Learnlets

Do you know how to learn?

10 November 2026 by Clark Leave a Comment

Do you know how to learn? Does your organization expect you to be capable of keeping up with changes and adapting to new situation? I bet they do! And, what are they doing to help you learn how to be effective at that? Because the evidence is you’re not now. And unless someone’s doing something about it, it’s going to stay that way.

Sure, you survived college most likely, or at least high school, and in theory they teach you how to learn. Yet, they don’t. Yes, the require it, and expect it, but actually teaching you specific skills is rare. There are a lot of things people do that aren’t effective. And, let’s be honest too much of high school and college is about passing the test to achieve the grade anyway. Or any possible way to achieve marks without actually learning anything. The few who actually want to learn may become better at it, but that’s not the way to bet.

We have strong evidence that people aren’t good at learning. They use ineffective methods like highlighting. They get the experience of feeling like they learning when it’s easy and they can pass test right away. However, there’s strong evidence that the outcomes of such don’t linger. And most educational interventions and organizational training doesn’t actually look at long term bridging. Even mine! (OK, I’m trying to get better. ;) We ask it of you. We expect it of you. And, we don’t support it for you.

By and large, we ignore most folks on self-development. Yes, when people show executive potential, they may well have it as an objective for leadership development. Except, as Jeffrey Pfeffer (c.f. Leadership BS) will point out, most leadership development is ‘make you feel good’. It doesn’t teach much of anything, and rarely ‘how to learn’. Worse, that’s still only for the select few! What about frontline workers? Jay Cross’ back of the envelope calculations suggested even a 10% improvement across the board would be valuable.

There are significant things we can do to be better learn. Harold Jarche‘s Personal Knowledge Mastery (PKM) talks about seeking knowledge effectively. How do you make the feeds that continue to give you new ideas without overwhelming you with information? How, when you’re searching do you get good answers not the easy answer but the right answer? That’s not a given, possibly even less so with AI summaries! Then, you make sense: you can experiment or re-represent, but too often we stop there. We’ve got the information, we processed it. We think we’ve learned, and that’s not the case. If you wanna have ownership of it, you have to take tested and refined it. Repeatedly, over time.

And, we’re not even talking motivation! Do you believe it’s a professional obligation to continue to improve? Too many people are happy being complacent. Yet, motivation is addressable as well (c.f. The Motivation Blueprint). And, of course, it’s an important part of learning (c.f. Make It Meaningful). Even if they are motivated to learn, again it takes the skills as well.

Then, if you’ve learned something you should share it so others don’t have to do the same struggle you did. Or, they can give you feedback to improve you because maybe they’ve been there before. You should be learning ‘out loud’. These are all mechanisms we can get better at if we have the guidance. If we are supporting learning how to learn, we need to make that explicit. Teach it in courses and universities, and until then even organizations.

I’ll suggest the answer to the question “do you know how to learn?” is ‘no’. Sure, most of the readers of this blog will, I reckon (or you wouldn’t be here). But the average answer will be ‘sure’, but they’ll be wrong. Heck, it may be that you are too (c.f. The Accidental Instructional Designer).

What does your organization do to make sure you’re a more effective learner? Not just access to information, by the way, but actually making use of it? (Access to information can be useful, but again it’s not the way to bet.) Organizations should be providing specific training on effective learning skills, particularly in this era where the tools are getting more seductive, but potentially actually decreasing learning. There are ways that let you stay up-to-date, adapt, and be agile to meet your needs. For biologically secondary learning, your natural instincts are likely to be wrong. You can make it explicit in your pedagogical practices, but explicit instruction is the best way to actually provide a positive answer to the question.

Getting real

3 November 2026 by Clark Leave a Comment

So, one of the things that’s happening is that my ideas on learning science are driving the development of elevator9‘s platform. And, that’s a good thing, I think, because my postulate is that too many platforms have nobody in the C-suite who understands learning. I’ve talked before about learning from this endeavor, but I want to take a step back and talk more in principle about getting real in practice.

