What Is CoffeeBreak? | AI Orchestration, Memory, and the Future of Intelligent Systems
Jake Dubin
0:00 Hi, I'm Jake Dubin. I'm a technology
0:03 Hi, I'm Jake Dubin. I'm a technology
0:03 Hi, I'm Jake Dubin. I'm a technology veteran of over 25 years professionally
0:05 veteran of over 25 years professionally
0:05 veteran of over 25 years professionally and a lifetime of technology tinkering.
0:08 and a lifetime of technology tinkering.
0:08 and a lifetime of technology tinkering. If you follow me on social media lately,
0:10 If you follow me on social media lately,
0:10 If you follow me on social media lately, you've probably heard me mention coffee
0:12 you've probably heard me mention coffee
0:12 you've probably heard me mention coffee break more than once. In fact, you
0:14 break more than once. In fact, you
0:14 break more than once. In fact, you probably seen me use that coffee emoji
0:15 probably seen me use that coffee emoji
0:15 probably seen me use that coffee emoji quite a bit. Um, today I'm going to
0:17 quite a bit. Um, today I'm going to
0:18 quite a bit. Um, today I'm going to explain what it actually is and why I'm
0:19 explain what it actually is and why I'm
0:19 explain what it actually is and why I'm building it. There's a lot happening in
0:21 building it. There's a lot happening in
0:21 building it. There's a lot happening in AI right now. Every week there's a
0:24 AI right now. Every week there's a
0:24 AI right now. Every week there's a smarter model, a faster model, a cheaper
0:26 smarter model, a faster model, a cheaper
0:26 smarter model, a faster model, a cheaper model. And honestly, this is all
0:28 model. And honestly, this is all
0:28 model. And honestly, this is all exciting. Maybe a little exhausting, but
0:30 exciting. Maybe a little exhausting, but
0:30 exciting. Maybe a little exhausting, but very exciting. And I've been in
0:32 very exciting. And I've been in
0:32 very exciting. And I've been in technology a long time, and this is one
0:34 technology a long time, and this is one
0:34 technology a long time, and this is one of the biggest shifts I've ever seen.
0:36 of the biggest shifts I've ever seen.
0:36 of the biggest shifts I've ever seen. But at the same time, I think a lot of
0:38 But at the same time, I think a lot of
0:38 But at the same time, I think a lot of people are focusing on the wrong layer.
0:40 people are focusing on the wrong layer.
0:40 people are focusing on the wrong layer. The problem isn't really whether AI can
0:42 The problem isn't really whether AI can
0:42 The problem isn't really whether AI can do impressive things anymore. It clearly
0:45 do impressive things anymore. It clearly
0:45 do impressive things anymore. It clearly can. The real problem is how all these
0:48 can. The real problem is how all these
0:48 can. The real problem is how all these systems actually work together in real
0:50 systems actually work together in real
0:50 systems actually work together in real life. That's the part that still feels
0:51 life. That's the part that still feels
0:51 life. That's the part that still feels messy. Things drift. Context gets lost.
0:55 messy. Things drift. Context gets lost.
0:55 messy. Things drift. Context gets lost. Systems break in weird ways. Sometimes
0:58 Systems break in weird ways. Sometimes
0:58 Systems break in weird ways. Sometimes things just stop talking to each other.
1:00 things just stop talking to each other.
1:00 things just stop talking to each other. AI gets things almost right surprisingly
1:03 AI gets things almost right surprisingly
1:03 AI gets things almost right surprisingly often, which is sometimes more dangerous
1:05 often, which is sometimes more dangerous
1:05 often, which is sometimes more dangerous than being obviously wrong. And
1:07 than being obviously wrong. And
1:07 than being obviously wrong. And sometimes uh people are using AI like a
1:11 sometimes uh people are using AI like a
1:11 sometimes uh people are using AI like a tool you pick up and put down. You ask a
1:14 tool you pick up and put down. You ask a
1:14 tool you pick up and put down. You ask a question, you get an answer, you move on
1:16 question, you get an answer, you move on
1:16 question, you get an answer, you move on like chats and prompts. But I don't
1:18 like chats and prompts. But I don't
1:18 like chats and prompts. But I don't think that's where this is heading. I
1:20 think that's where this is heading. I
1:20 think that's where this is heading. I think we're moving towards a world of
1:22 think we're moving towards a world of
1:22 think we're moving towards a world of continuous agents, longrunning systems
1:25 continuous agents, longrunning systems
1:25 continuous agents, longrunning systems that always work on our behalf in the
1:27 that always work on our behalf in the
1:27 that always work on our behalf in the background.
