I’ve spent roughly 7,000 hours working with AI since ChatGPT launched. You’d think that would mean I know what I’m doing. Then I asked ChatGPT to remove a few words from a diagram.

To me, the task was almost deceptively simple. Preserve a diagram, but remove some text from what were meant to be blank placeholders. I knew I needed to be specific, so I wrote a detailed prompt and hit Enter. What I got back was what I described to ChatGPT as a “big blank.”

I argued back and forth with the tool for far too long. I could have made the changes in Canva in a fraction of the time. And in the end, I still never got it quite right.

I was frustrated about wasting so much time, but mostly I was confused. After thousands of hours working with generative AI, why couldn’t I get this one simple task right?

Did the model change? Was I forgetting some brilliant prompting technique? Do I really just not understand AI?

The gap between use and understanding

What I realized is that using AI teaches us a rhythm of request, respond, receive a revision, and repeat. Somewhere in there, hopefully, is a bunch of critical thinking and judgment.

Over time, and with enough practice, we become more efficient at getting more useful output. But simply using AI doesn’t give us a mental model for how it actually works. This is why an output that doesn’t quite match what you expected can sometimes be puzzling.

And it isn’t just a beginner problem. Someone who opened Claude yesterday can share the same uncertainty as someone who has used generative AI for years.

The operative word is generative.

What “generative” means

To understand why my prompt produced such stellar garbage, I had to go back to the basics. What exactly is generative AI?

Unlike other types of AI, generative AI produces new outputs. More technically, a generative model learns a statistical representation of patterns in its training data. It can then use that representation, together with the context we provide, to produce a new output.

For an LLM, that output is built one token at a time. The model estimates which tokens could come next, selects one, and repeats the process.

Unlike a search engine, it isn’t searching an index and pulling out a completely finalized answer that matches the request. But a product like ChatGPT can also search the web, retrieve files, and use other tools.

These capabilities supply information to the model, but they don't make the model generative.

Our instructions, examples, and files also provide information and influence what the system creates. But we aren’t selecting a finished answer from a shelf. The system constructs an output for this particular interaction.

Impressive results and garbage come from the same process

This process is what gives generative AI its awesome flexibility.

It’s why an AI tutor can explain a difficult concept to a student, even if that exact explanation never appeared in its training data. It’s why an image model can create a visual for a lesson concept that is made up of a combination of details the system has never encountered together.

It is also why, on occasion, the system produces garbage. Sometimes the process matches our needs incredibly well. Sometimes, it fills a gap with something plausible, but incoherent. Sometimes, it changes a detail we expected it to preserve.

That’s why the results can be coherent and convincing, without being instructionally sound.

Why this matters for instructional designers

Instructional design artifacts follow recognizable forms.

Learning objectives generally start with verbs. Course outlines follow traditional course, module, lesson structures. Multiple choice assessments tend to follow common psychometric conventions.

Generative models can learn those patterns and produce something that closely resembles our work. Increasingly well I might add. The output can look like instructional design before anyone has done any instructional design thinking.

I think this is why many instructional designers, myself included, rushed to use ChatGPT when it launched nearly four years ago.

But after over 7,000 hours of using AI, I now understand that resemblance differently. The appearance of a finished artifact is not proof that the design work is finished.

This raises an important question that I want to come back to in later articles: How should we decide where generation belongs in our work? What standards should its output need to meet? And how do we keep using AI as the tool it is, to help us meet the ever-increasing demands of our stakeholders and clients?

If you enjoyed reading this, please consider subscribing. We plan to keep this publication free for all, and we’d love to have your thoughts as part of our discussion.

Subscribe now

Bill Gates recently released an essay on his blog, “Gates Notes” outlining the downsides and upsides to the AI era. This era is already upon us and Gates is both optimistic and worried. Common sentiments for sure. The essay is urgent and calls for action.

People in the education space are feeling much of the urgency that Gates describes. How instructors, instructional designers, and content production folks do their jobs is changing. In some cases, those jobs are changing dramatically. And learning itself is experiencing a shift. Digital learning impacted the teaching vocation in important ways and now AI tools are amplifying those changes.

How do the changes Gates outlines impact the activity of learning and the vocation of teaching? More importantly, how should instructional designers, subject matter experts, and content production people think about these changes? Here are some ideas that may help.

Adoption can keep poor content in check

It can be tempting for people in the education space to resist using new tools. Learning is guided by some basic and fundamental principles and new tools and different ways of doing things may seem, on the surface, to not respect those fundamentals. But resistance may only enable people that don’t really understand learning to use these tools and overshadow people that have spent their lives fine-tuning their skills.