Of course, right now it’s easy for anyone to make a platform, and many are. Which doesn’t mean the design is based on learning science, to be sure. I recall meeting a new platform provider’s team, and their answer to the guiding principle was what the customer wanted. Which, to be clear, isn’t what the customer needs. (This was all pre-GenerativeArtificialIntelligence, of course). And I get it, but it’s not a long-term strategy.

What, instead, is that the goal here is to take the principles of spacing, and see how that works for learning. One of the things to realize is that, for pragmatic reasons of access and c0st, many studies are done a) with college students in psych, b) in quite small windows of time, and c) with only one learning goal. I get it, but…that’s not how the real world works. So, we’re facing some challenges.

What the intent is is to augment face-to-face events with the reactivation over time and bridging the gap from event to ultimate performance. However, most events have more than one learning goal. So, which to space? Also, if done right, there may be more than one event (certainly for leadership, one of the areas that prompted CEO David Grad to start this whole initiative). The events may also have new and different learning goals. So, for instance, evaluating self and others’ emotions at one event, and moving on to coaching at the next meeting). Which means that we’re starting some goals new, and there are likely still some previous goals that need addressing.

It gets more complicated, too. What if the audience is busy and/or unmotivated? How do we make it easy to accomplish? Short interventions, to be sure, but how abour rescheduling on an individual basis? These issues begin to bring in interface design issues on the part of the learning designer and the learner! What’s happening, of course, is iteration: trialing, evaluating, revising, etc. Learning, in short.

What is a positive, of course, is the learning that accompanies this effort. elevator9’s going from an initial vision to a combination of what folks know they need, and what we know they need! We’re also going from a app-building environment to a true developed system. We’re going from a couple of folks and a drawing board to a full team. Along the way, lots of issues emerge. I’m don’t have much experience outside of learning design and cognition, so I’m listening and learning. I of course can’t keep my big yap shut all the time, and weigh in on some things (even, occasionally, usefully!).

In retrospect, that’s also true of what I’m doing with LDA, of course. We’re trialing different curricula, testing out things about the offerings, and trying to not just survive, but succeed in our goals to lift the game of the industry. In both cases there’s established competition, so it’s not an easy step. Still, it’s the right one (and I’m stubborn ;).

The reality, not really new but always a revelation, is that it’s great to know theory. It’s great to solve real problems. Mixing them together, trying to apply theory to real problems, is even better. It improves the solution, and informs the theory. It’s all about getting real, and that’s a blessing. That’s my take, what’s yours?

Context in context

27 October 2026 by Clark Leave a Comment

Context is something, I think, that we don’t think enough about. It plays a fundamental role in our cognition, and so, too, our learning. It also, then, plays a role in our use of technology. Which means we should be thinking about context in context.

So, I’ve talked before about how context matters. Our current models of cognition suggest that we’re not the logical reasoning engines that we used to believe we were. That is, our interpretation of the situation we’re facing is a complex interaction between our previous knowledge and our current situation. We can, in fact, completely miss something, because we’re assuming it’s a situation we’ve seen before. (I’m guilty of this, for instance!) Similarly, we can be completely flummoxed by a situation we should be familiar with, because something about the context has led us to mis-categorize the situation.

Which implies, strongly, that making sure we’re seeing the right contexts across examples and practice is important in designing learning experiences. And that we should facilitate reflection to help learners recognize how things transfer. We can’t assume that folks will transfer what we’ve presented to appropriate situations, we need to engineer it!

This applies to technology, too. We can’t assume that it’s aware of the current context. As I stated in a link above, we can tie technology to a particular context. Or, harder, we can try to make the technology ‘context aware’. That is, have enough appropriate ways of sensing the environment that the system can know what we’re doing, or where we’re looking. Augmented reality, for instance, depends on this.

One other approach, as we’re currently seeing in generative artificial intelligence, e.g. Large Language Models, is to convey the context to the system. This is what a lot of ‘prompt engineering’ is about; getting the system to understand enough of what we are asking to appropriately provide a response. And, that’s not a bad approach. With one caveat; we can be blind to some of the contextual elements. And, until we provide them, the responses we get are less likely to be useful. (And, they’re never 100% trustworthy even then, they’re still probabilistic. We need experts in the loop.)