1:29 background.
1:29 background. One example I gave my wife recently was
1:31 One example I gave my wife recently was
1:31 One example I gave my wife recently was this idea of a personal chef agent. It's
1:34 this idea of a personal chef agent. It's
1:34 this idea of a personal chef agent. It's not just a chatbot that gives you
1:35 not just a chatbot that gives you
1:35 not just a chatbot that gives you recipes that's been done. I made a
1:38 recipes that's been done. I made a
1:38 recipes that's been done. I made a system whose actual job is helping keep
1:39 system whose actual job is helping keep
1:40 system whose actual job is helping keep your family fed and healthy over time.
1:43 your family fed and healthy over time.
1:43 your family fed and healthy over time. Uh the system knows what foods you like.
1:46 Uh the system knows what foods you like.
1:46 Uh the system knows what foods you like. It remembers the all your allergies. It
1:48 It remembers the all your allergies. It
1:48 It remembers the all your allergies. It tracks what's running low in your
1:49 tracks what's running low in your
1:49 tracks what's running low in your fridge. It'll remind you when you're out
1:51 fridge. It'll remind you when you're out
1:51 fridge. It'll remind you when you're out of milk. It'll suggest healthier food
1:53 of milk. It'll suggest healthier food
1:53 of milk. It'll suggest healthier food options. It might coordinate grocery
1:55 options. It might coordinate grocery
1:55 options. It might coordinate grocery deliveries. Maybe it notices you've been
1:57 deliveries. Maybe it notices you've been
1:57 deliveries. Maybe it notices you've been eating badly for a period of time. Says,
1:59 eating badly for a period of time. Says,
1:59 eating badly for a period of time. Says, "We need to start shaving some calories.
2:00 "We need to start shaving some calories.
2:00 "We need to start shaving some calories. Let's adjust the meal suggestions."
2:04 Let's adjust the meal suggestions."
2:04 Let's adjust the meal suggestions." But most importantly, it checks with you
2:06 But most importantly, it checks with you
2:06 But most importantly, it checks with you when it matters. Hey, I noticed this.
2:08 when it matters. Hey, I noticed this.
2:08 when it matters. Hey, I noticed this. Are you okay with me ordering this? Do
2:10 Are you okay with me ordering this? Do
2:10 Are you okay with me ordering this? Do you want me to adjust the meals this
2:12 you want me to adjust the meals this
2:12 you want me to adjust the meals this week? Should I prioritize budget foods
2:14 week? Should I prioritize budget foods
2:14 week? Should I prioritize budget foods or healthy options? That's the human in
2:17 or healthy options? That's the human in
2:17 or healthy options? That's the human in the loop. And that part's really
2:19 the loop. And that part's really
2:19 the loop. And that part's really important to me.
2:21 important to me.
2:21 important to me. I don't think the future's fully
2:23 I don't think the future's fully
2:23 I don't think the future's fully autonomous AI making every decision for
2:26 autonomous AI making every decision for
2:26 autonomous AI making every decision for us. I think it's systems that
2:28 us. I think it's systems that
2:28 us. I think it's systems that collaborate with us. What gets
2:30 collaborate with us. What gets
2:30 collaborate with us. What gets interesting is when you stop thinking
2:32 interesting is when you stop thinking
2:32 interesting is when you stop thinking about AI as a chatbot and start thinking
2:33 about AI as a chatbot and start thinking
2:34 about AI as a chatbot and start thinking about it more like a longunning system
2:36 about it more like a longunning system
2:36 about it more like a longunning system with responsibilities.