A better approach is to learn how to use the tools to amplify and energize the skills you have. For example, learning how to incorporate time-tested instructional design principles into prompts that you can use to draft content is a way to use AI to amplify those principles. If your prompts create more instructionally sound content than someone who doesn’t know those principles, you can create powerful teaching materials by using the tools rather than resisting them.

Gates gives a great example of how AI can amplify teaching rather than subvert it. He writes,

AI can be a boon for education as well, despite the concerns I mentioned earlier. It can free teachers up to spend more time working with students one on one or in small groups and give them a clearer view of where the whole class is struggling. For students, an AI tool that preserves what researchers call “productive struggle”—the cognitive work that builds understanding—can strengthen learning.

Market forces will influence but don’t need to define education

Gates rightly warns that AI is and will continue to take jobs away from humans. He argues that systems need to be put into place to assist those in “transition” to new activities. I agree. He also intriguingly argues for a policy he calls “Human Reserves” — jobs that are set aside only for humans.

In our work in the education space, Kasie and I experience the pushback on AI-developed content mainly in the instructional design space. Some of it is from people fearing job loss and a general pessimism about whether AI can do as good of a job as they can.

But much of the resistance is based on principle: even if a learner can’t tell whether something is AI developed or created by a human, this group doesn’t want AI-developed (or at least fully AI-developed) content. This sentiment is based on belief that using lived experience to influence content design produces a different kind of thing from synthesized content that is generated from a computer even if the end result from both has a strong family resemblance.

I think this will be true across industries but particularly in the education space.

Now it’s true that low-cost, mass-produced, low-quality products and services many times win. People can easily buy and consume mass-produced, lab-driven food products. But it’s also the case that many people simply do not want these types of products. Unfortunately, the option for better quality food typically is only available for people with means and time (you’ve probably heard the original Whole Foods Market referred to as ‘Whole Paycheck’). Places that sell and serve higher quality products are being relegated to ‘boutique’ status.

Given this, there’s an uncomfortable analogical tension here. If AI-led, mass-produced education is like mass-produced products, where does that leave us as educators? What do people that want to build solid learning experiences with more intentionality, personal investment, and higher-quality do? And how do we ensure that everyone can get access to high-quality education in the potential future of what is being called ‘AI slop’?

I don’t have the answers to all these questions. There are two obvious options: either go all in on being boutique in your vocation or embrace the AI shift and find other ways to support the educational content it produces. Gates suggests some ways to use AI in just this way in the quote above.

There is a third way which I alluded to above. Continue to excel in your discipline and find ways to use AI as an accelerator and amplifier. What this looks like will most likely be unique to each person or industry. In our business, we’re using our decades of instructional design and classroom experience to heavily influence how our AI-generated content gets created and the more we refine and influence the tools using that experience, the higher-quality content we’re seeing.

There’s another result. By using AI to amplify our work, we can make higher-quality education available to more people. As Gates implies, as AI takes on more mundane work, it not only frees us up to do more meaningful tasks, but it can make access to the work we do cheaper and more widely available. We spend our time on depth and AI tools help us with the breadth.

First principles should guide the way forward

People like Gates who focus on fixing things tend to gravitate towards complex solutions because the problems are complex. Sometimes complex solutions are needed. But as William of Ockham has taught us, they should be no more complex than they need to be. Expanding on Ockham’s principle, I’ll also add that solutions should excel at the fundamentals.

Part of determining what an essential, first principle is in any domain is to use an evolutionary principle: what are those things that, despite numerous changes, persist over time? What has survived? Just because the answer to these questions have staying power doesn’t guarantee that they’ll be applicable to the future. But it does tilt the probability that they will heavily in their favor.

We can come up with guiding first principles in education by looking at the areas that AI is currently creating risk:

• People learn best through human connection both with instructors and other learners.

• People want to actually learn what they’re being taught. That is, true learners aren’t passive. They want to be actively involved in their learning experience.

• People want education that moves them forward. This could be intellectually, vocationally, or just in ways that widens their world.

• People need to learn things that are meaningful to them and that give them purpose.

For most if not all of these items, personalized, human involvement is essential for meeting the needs. When educators use AI tools, they can amplify their influence in digital education. These tools also can provide more time and resources for educators to work directly with learners or to spend more time on really crafting experiences they previously would not have had time to build.

Despite any change that AI and robotics may bring, these are some of the basic things humans care about and most likely have always cared about. The solutions educators come up with for dealing with the AI shift should focus primarily on preserving or fostering these basic educational drives. This can be done with AI as a fully integrated part of our learning experiences—AI and strong educational programs need not be at odds.