My point is that we need to be more aware of context than we are. We need to account for it in learning design, in technology use, and more. We should be going as far as what Don Norman suggests (and I’m on board), that our contextual awareness needs to extend to the implications, e.g. environment, society, etc. If we’re taking context in context, we’re more likely to be addressing the immediate, and long-term need, than we are if we don’t. It’s that simple. It’s hard, I know, as I struggle with it too. Still, it’s important. So, shall we?

A whinge (you can avoid as you wish)

20 October 2026 by Clark 1 Comment

I recently received an email, that I responded to. One of the weird things was that it said it came from one person, but the email body said it was doing it on behalf of that person. ? So, I asked. I’m not sorry I did, though I was quite sorry to see the response. So let me share a bit of a whinge (you can avoid as you wish).

So, with the setup above, here was the (anonymized) response:

You’re right about the confusing sender. I’m an AI assistant helping Xxx with Xxxxxxx correspondence, using his mailbox with his authorization. These replies are from the assistant, not Tim personally. Sorry that wasn’t clear.

Several things are, to me, very wrong with this. First, the original message didn’t state that it was an AI. The original message simply said:

I’m helping Xxx, founder of Xxxxxxx.

I feel like I should be informed about the use of technology instead of a person from the start. Yes, it’s not a legal requirement, but I’m thinking perhaps it should be. Because a person, even me, could be misled otherwise, and misleading is really fraud.

Then, there’s the presentation as being intelligent. The characterization of “I”, implying a person, doesn’t sit right. Also, stating that they’re “helping” instead of working on behalf of is important, it would seem to me. This is, after all, a probabilistic machine-generated message, without any consciousness nor contextual awareness. Implying sentience when such isn’t the case again feels misleading.

Of course, anyone giving an AI authorization to generate mail is asking for trouble. It’s not clear if it’s a local AI or using one of the big systems (OpenAI, Anthropic, Gemini, etc) using stolen IP. (Yeah, I know, my general rant, but I still find it odd that our society seems to countenance stealing people’s IP and then charging them to access it. Theft as a business model? What’s next?) The former would be more scrutable, of course. Regardless, when agents have hacked external databases and eliminated internal ones, that seems like a dubious step. Note that these systems have gone around explicit prompts to behave, so that’s not a solution.

Then, there’s the implication that this is being done without Xxx’s input. Or, rather, I’m not clear what Xxx’s input is!

Xxx would decide after seeing the quote.

Has Xxx been involved to this point? If so, how? Or am I negotiating with a stochastic parrot? If so, I feel kinda demeaned. Clearly, Xxx doesn’t seem to believe it’s important for me to know. Ahem.

Overall, this is just the type of scenario I fear happening. Xxx is presumably saving time, but also generating ill will, at least amongst those of us who don’t automatically expect that this is just fait accompli. So, maybe this is just a whinge you can avoid as you wish, but it, to me, is a stark example of just the sort of situation we’re devolving to. Is Xxx actually developing any skills here? Making decisions, or cognitive offloading? Further, do I want to assist an organization that could be helping to decrease intelligence? (And, can I afford to be so principled in this environment?) Maybe I’m way off. What do you think?

(To be clear, I get lots of unsolicited offers – loans, GoodReads Listopia, magazine ‘profiles’, etc – which I don’t like but I understand. Many are probably AI generated too; the Listopia ones in particular seem to be so. This, however, rankled in a different way. FWIW.)

You. Need. To. Know. Practice!

13 October 2026 by Clark 4 Comments

Everyone’s fretting about the role of IDs (instructional designers) in an age of AI (artificial intelligence). And, there’s so much wrong with this! Quite simply, IDs are still going to be necessary. They always have been, except too much of society has a legacy belief that undermines that awareness. And, to be fair, too few IDs actually know what they should. So, I want to make it perfectly clear: You Need To Know Practice!