2:38 with responsibilities.
2:38 with responsibilities. That changes everything. Because if you
2:40 That changes everything. Because if you
2:40 That changes everything. Because if you really want an AI system helping your
2:42 really want an AI system helping your
2:42 really want an AI system helping your family over time, it's going to need to
2:45 family over time, it's going to need to
2:45 family over time, it's going to need to have different kinds of memories. Some
2:47 have different kinds of memories. Some
2:47 have different kinds of memories. Some memories going to be temporary, like
2:48 memories going to be temporary, like
2:48 memories going to be temporary, like what are we eating tonight, this week,
2:52 what are we eating tonight, this week,
2:52 what are we eating tonight, this week, what's already in the fridge,
2:54 what's already in the fridge,
2:54 what's already in the fridge, you know, what grocery store are we
2:56 you know, what grocery store are we
2:56 you know, what grocery store are we using? That might be something that you
2:57 using? That might be something that you
2:57 using? That might be something that you want to remember for a short period of
2:59 want to remember for a short period of
2:59 want to remember for a short period of time. Other memories are going to be
3:01 time. Other memories are going to be
3:01 time. Other memories are going to be long term. maybe food allergies, your
3:03 long term. maybe food allergies, your
3:03 long term. maybe food allergies, your food preferences, birthdays, health
3:05 food preferences, birthdays, health
3:05 food preferences, birthdays, health goals, um things your family
3:07 goals, um things your family
3:07 goals, um things your family consistently dislikes or likes, and you
3:11 consistently dislikes or likes, and you
3:11 consistently dislikes or likes, and you have a picky eater. Uh then there's
3:13 have a picky eater. Uh then there's
3:13 have a picky eater. Uh then there's another layer entirely. We're talking
3:15 another layer entirely. We're talking
3:15 another layer entirely. We're talking about sensitive information, right?
3:16 about sensitive information, right?
3:16 about sensitive information, right? Credit cards, medical conditions, family
3:19 Credit cards, medical conditions, family
3:19 Credit cards, medical conditions, family uh details. That stuff isn't going to
3:21 uh details. That stuff isn't going to
3:21 uh details. That stuff isn't going to you don't want that floating around
3:23 you don't want that floating around
3:23 you don't want that floating around loosely in some giant AI context window.
3:25 loosely in some giant AI context window.
3:25 loosely in some giant AI context window. The system has to have boundaries. needs
3:28 The system has to have boundaries. needs
3:28 The system has to have boundaries. needs to understand what is supposed to
3:29 to understand what is supposed to
3:29 to understand what is supposed to persist, what should expire, what should
3:32 persist, what should expire, what should
3:32 persist, what should expire, what should stay private,
3:34 stay private,
3:34 stay private, uh what what should ever leave a trusted
3:36 uh what what should ever leave a trusted
3:36 uh what what should ever leave a trusted scope. And honestly, this is where I
3:38 scope. And honestly, this is where I
3:38 scope. And honestly, this is where I think a lot of current AI thinking is
3:40 think a lot of current AI thinking is
3:40 think a lot of current AI thinking is still very immature. Right now, we're
3:42 still very immature. Right now, we're
3:42 still very immature. Right now, we're throwing massive models at everything
3:44 throwing massive models at everything
3:44 throwing massive models at everything because they're impressive. But if these
3:46 because they're impressive. But if these
3:46 because they're impressive. But if these systems are eventually going to exist
3:48 systems are eventually going to exist
3:48 systems are eventually going to exist everywhere, we're talking about your
3:50 everywhere, we're talking about your
3:50 everywhere, we're talking about your home, your business, your car, you know,
3:54 home, your business, your car, you know,
3:54 home, your business, your car, you know, factories, health care environment, uh,
3:58 factories, health care environment, uh,
3:58 factories, health care environment, uh, robots, personal assistance. Um, these
4:01 robots, personal assistance. Um, these
4:01 robots, personal assistance. Um, these systems have to become dramatically more
4:03 systems have to become dramatically more
4:03 systems have to become dramatically more efficient. Not just computationally, but
4:05 efficient. Not just computationally, but
4:05 efficient. Not just computationally, but we're talking about operationally. The
4:07 we're talking about operationally. The
4:08 we're talking about operationally. The cost to run the system has to be better.
