988: In Case You Missed It in April 2026

0:00Jon Krohn:This is episode number 988, our In Case You Missed It in April episode.

0:08Jon Krohn:Welcome back to the Super Data Science Podcast. I'm your host, Jon Krohn. This is an In Case You Missed It episode that highlights the best parts of conversations we had on the show over the past month, in this case, April 2026. My first clip is from episode number 985. Regular listeners will know that I have a background in neuroscience, and I was particularly excited to speak to Oracle's director of AI developer experience, Richmond Alake, about his thoughts on agent memory and how its foundations cross-sect with animal memory systems. I have a neuroscience background. And so these are the kinds of things, doing a neuroscience undergrad in particular, where you're just kind of serving across different psychological disciplines.

0:52Jon Krohn:I spent a lot of time learning about different kinds of memory. And so it's interesting now to see a lot of those same kinds of animal memory systems being replicated in agents. And so according to you, agent memory can be broken down into four categories, episodic, semantic, procedural, and working memory. Can you break down each of those for us here? Yeah. So first thing I wanted to point out is this is the first time I have someone that can fact check me on all of this. So this is good. Live on air. Exactly. Because one of the things that we've done as, I guess, a society or humankind is look to nature to inspire our technological advances.

1:36So I always use the example of planes, right? The Wright brothers had inspirations from the early inventors of vehicles or flights that actually took inspiration from birds. So you always look to nature to try to inspire technology. Same with convolutional neural networks, right? There was an experiment by Opel and Wiesel back in the 1970s where they experimented with the visual cortex of cats to realize how neurons work. But anyway, the long story is we always draw from biology to inspire our technology advancement.

2:14Jon Krohn:Yeah, if you don't mind me interjecting just for a moment. Yeah, that Hubel and Wiesel experiment, absolutely, you know, a critical moment in the development. John, I'm so excited. This is good because I always talk about the experiment. Not everyone knows it, but of course you know it because you did neuroscience. I did, but it was also, this was the Hubel and Wiesel experiments. They were a key in for years when I was about a decade ago, when I started teaching deep learning to the public, when I was giving talks to technical or non-technical audiences, I would often use the Hubel and Wiesel experiments to illustrate how neural networks work when you layer deeply.

2:54Jon Krohn:Because yeah, they had, it's a really interesting story. If you don't mind me taking a couple minutes to explain this. It's all here. We'll switch roles. Hubel and Wiesel, they were trying to figure out how the vision system worked. And they could see in the cat's brain that there were all of these nerve fibers going from the eyes to the back of their brain, the visual cortex. And so they tried for ages and ages to, they would have a cat sitting in a harness. They'd have recording electrodes in what they could see was the right part of the brain. and they would show different kinds of images to the cats.

3:30Jon Krohn:And day after day after day, there'd be no readings until like many of the great discoveries and histories like x-rays, they discovered by accident, they were frustrated at the end of the day. And instead of taking the cat out of the harness before starting to take apart the rest of the recording equipment, they removed a slide that they'd been showing on the projector. and the slide had like a straight line edge. And as that straight line edge passed through the cat's field of vision, boom, the neurons lit up. And that's how they made the discovery that the first brain cells that receive information from the eyes, they're responsible only for straight lines at a specific orientation.

4:13Jon Krohn:And then they later discovered that that information, what they called simple cells, detecting those straight lines, it gets combined into edges and corners to get a second layer. layer, and then that second layer gets combined into a third layer to get even more complexity, more abstraction, which is exactly how deep learning systems work. Yeah, exactly. Convolutional neural networks. Exactly. The early layers there, they capture the edges, the features of the edges. Then as you get through the deeper layers, you see the abstract shape starts to form. I'm so glad that you explained that experiment better than I have ever done.

4:47Jon Krohn:It's kind of wild where we've come so far with agents and this kind of abstraction where like a decade ago, it seemed so important for us to understand how neural networks and deep learning work. And now you don't even hear people talking about deep learning that much, even though that's how LLMs work, which is how agents work. It's kind of interesting how you just keep tacking on more and more to get more complexity out of these systems, more abstraction. Yeah. And going back to the whole form, the whole memory, agent memory types that we mentioned, right? Semantic, procedural, working memory, and what was the other one?

5:21Jon Krohn:David Morgan Episodic, semantic, procedural, and working. Your list. I'm just reading it. David Morgan Exactly. My list, right? I'm going to go, my memory is not that great. I'm going to go through them and just talk about examples of how they could manifest, right? Or how they could be implemented. So let's go and keep me on a stair with the list. Let's go. It was a long flight from London to New York, so I'm a bit slower. When you're working without notes, for people who aren't watching the video version, Richmond is just freewheeling everything there. Well, the 100 days of agent memory helped.