Let’s start with the need for practice. Despite being obvious, it’s seemingly unknown! Or, rather, it’s got a mistaken interpretation. For one, folks think that if we present information, that learning will occur. Er, no. That’s definitely wrong; we need to practice retrieving the knowledge. Information dump means, quite literally, wasted money. It’s highly unlikely that presenting information will stick in our brains! If we need information (and we devolve to this all too often), put it in the world. If it absolutely has to be in the head (And yes, there are times this matters. Medical vocabulary is the example I always use; you don’t want to remove an elbow because you didn’t know the term for knee), you can use drill and kill, but that’s highly unlikely to be the case. Good analysis here is the first step. As the old canard has it: tellin’ ain’t training.

Then, it’s not knowledge retrieval we need to practice, or even test. Pooja Agarwal’s research (warning, PDF) shows that what we need to test is the ability to use the information. Too often, we have a legacy belief that we should knowledge test, and that’s sufficient: if they have the knowledge they can do the task. Again, no! It turns out that asking learners to use the knowledge already requires them to access it, so we don’t need that test. Yes, we need generative/elaborative activities to cement that understanding, but that’s prior to retrieval practice.

What’s critical to learning is practice using the information in the way that it will be used after the learning experience! What that means is understanding the actual performance, and designing (essentially) simulated practice. They can be mini-scenarios, branching scenarios, or full simulations. This is where IDs do what can’t be done in other ways. There aren’t technologies that can create meaningful practice; it takes IDs to design them. Technology may be able to play a role, but an ID needs to determine the necessary tasks. (Note: SMEs can’t do this, no matter what technology they use. For instance because they don’t have access [again, PDF] to about 70% of what they actually do.)

Another thing IDs do that I don’t have any reason to believe technology can solve is to choose the suite of contexts. That is, across examples and practice, the minimum set of situations that will foster transfer to all appropriate situations, and no inappropriate ones! Technology doesn’t understand context in any meaningful way. Again, the ID is the one who can (and should) select the minimum appropriate set of contexts.

These two core roles are, to me, what an ID alone can do. There’s probably more, like determining the best way to present information to the learner. What media should we use in this particular instance (is virtual reality [VR] necessary here?), etc. One of the tasks that I think IDs particularly need to perform is the afore-mentioned analysis. How can you determine what people actually need to know if you don’t know what performers are actually doing when they do it right? This sort of analysis can’t be put to technology, either.

Note that feedback is a necessary component of practice. It’s the reflection to the action, but it needs to be nuanced. It needs to be informative (read: useful), minimal, and non-personal. Getting this right is another aspect. Spacing out the practice, too, ends up being important (hence my engagement with Elevator 9). While technology may be able to help here, most likely on the spacing, it could err on the side of not getting feedback right because context again. We learn, after all, through action and reflection, and practice is action, and feedback is reflection. Even things that aren’t biologically primary, in the Geary sense, should be learnable by being driven to resources by practice and feedback.

Of course, all of the above does presume that the ID actually has the necessary skills. Which comes from applying the foundational knowledge of how people learn, e.g. learning science. (Which reminds me that we [the Learning Development Accelerator] are having a Learning Science Conference for just this reason!) We do need to have the professional background to adequately execute the necessary tasks. I’ve pointed before to the recommendation that the next economy will be a transformation one. That is, where we pay for experiences that change us in ways we want or need to be changed. And, that’s what IDs do, when we’re on our game. We are the future!

Tech isn’t, or shouldn’t be the worry. We can (read: should) be the necessary component. Of course, not only we, but society needs to know this. So we can and should promote this awareness. And, of course, counter any vendor who wants to pretend otherwise ;). ID skills are necessary. You need to know practice! You also need to know when to practice, and when to find other solutions, like job aids. But if you do, you’re increasingly invaluable. That’s my stance; am I missing something?

Smauthoring tools

6 October 2026 by Clark Leave a Comment

Is it smothering, or authoring? Are we authoring actual learning, or creating smothering amounts of content? Maybe the answer is ‘yes’…I’m suspicious, however. So, I’ve had opinions on authoring tools, though they’ve changed over the years. Still, there are some things that need to be core. And, of course, they’re violated. So here’re some thoughts on smauthoring tools.

So, first, to achieve real learning – acquiring new skills – you need real practice. Knowledge presentation is unlikely to yield any meaningful changes, because unless folks apply the knowledge, it’s just knowledge. And, in fact, you should be evaluating to see whether you’re having an impact. You might make content available, or present beforehand, but the important aspect is the practice. Repeatedly, over time. (Yes, there are exceptions, with experts, and serendipitous outcomes, but it’s not a fruitful bet.)