The future is ours to create

Karl Marx famously said, “Philosophers have only interpreted the world in different ways. What is crucial, however, is to change it.” This appears at the end of the eleventh note he wrote on the German philosopher Ludwig Feuerbach’s ideas. Earlier, in note three, he wrote this:

The materialist doctrine that humans are products of circumstances and upbringing and that, therefore, men who change are products of new circumstances and a different upbringing, forgets that circumstances are changed by men themselves . . .

One doesn’t have to buy into Marx’s entire philosophy (I don’t) to agree with this sentiment. When it comes to the impact of AI, we should create the future we want. AI is a tool. Let’s work to make it a powerful and useful one that serves us and those we’re trying to educate.

A few weeks ago, I decided to make the leap from ChatGPT to Claude. New era, right? Except there was no magic button to bring my digital life with me.

It’s the middle of the afternoon. I’m sipping on my third cup of coffee for the day, my laptop open, and I’m staring at a folder of JSON files I just exported from OpenAI. If you don’t know what JSON is, you’re not alone. It’s basically a wall of curly brackets that might as well be written in Elvish if you’re not a developer.

So here I am, wondering how long it's going to take to dig through code and transfer three years of my life from ChattyG to Claude. So I spent days building custom prompts and skills specifically designed to analyze my exported files and reconstruct my memory from scratch — prompt engineering, testing, iterating, the whole thing. Hours of work.

Naturally, right after I finished, Anthropic dropped a shiny one-click memory import that does the whole thing for you. Of course they did.

So here’s what I’m walking you through today: what AI memory is, and exactly how to drag yours from ChatGPT to Claude.

Your AI Has Been Taking Notes on You

Before we get into the how-to, here’s a 60-second crash course on memory.

Every time you mention the grade level you teach, your instructional design standards, or which LMS you use — it saves that. Quietly building a profile of who you are and how you work.

That’s why it gets more useful over time. Less "who are you again" and more "I already know how you work."

Both ChatGPT and Claude do this. And both let you see, edit, and delete what they’ve stored. The problem is that all of that context, the stuff that makes your AI useful, lives inside whichever tool you started with. And if you want to switch? You’re starting from scratch.

Or at least, you were.

Enter Import Memory

Just a few weeks later, Claude launches Import memory. This tool helps you bring over your memory in an organized way.

So how can you use it so you don’t need a computer science degree to move your data over?

Step 1: Open the Import Tool in Claude

Log into Claude (web or desktop app — both work). Then navigate to:

Settings → Capabilities → Memory

Scroll down until you see “Import memory from other AI providers.” Select Start import.

Step 2: Copy the Export Prompt

Once you select Start import, Claude has a ready-made prompt for you. It’s a short message specifically designed to ask your old AI to hand over everything it knows about you — your preferences, your context, your patterns, your tools — in one clean, organized block.

Copy that entire prompt.

Step 3: Paste It Into ChatGPT

Open ChatGPT. Start a brand new conversation. Paste in the prompt Claude gave you and hit send.

After just a few seconds, ChatGPT creates a highlight reel of everything it’s been hoarding about you. Job title, tools, preferences, formatting quirks, projects. All neatly packed into one code block.

This is the part that took me days of building. You? Thirty seconds, tops.

Copy ChatGPT’s entire response.

Step 4: Paste It Back Into Claude

Go back to Claude’s import window — the same one from Step 1. Paste everything ChatGPT gave you into the text box.

Select “Add to memory.”

That’s it.

Step 5: Wait

Claude needs about a day to chew through what you imported.

After 24 hours, go to Settings → Capabilities → Memory and select “See what Claude learned about you.”

It’s time for a quick audit. Skim what Claude saved, change or delete anything that seems old. Add anything chatGPT missed. Ten minutes here saves you hours later.

Step 6: Take It for Test Drive

Ask Claude to do a task that you used to use ChatGPT for—and see what happens.

Have it draft a lesson plan in your voice. Make it write parent emails the way you do. Test it on what matters to you.

That’s when you’ll feel the difference.

What Comes Over (And What Doesn’t)

This is a great tool, but let’s set some honest expectations.

Transfers well: Your general preferences, your professional context (what you teach, where, what tools you use), how you like things formatted, and the big-picture patterns ChatGPT picked up about how you work.

What doesn’t: If you had Projects, or Custom GPTs with all the bells and whistles, those don’t transfer with this. You can use a similar method to transfer your projects and Custom GPT’s, but it’s slightly more complicated.

Worth leaving behind: Not everything ChatGPT “learned” about you was accurate. Some of what mine had saved was outdated. Some of it was just wrong — assumptions it made early on that calcified into "facts." The import process forces you to read through all of it, which means you can finally clear out the bad takes your AI has been carrying about you for months.

Educators Don't Have Time for a Learning Curve

If you’re a teacher, instructional designer, curriculum specialist, or coach — you have so much context that your AI needs to be useful:

• Your state standards

• Your scope and sequence

• Your LMS

• The specific way your school does things.