5:52All right, so let's go with episodic. Episodic is one of the easiest ones to actually understand. Episodic are essentially memory that has some association with time, right? And the easiest one within an agent is back and forth conversation, right? So the actual conversations you're having for developers or engineers listening, this would be the user would be the assistant or the user, then the content would be the message, then you have a timestamp. The timestamp is always important in episodic because that's how you recall that particular type of memory type. Then that's episodic. Another example would be procedural.

6:30One thing that's very popular today is skills.md files.

6:33Jon Krohn:Let me just quickly before you move on from episodic, maybe like I'll try to give like an example. So, you know, you give the travel agent example. So the instance of, you know, somebody, if you imagine a human travel agent and somebody come, you're a human travel agent, somebody comes into your office on Monday, they say, they describe, you know, the kinds of things that they're looking for, cultural, lots of great food, that kind of thing, set up a trip for me. And they say, I'll be back on Wednesday and, you know, hope, hope you have something ready for me. That's like, it's an episode. It's a specific, you know, if you were the travel agent, you could mark that down in a notebook.

7:08Jon Krohn:On Monday at 10 a.m., Richmond came into my office. It's a discreet episode where you can, as a human, you can think back by Wednesday, you should be able to think back to Wednesday or think back to Monday and think, yeah, at 10 a.m., Richmond was here in my office. I remember what he's looking for. So you recall that episode. Exactly, nice. So let's go over to procedural memory. An example of that would be, within the agentic context, would be maybe workflows. And I have another content, another example of procedural memory called using the database as a toolbox. But let's talk about workflows.

7:46You can think of things like skills.md file. They're very popular now at the time of this recording. And skills.md file are basically marked down that covers instructions on how to do certain tasks that are given to an agent. I call it the SOPs for agents, essentially. Standard operating procedures.

8:04Jon Krohn:Exactly. But these are basically procedural memory because humans have the same concept of these routines and skills. We store them in a particular part of our brain. And you're the neuroscientist there, so you can fact check me on this. We have this part of our brain called a cerebellum. Cerebellum. Cerebellum, yeah. And that stores all our routines and skills. So you can actually think about that in the same context of agents as well. We are writing a bunch of skills as an MD file, and we're storing them in... People are using file systems, and we could talk about file systems versus databases for agent memory.

8:39But over in Oracle, we're experimenting with putting these skills into the database, right? And actually putting them within tables that have some different representation of this data. One of them would be a vector representation. And then we can progressively expose the skills at the right time. So rather than giving the agent a hundred skills IMD file within a context, we can just retrieve the actual skills that we need at the time, which allows you to start to scale essentially. But that's an example of procedural. Do you want to do it?

9:13Jon Krohn:I'll try to come up with an example here. So if you think of some kind of thing that you do regularly, like making scrambled eggs, Yeah. You know, it's not that complicated. And you probably don't remember the first time somebody showed you how to scramble eggs. It's not like an episode that you're like, oh, yeah. I mean, maybe. There might be some people who are like, I love the time my grandma showed me how to scramble eggs. You might have the kind of episode that you can recall. But for most people, it would just be kind of something you can make scrambled eggs now. And you don't really remember how it happened.

9:43Jon Krohn:But you just kind of figured out through experimentation. Maybe some things you read online or some things people taught you, a cookbook. There could be all these different inputs that over time created this procedure that you're aware of. You're not really, there's not any specific episodes tied to that procedure. You just know how to scramble eggs. Exactly. Exactly. That's a very good example. So let's go to semantic. Semantic can be a bit easier. So semantic is actually just, the easiest one would be world knowledge, right? You can have knowledge about a certain topic. That's semantic memory.

10:13And for agents, that could be all of the, let's say, the institutional knowledge within your enterprise data, right? How to do certain things, some abbreviations or some words. And you could just bring that into your agents to get them to actually start to have the knowledge that the same your employees have in terms of that particular use case. So that's semantic memory. There are different types, but that's the one we're going to use today. So over to you.

10:39Jon Krohn:Yeah, exactly. So this is kind of like, you know, any kind of knowledge that's in like an encyclopedia, like Wikipedia, these kinds of key pieces of information. Again, so it's different from procedural in that it's not necessarily a sequence. It's just, it's a fact. It's an association. Exactly. You know, Richmond lives in London. It's like, I know that semantic fact. It's not associated with a specific episode. It's not associated with a procedure. It's just a piece of information that I know. Exactly. And then we can talk about working memory, which is another type of memory, right? Working memory is this short term, specifically a subset of short term memory.

