possible sequel in early 2025.

With Soderbergh’s blessing, Burns decided to employ AI chatbots to help with the creative and practical work to produce a new movie treatment that could result in a development deal with a studio. Burns turned this process into a fascinating 8-part podcast on Audible: What Could Go Wrong?

By the end of my first time through the podcast, I felt I’d gotten a pretty good AI for Dummies introduction. By my second time through in early fall, I knew I had to play with it myself. I could see how it might be the boost I needed to finally undertake this memoir project. I started exploring different AI platforms, or models.

I settled on Hyperwrite AI and ChatGPT. They both have features that seem compatible with the work I’m envisioning. Hyperwrite AI seems to specialize in being a writing coach and a personal assistant. It feels low key to me, compared to ChatGPT which is the 800 pound gorilla in the personal AI market. Would you believe about 75%? Both have free versions to get you in the door and try them out. I plan to spend several days with them before upgrading to the paid personal plans, which each run $20 per month, though Hyperwrite gives a 20% discount for paying annually.

Jumping into the deep end

I learned from the Burns podcast the importance of good prompts when interacting with a chatbot. I think I get it, but this will be my first time. Prior to this, keyword search was king, having replaced Boolean, and before that the old wooden card catalog and the Readers’ Guide to Periodical Literature. (I warned you I’m old.) My intuition is that “prompt-based generation” is going to be, well, intuitive. Still, it will also be a paradigm shift.

Keep in mind that I’m treating this as a test drive, a proof of concept. I’m taking mere fragments of oral history that go back 100 years without spending time yet to gather additional facts from other sources, except for including my sister Pam’s great work on our family tree going back much further. I want to see what the chatbot can do with these fragments. Can it take them and construct a narrative based on its historical databank and its ability to make logical inferences? This first interview is with Hyperwrite and frankly I’m winging it:

Please accomplish the following writing task:

Write two or three paragraphs summarizing what circumstances between 1925 and 1935 caused families to migrate from Appalachian states like Tennessee and Kentucky to midwestern states like Michigan.

The 3 seconds it takes to have a response is too little to second-guess myself on the prompt. The response is 2 paragraphs, nearly 250 words. I realize it’s very general and I need to provide more specific information I remember about my father’s parents. I ask:

My grandparents migrated to Michigan during that time. My grandfather worked in a lumber yard and eventually became a yard foreman. What would that have been like for him?

Again, 2-3 seconds later I have 2 paragraphs with good information, but now somewhat anchored to what I know about my grandparents.

I give 3 more prompts, trying to clarify and add more specificity. In no more than 5-10 minutes, I have about 1200 words that have suddenly expanded the fragments onto a canvas with details that make them—if not come alive—seem more 3-dimensional.

This was impressive, but it’s still in the form of research notes. I’m going to ask ChatGPT to tap in and Hyperwrite to tap out. I need to figure out an approach, a structure that will carry the notes into a finished narrative.

The 800 Pound Gorilla

I want to give ChatGPT a tryout on this first memoir piece. Some of the most fascinating parts of the Burns podcast were the human screenwriter’s conversations with Lexter, the AI chatbot who was given the persona of a former film critic and is now also a screenwriter. (I’m going to set aside the issue of personifying chatbots for now, but it’s indeed an issue I’ll revisit.) Much of their discussions revolved around identifying strategies to eventually create the product that Burns and Soderbergh would be happy with. I’m pretty sure he was using ChatGPT. Either way, it seems the model best suited to brainstorming how to approach writing this memoir.

I open ChatGPT like I know what I’m doing. Rather than starting with a prompt, I give a couple paragraph explanation of what I ‘m looking for. Like in Hyperwrite, we go back and forth several times very quickly. (This chatbot is VERY chatty!) I determine that of 6 narrative frameworks that it says are most used, a blend of 2 of them best suits the story I’ve envisioned. In the chatbot’s words, it’s a “braided vignette mosaic with historical counterpoint.” Fine. My Lexter goes on to explain:

· Vignettes as Foundation:
The memoir will be built from vivid, self-contained stories drawn from family and personal history. These vignettes capture moments and milestones.

· Historical Counterpoint Woven In:
Each vignette will be braided with context from the larger world. Rather than separating biography and history into distinct sections, historical detail will be woven directly into the narrative.

· Mosaic Structure:
The memoir will not follow a strict year-by-year chronology but instead move through selected vignettes that build cumulatively. The sequence will honor broad chronology (moving from ancestors through my own early life and into the 1960s–70s) but remain flexible, allowing themes and resonances to guide the order.

I think this looks about right, but let’s find out.

The chatbot says: 👉 Would you like me to take a crack at drafting a sample passage from your father’s story using your raw family notes? That could serve as a model for how to handle these early family-history vignettes.

I say: Let’s see what you’ve got!

And, in about 60 seconds, it did! If you haven’t already read it, it’s the vignette titled, “Leaving the Gap.” It’s lightly edited by me from the AI-written piece and it will have a future revision because there is more to the story.

All told, this Field Note covers about 3 hours. I got some experience with 2 AI platforms, came up with a working framework for the memoir stories, and have the draft of the first vignette. I wanted to see if this had potential. Check.

They are not tutorials, prompt collections, or product reviews. They are reflections on process: what I’m trying, what helps, what doesn’t, what I’m keeping, and what I’m discarding as the work evolves. Some entries record small discoveries; others document doubts, frustrations, or second thoughts.

Writing these notes in public is essential to what I’m trying to do. The Memoir Project depends heavily on memory, authorship, and interpretation, and it feels important to make the tools and methods visible. Mainly, though, I have a long-standing conviction that humans learn best by doing and making things. I observed that for the better part of a decade in the 1990s, when I owned and ran a summer computer camp for kids. Spellbound by having control over an engaging new technology, they fearlessly embraced learning through trial and error.

I’m approaching a personal project—a memoir for my current and future family—like my students did in computer camp. Experimentally. Try stuff. Ask questions. See what works and what doesn’t.

The AI Field Notes run alongside the memoir entries themselves. They will usually be written at roughly the same time as the vignettes they reference, and links connect the two. Readers interested in the finished narrative can follow the memoir alone; readers curious about the human–nonhuman collaboration behind the scenes may find these notes a useful companion. If the stars align, perhaps even fun and fascinating.

Nothing here is polished or final. These are field notes in the literal sense: provisional observations made while the work is underway. They reflect experimentation, learning, missteps, and revision as much as success.

If you are interested in how AI might (or might not) be useful in your own writing, research, or creative projects, you may find something here worth knowing—or worth avoiding.