What this implies, then, is that authoring tools should focus on practice. That is, they should make it easy to make meaningful activities. If they just support knowledge test, they’re not going to support the important learning we need. Sure, you can turn multiple-choice into mini-scenarios (I’ve run a whole workshop on that in the past!), but… You’re really working around the limitations of the tool. It took years for them even to have more than one response for all the wrong answers, for crying out loud! In short, authoring tools haven’t, by and large, provided good support for designing learning that sticks. (There have been tools to support building branching scenarios at scale,

However, folks who don’t know learning (or learning science, hence our conference, *cough*), can often believe that presentation = acquisition. That’s a legacy, I suggest, of old beliefs about how our brains work. And, authoring tools have been promoted then, and again now, to not require instructional designers (IDs)! Yes, you can just have experts create learning! Ahem. Sure, it’ll sell, and the executives and managers and learners will feel good about having seen the content. But, you have essentially zero change of actually having an impact.

So, you really can’t have domain experts design learning. And, you shouldn’t. Experts should be experts in their domain, and not be experts on learning. That’s a waste of resources. You should have IDs who know learning, and they should work with the experts to create experiences that develop abilities. The problem is compounded, of course, by having folks who are good at the job being moved to trainers and then instructional designers without any real experience with learning. What Cammy Bean aptly called the ‘accidental’ instructional designer. Give them tools, and they too can create elearning! Er, no.

AI compounds the problem, of course. It can take the content and do the manipulation into information and knowledge test quite well. Yet it’s still not learning. And, really, AI doesn’t know context, and so can’t really make meaningful practice. Again, what we need if we want to develop skills, not just knowledge, is contextualized practice. I don’t think AI can do that.

In the (very) old days, it took teams to develop digitally-delivered learning experiences: they had folks who knew how to program, and folks who knew interfaces, and folks who knew the domain, and folks who knew interfaces. I understand the desire to reduce head count to keep costs lower, and we now do have technology that can support meaningful learning. (I should know, I’ve tracked many and even advise one!) It’s too bad, however that the vendors want quick sales and promote the knowledge, and not the skill. We have to measure, to know, of course. (Completion, and ‘liking’ the course aren’t good measures, by the way.) Once we do measure, we can start erring on the side of making learning that moves necessary needles. We need real practice development tools, not smauthoring tools.

More on evaluation

29 September 2026 by Clark 2 Comments

So, just as I’ve been reviewing a book on evaluation (and sharing one issue), I also get my mitts on a copy of an evaluation report. And, yes, I tend to blog about what I’m currently doing (this is learning ‘out loud’ after all!) Further, while I’ll freely admit that I’m not an evaluation expert, I do know a wee bit about methodology, who’s said what, and bad graphics. All of which were in the report. So, here’s a bit more on evaluation.

So, first of all, it’s not a bad report. Will Thalheimer did one at the beginning of the year, and having more is better (a lesson I learned from Don Taylor about AI surveys ;). It asks respondents about their practices. It also includes quotes from noted names (not including me ;), and case studies. The data are interesting. Oddly, it does not mention Will’s survey!

However, there are problems. First, as a survey, responses only characterize those who respond. Given that it’s by a vendor, that likely biases the sample. Also, surveys tend to get folks who want to respond. We get no details on how respondents were even made aware, solicited, nor the actual survey questions (tho’ you can infer). There are also, supposedly, 15 in-depth interviews. Maybe those are the noted names and case studies? To be fair, the results don’t suggest a bias towards those who want to respond, as the indications are that folks still aren’t doing a good job. Which isn’t surprising, though dismaying. Still, there should be qualifications about limitations for considering the results.

Another issue is that the report is ostensibly about Return-On-Investment (ROI). Which I consider to be the area of expertise of Patti and Jack Phillips (rightly or wrongly ;). They have extended the Kirkpatrick-Katzell model explicitly to include ROI. Thus, I find it odd that they receive almost no mention (except for one minuscule print reference). Nor, for that matter, is there a clear definition nor formula for ROI! Another strange omission.