11:19Working memory is what you're using in real time, in the context. There's a bunch of information that is my working memory that I'm using to speak to you right now to actually interact with you. I'm not having to think for an extensive period of time, right? So that's my working memory. The best way I would describe this and within a genetic context is the context window of the LLM. That's working memory.

11:40Jon Krohn:I recommend listening to the whole interview to hear Richmond walk me through every one of the memory types he outlines. We move now from theory to practice with my next clip, which is from episode number 981. I invite Matt Glickman, who is co-founder and CEO of Genesis Computing, on the show to talk about how Genesis helps practitioners use data engineering agents to create at scale. It sounds like a no-brainer to be taking on these kinds of data engineering agents to help myself and the audience better understand how this works in practice. Are you able to walk us through one or two use cases, maybe anonymized use cases with clients of yours in terms of how you implemented your solution and what the impact has been on that business?

12:23I mean, part of, I guess, the third pillar that I'll kind of mention, and then I'll I'll take you through the example is we're agnostic to the actual data engineering tools and platforms that people are using. These platforms will come and go. The frameworks will come and go. What we're replacing or really augmenting is the people. So now everyone out, to be clear out there, anyone who's panicking, no one's losing their job. I had that question coming up. No one's losing their job. What's going to happen though is that people are not going to be hired as much. So the next wave of people who would have been the junior data engineer or the junior analyst or the junior operator or just honestly any junior role in enterprises, if you want to get kind of a little more dystopian about it, but the junior lever hiring is the problem that AI is basically going to wipe out.

13:22And it's already starting to happen. Hiring projections are just not going to happen. people have the skills already there. There's no reason to replace them, particularly in the data space because they were limiting, in fact, they just didn't have enough of them. Now you just can get 10x more power out of them and they can move on to things that they wanted to do and not the tasks that they were basically bogged down to do.

13:43Jon Krohn:Yeah, maybe I'm being too optimistic here, but don't you think there are some scenarios in some organizations where they actually will want to do more hiring because, and even of junior people, because they're being so much more productive with the individuals that they have. They're getting so much, like you said, a kind of 10xing the capability there. You were talking about how previously data engineer was this extremely stressful role. They never felt like they had enough time for anything. And I suspect that even if you're 10xing your capability, there's still a lot to be keeping an eye on and making sure of.

14:18Jon Krohn:And so as that ecosystem grows, you can imagine some organizations also being like, you know it'd be good to have a few more eyes on this and i think that the few more is the key part of this because there's um my co-founder justin published a piece that if you haven't seen on our it's on our linkedin on genesis linkedin there's going to be certain people who can who can actually be these kind of conductors of these ais at scale and and actually use them better than others right He calls it spinning plates, right? It's where, you know, he now has like six different kind of agents working in parallel on different things that we're building, right?

15:02I've only gotten to like two and a half, right? It's like, so there's going to be certain people who can come in and be these force multipliers and say like, maybe that you're working with one agent is doing something with you. I can actually command 10 of them and I'll have the context switching in my head to figure out how to put them all to work. So there'll be some of that hiring. But that's, in my opinion, that's going to be, you know, kind of the few special hires. Because the number of people who can do that, it's just a limited kind of audience. I think it's more about, like, just the impact it's going to have.

15:44Again, it's just everyone who's doing these kind of operational roles are going to be the ones that now just get a force multiplier applied to them. And I just see it hard pressed to want to bring on more humans when you can just one of 10 agents, 20 agents who do that task instead.

16:06Jon Krohn:It wouldn't be surprising to me, and I'm open to your critical feedback on this, but it seems to me like some junior hires might be more likely to be that kind of 10x agent orchestrator as opposed to two and a half x because they could be growing up in this ecosystem where it's like vibe coding first i hope so um the thing i worry about and this is you know maybe a little off topic but i think it's relevant is that is that the education system has not yet figured out how to teach AI. Yeah, a lot of places haven't. Some places are. And I think if that happens, maybe this can change dramatically.

16:52But right now, most schools do not, in high school and college, discourage people from using AI as part of their process. So it's like a double whammy. right you have basically you're going to have companies that are just you know if anything like i said if anything they're looking people who are masters of how to use this tech any people come out of school that have learned it despite their schools you know kind of not encouraging it right um so yeah it's a bit of a double whammy and it's going to happen i mean i i'm you know i'm actually trying to give back and work with you know the schools that you know my kids have gone to and just to try to express the urgency.