Every time you open a new AI chat and have to re-explain your whole world, you pay a time tax. And time is the one thing educators never have enough of.

This feature means you don’t have to pick between the AI I’ve already trained and the AI I actually want. Day one, Claude already knows your classroom.

If you’ve been eyeing Claude but staying put because you didn’t want to start from scratch, that excuse just got retired.

You don’t have to do it the hard way, I already took that hit for you.

Next, we’ll dive into Claude Projects. If you’re thinking about making the switch, follow along — I’ll tell you everything the product pages won’t.

Break Up With ChatGPT Workshop— Want to go deeper? Check out the workshop we hosted on making the switch.

Making the Switch Resources

We’ve all said it. And it feels right, doesn’t it? Because the frustration is real. You were hired for your expertise in learning science, but you spend your days formatting Word docs and chasing SMEs who won’t email you back. Then you get pulled into a launch 48 hours before go-live, like you can magically “fix” how 10,000 people work over a weekend.

The story we tell ourselves is simple: They don’t get it. They don’t value the craft. If they just understood what we actually do, things would change.

It’s a clean explanation. It’s got a villain (leadership) and a victim (us). But honestly? I think that story is exactly what’s keeping us stuck.

The comfort of being a victim

When I started in ID, I was all about learning science, and cognitive psychology, and constructivist theory. But, six weeks in, I was a glorified template-formatter. It was demoralizing. So, I did what we all do: I fought for “respect.” I took time to explain what ID actually is. I tried to educate them on cognitive load and evidence-based practice.

And man, I loved that work. By nature, I am an educator right? It felt good to educate people on what was needed for strong instructional design. It was comfortable.

But then I realized, some teams actually win that fight. They get the executive sponsor. They get the “seat at the table.” And you know what? The work still gets squeezed. The needs analysis still gets cut from two weeks to three days. The pilot still gets skipped.

When we focus on “respect,” we’re looking at attitudes. We’re watching people’s faces in meetings for signs of validation instead of checking the project plan for signs of structural failure. And worse, we become permanent complainers. If our identity is “the undervalued expert,” we actually need to stay undervalued for the story to work. Not comfortable.

You're not undervalued. You're structurally unprotected.

I was reading about this concept the other day called Hyperbolic Discounting. It’s a behavioral econ term, and I think it’s the “missing link” for us.

Basically, humans are wired to overvalue a tiny, immediate win over a huge, delayed reward. When a manager cuts our budget, they get an immediate “win” for their quarterly target. The “cost”—the higher error rates, the support tickets, the slow adoption—doesn’t show up for six months.

Training and learning are delayed rewards. And when they’re not done well, they’re a delayed cost.

An executive can genuinely respect your work and still trade it away. Not because they’re jerks, but because nothing breaks immediately when they cut our time.

When budgets get tight, people look for the stuff that has a hard ‘No.’ You don’t skip Legal because you’ll get sued. You don’t skip the product launch because the revenue disappears. Those things have teeth.

But Learning? We’re treated as optional time. Everyone assumes our timelines are ‘flexible.’ They think they can compress a month of design into a weekend, or ‘just make it a PDF,’ and the project will still move forward. And it will move forward—at first. The consequencesare real, but they’re delayed. And by the time the bill comes due, nobody remembers that the training was the thing that got cut.

This isn’t a respect problem. It’s a prioritization problem. We are structurally unprotected because our failure is silent and slow.

Stop Asking, Start Accounting

If we want to change how we’re prioritized, we have to stop absorbing the blow. We have to stop saying "we'll make it work" and start making the trade-offs explicit.

1. Force the trade-off. When the timeline gets crushed, stop being “helpful.” Say: “We can hit Friday, but that means we’re skipping the pilot and the needs analysis. That usually leads to a spike in support tickets and lower adoption. Are you okay with that trade-off?” You’re forcing them to actually prioritize—to choose between the date and the quality.

2. Define what’s “Non-Negotiable” early. You can’t prioritize in a crisis. You have to decide what the “Strategic Core” is on day one. Tell them: “The formatting and video creation are flexible. The needs analysis and the assessment are not. If we lose those, we lose the project’s ROI.”

3. Show the “Price” of the last cut. Next time they try to squeeze you, bring the receipts from the last time they did it. “Last quarter, we skipped the pilot to hit the launch date. It resulted in 40% more rework hours in month two. Do we want to repeat that, or prioritize the design phase this time?”

The “respect” narrative makes us feel like victims waiting for a hug. The “prioritization” frame makes us partners managing a project. Stop trying to change how they feel about you—start making them own the consequences of their choices.