A further omission is any talk of Will. Not only does he have a survey, but he has a worthwhile instrument! At least Phillips & Kirkpatrick (who are better known of course) get a mention, no matter how obscure. Of course, others like Brinkerhoff also don’t get a mention, so probably not a deliberate slight of Will. However, I think helping folks know that you know the players is a sign of credibility. Maybe the academic in me?

There are some hiccups in the formatting, but that’s not a big deal, except that it purports to be official output of said vendor. Which, to me, undermines credibility. If they’ll make mistakes in formal output, does that indicate a lack of diligence in other things they do? I don”t know the answer, but it’s not salutary.

I was given a chance to publicize the report, and I’m sure this post doesn’t count ;). (Once again, my principles are interfering with my success…) I’m not naming names, because the goal is not to shame, but educate. And I think it’s worth knowing what good methodology requires, and that includes references to relevant works and qualifications on what the data shows. There’s lots more on evaluation, but it’s a start.

Into action?

22 September 2026 by Clark Leave a Comment

I’m having a look at a book on evaluation that I am finding has a lot of value. A familiar model (not well-referenced, mind you), for one. Not as nuanced as Thalheimer’s Learning Transfer Evaluation Model (LTEM), but comprehensible. And importantly, really good analytics of problems and ways to communicate them. However, there’s one aspect that I’ve seen elsewhere, and it’s bothered me without me being able to articulate why, until now. And I realize it’s about putting learning into action. What do I mean?

So, the book talks about three types of learning outcomes: basically informational, then changing behavior but not impact, and then finally impact. The explicit belief is that organizations have learnings of all three types, and therefore they’ve aligned evaluations to that belief And, as I said, they have (so far, I’m not done ;) good evaluations at each step that identify ways in which the learning can go wrong and remedies. In many ways, reminiscent of Julie Dirksen’s latest and important book Talk to the Elephant. All good, but I have a quibble about that initial breakdown.

To me, you don’t give people information and expect them to do nothing with it. You give it with a reason: whether it’s what HR policies are available, or how you handle commission, or what have you. You want people to use the available HR opportunities if relevant, and likewise you’d like commission to be motivating. Sure, you can see whether people passed a test on the knowledge, but why would you give it to them unless you expect them to do something differently as a result? That, then – doing something differently – is really your goal.

The same with workplace behavior change. You don’t want it to lead to anything meaningful? I doubt that. Why would you ask for things to change if you didn’t think it mattered? If you do, then there’s got to be some reason, some outcome, and therefore some way to measure it. You want people to do things different for a reason! That is, there’s some impact. And that, to me, should be checked.

In both cases, ultimately your motivation is for something different to happen. Which should impact some organizational metric. Otherwise, why did you bother? The point being, I don’t think giving people information, or asking them to do something differently but it’s not meaningful to the org, is done just because. I think it’s because you want some thing to change, you want the audience to put it into action.

I have to ask myself, of course, whether folks might do something and not care to measure the impact. Ok, for Cover Your Assets type of stuff maybe, but that’s learning for legal’s sake, not learning’s sake. Is this just too persnickety? I suggest that it’s not, in that assuming you’ll have a good impact is kind of naive. Don’t you want to ensure that you are?

Look, it’s good at each stage to look for reasons it might not be working, and remedy thereto. But that’s doesn’t mean you can stop after one or another stage. You should have a reason, and you should therefore have an outcome that you’re looking for, and therefore that you can measure. It could just be observation, and it may be subtle and/or long-term, but I’ll suggest that there are measures. I think you do want to put learning in action, and thus have an impact. Yes, I recognize that’s an ideal, but is there really any justifiable reason to stop short? I’m perfectly willing to hear that there aren’t the resources to do more, but that’s pragmatics, not principle. On principle, am I wrong? Seriously? Let me know.

Two things I’m currently on about

15 September 2026 by Clark Leave a Comment

I’ve been talking a fair bit about AI, but that’s not really what I’m interested in right now. So maybe it’s worth talking a bit more about two things I’m currently on about. I’d welcome your feedback!