17:34And I think there's appetite, but it's tough. I mean, it's how do you incorporate this? How do you assign papers? Like, what does that mean? Like the idea that you're gonna be able to tell, you're not, that's a farce. No human is gonna be able to tell and it's just gonna get worse and worse.

17:49Jon Krohn:One person who might've cracked AI in education is Linda Haviv. I speak with Linda, a prolific content creator in episode number 987, about her work in lowering the barrier to access learning about AI. Yeah, let's talk about this fast-moving, not needing to code anymore, scary business. Because I think it is, I don't think anyone would argue that it's still an advantage to be literate in code, even just for keeping track of what your agents are doing. But it is, you know, as you said earlier in this episode, things have always been moving fast in this space. this is a really big change where now theoretically you don't need to be able to write code at all to be working in AI, to be training models, downloading open source model weights, fine-tuning them to some specific task, labeling the data, getting that into production infrastructure, building the website that's running that AI application in the back end, theoretically now could be done all with natural language prompts.

18:57Jon Krohn:And that is a big change. And that's really just a couple months now that you could do that. It's a couple months. And I do think, though, the more and more we're dealing with a lot of toy examples, as people move to production and scale, we're going to hit a lot more system thinking. And you know what stopped me in my tracks this week? Somebody who's not technical talking about memory. And I was thinking about that a few years ago. Who would be talking about memory? Like, it was always like an AI infra, you know you don't see this part it's invisible like you know i used to build like maps for elections right nobody gave thanks to the devops engineers okay and i do think we're in this phase where ai infra is actually becoming the problem that many people are just wording differently that they need to figure out and maybe maybe they don't need to know how like pre-filled decode desegregation or like kvcash works or stuff like that but they are saying the same thing in different ways, right?

19:50They are saying, oh, this context is not working for me. Why does it feel like it's, you know, it's lagging? Why, you know, it's things that even now that they're building and vibe coding, whether it be the founder, the AI engineer, or the AI infra engineer, you know, they're kind of all saying the same thing.

20:07Jon Krohn:Yeah. So in 2016, when you were getting, well, actually, I don't know if it was, was it 2016 that you got started in your tech career or that's just content creation? When did you start oh yeah so i it was 2015 2016 yeah i was like i was teaching myself for a year and 2015 2016 is when i got my first uh like javascript i was a javascript developer that's my first role and you you were in media still for a while so i so the funny part talking about transferable skills i think people are all dealing with non-linear career paths right now right that corporate ladder it's not exactly i don't know how it's going to play out for my kids i don't know how it's going to play out but i but i can say that i've navigated non-linear career paths maybe not in the speed of now but this is where i think transferable skills and people have engineering background it's really really a leg up and you have to like lean into it um and what i mean by that is when i was working in media i was working first on the other end of media and i was like well where would i have a leg up if i go to a random tech company i'm probably not yet you know i didn't have a traditional computer science background but i do understand the user of media because I come from media.

21:10And if I'm building the CMS or the journalist writing the article, I understand what they need because I wrote the articles also on the other end.

21:20Jon Krohn:Yes, the content management system. Exactly. It was like my life for a while. I'm speaking with my hands and I always hit it with my chin. When I used to sing on stage, I used to always hit my teeth with the mic. Sorry, tangent. There may have been a sound in the audio-only version just now, and that sound was Linda hitting her face on the microphone. You know, this is why I'm chaotic. I shouldn't drink so much coffee. But that part of transferable skills, like, it is more about now. You understand the user. You understand system thinking. System thinking, that is part of everything we're doing.

21:56If you're building, like, an agentic AI, you know, workflows, there is, like, a distributed systems layer here that you have to understand. and a person who is maybe, the people that are non-technical might be coming also with a very good transferable skill, right? They're coming with understanding their niche. They're understanding real estate. They're understanding health. They're coming from their niche. And what I think engineers bring is they understand where things are failing. It's not completely like a mystery to them. Whether they come from software development, whether they come from the infra end, you know, even someone coming from DevOps, they might not be like an AI ML expert.

22:28But if you think about it, even the training of a model is an orchestration or distributed systems problem. A mixture of expert models and how they work. If you think about it, the large language models and how they're trained, that's also, in a way, a DevOps skill set. I think a lot of it is getting abstracted, but what I think is bubbling up to the top is this AI infra challenges. To bring it back to the transferable skills point, I think leaning into not feeling like just because, yes, people could vibe code, I think everyone brings strengths to the table. Just like the culture of open source, when we build together and we're able to actually bring our strengths together, we will all win that way, right?