So, first, I’m on about bridging. For obvious reasons, I’ve been thinking about the gap between training and sustained impact on the business. That includes spacing (warning: PDF), reflection, planning, after-action reviews, coaching, evaluation, and more. I’ve already expressed some thoughts on these, and it’ll of course continue. The main thing is that too many orgs aren’t thinking about it. And they should, because without concrete plans to facilitate the initial learning, it’s lost. Not a good investment!

For one way my thinking on bridging will continue is that we’re beginning to see results from analyzing data being collected by the use of the Elevator 9 platform (caveat, I’ve been involved in the initial design). It appears that following the recommended spacing works. The corollary, of course, is that not following the recommended spacing doesn’t work. That includes not doing the reactivations, and also trying to do them in a rush at the end. Not surprising, but interesting nonetheless.

The other area I’m increasingly thinking is important is meta-learning, or learning to learn (see diagram). I like Harold Jarche’s Personal Knowledge Mastery, simplified, as a framework. Now, I’ve been a fan of meta-learning since at least grad school (took Jean Mandler’s class on it), and probably before. Right now, of course, I think it’s essential to make good decisions about how to use generative AI, if you’re going to. However, I reckon it’s also important in more general terms. It’s the key to innovation, for one, because (as I maintain), it’s a form of learning. (Can’t find where I’ve stated it in my blog, which doesn’t mean I haven’t, because I’m pretty sure I have!) It’s also the key to continual personal development, which means doing better at anything you care about. And you do care about something!

There likely is a third area (besides the general long-term interest in learning, technology, design, engagement, etc that are where my interests and experience lie). It’s just that I haven’t expressly thought about it, like I have the two above. I still think there’s not enough interest in the alignment between engagement and learning, which I explored in both my first book and my next to most recent one. For that matter, there’s not even sufficient understanding of learning science (topic of my sixth book), too many myths (my fifth book), and overall the wrong focus for L&D (my fourth book, if you happen to notice a trend :).

Of course, I’m basically on about helping orgs apply the cognitive and learning sciences to identify, understand, and provide remedies to problems they’re not quite sure whether or why they exist. (Hence the Learning Development Accelerator’s Learning Science Conference.) That’s the bigger picture. But, life’s short and there’s only so much one can focus on at any one time. So, bridging and meta-learning are two things I’m currently on about, while I continue to track what’s happening in all the related fields I care about. Am I missing anything?

A layer above

8 September 2026 by Clark Leave a Comment

I recently wrote a post about a secure, sustainable way to take advantage of artificial intelligence (AI, and all forms). The point was to not leave you vulnerable. And, I don’t think I was wrong, but I was a wee bit too limiting. Elizabeth Dalton (who I’m working with via Elevator 9) pointed me to an article that argues in much more depth about the necessary structure. So, my learning is that there’s a layer above that’s necessary as well.

In the post, I wrote about using small hybrid agents to do things, instead of using one monolithic entity to address all the needs. It was sort of a ‘use the right tool for the job’ article. And I do think it makes sense, for multiple reasons. For one, you control what you release when you control where and how when you host your own solutions. For another, they can be run securely on your or hosted storage. That means without connecting to anyone else. Yes, you do have to develop the systems, but that’s become relatively easy. And, you own what you create. That’s important.

What this article adds is the integrating layer. What they call sovereign AI, and that’s a term I’m going to use more frequently. You need to have the infrastructure that hosts the agents to be something you build and own as well. The article was by an entity that I don’t particularly know, but I trust more because it’s open. And, my inference is that there are other ways to do it, but the important thing is that you do do it.

If my understanding is correct, under such a scheme you can have an agent that’s one of those big ones (from Anthropic or OpenAI or…) in there as well. Just that they’ve containment and constraints, and so given specific access. It doesn’t allow the systems to run rampant, hacking your security for instance.

To me, that integration, that complete ownership, is the only scrutable way to do things. I’m not a developer, so I can’t weigh in more, but the article talks not only the importance, but the steps. A layer above is a necessary extension to the core idea of ‘build your own’ that has big and long-term implications for a scrutable and successful implementation. Why would you want anything else?