4:09 cost to run the system has to be better.
4:09 cost to run the system has to be better. It has to make sense. The energy has to
4:12 It has to make sense. The energy has to
4:12 It has to make sense. The energy has to make sense. The energy use has to make
4:15 make sense. The energy use has to make
4:15 make sense. The energy use has to make sense. The platform has to be smart
4:16 sense. The platform has to be smart
4:16 sense. The platform has to be smart enough to decide when do I use expensive
4:19 enough to decide when do I use expensive
4:19 enough to decide when do I use expensive intelligence? When is small specialized
4:21 intelligence? When is small specialized
4:21 intelligence? When is small specialized model enough? When should I avoid the
4:23 model enough? When should I avoid the
4:23 model enough? When should I avoid the model entirely and use determinist uh
4:26 model entirely and use determinist uh
4:26 model entirely and use determinist uh deterministic logic? Because not every
4:29 deterministic logic? Because not every
4:30 deterministic logic? Because not every decision should be probabilistic.
4:32 decision should be probabilistic.
4:32 decision should be probabilistic. Sometimes the answer needs to be
4:33 Sometimes the answer needs to be
4:33 Sometimes the answer needs to be predictable, repeatable, auditable,
4:36 predictable, repeatable, auditable,
4:36 predictable, repeatable, auditable, testable. The sometimes the answer
4:39 testable. The sometimes the answer
4:39 testable. The sometimes the answer shouldn't come from AI at all. Sometimes
4:42 shouldn't come from AI at all. Sometimes
4:42 shouldn't come from AI at all. Sometimes the system should just recognize this
4:44 the system should just recognize this
4:44 the system should just recognize this requires human judgment. I need to step
4:46 requires human judgment. I need to step
4:46 requires human judgment. I need to step away. That's a thing I I mentioned
4:49 away. That's a thing I I mentioned
4:49 away. That's a thing I I mentioned earlier that I I care about. I don't
4:52 earlier that I I care about. I don't
4:52 earlier that I I care about. I don't think the future works if humans
4:53 think the future works if humans
4:53 think the future works if humans disappear from the loop completely. The
4:56 disappear from the loop completely. The
4:56 disappear from the loop completely. The system has to know what it can automate,
4:58 system has to know what it can automate,
4:58 system has to know what it can automate, what it can recommend, what needs
5:00 what it can recommend, what needs
5:00 what it can recommend, what needs approval. And once these systems become
5:02 approval. And once these systems become
5:02 approval. And once these systems become more autonomous, governance becomes
5:04 more autonomous, governance becomes
5:04 more autonomous, governance becomes incredibly important too. You have to
5:06 incredibly important too. You have to
5:06 incredibly important too. You have to know what happened, when it happened,
5:09 know what happened, when it happened,
5:09 know what happened, when it happened, why it happened, what system made the
5:12 why it happened, what system made the
5:12 why it happened, what system made the decision or what person. Otherwise,
5:15 decision or what person. Otherwise,
5:16 decision or what person. Otherwise, people are going to stop trusting the
5:17 people are going to stop trusting the
5:17 people are going to stop trusting the system completely. And that's going to
5:19 system completely. And that's going to
5:19 system completely. And that's going to happen pretty quickly. That means audit
5:21 happen pretty quickly. That means audit
5:21 happen pretty quickly. That means audit trails, decision histories,
5:23 trails, decision histories,
5:23 trails, decision histories, observability, not because it sounds
5:25 observability, not because it sounds
5:25 observability, not because it sounds enterprise friendly, because it's
5:28 enterprise friendly, because it's
5:28 enterprise friendly, because it's necessary if these systems are going to
5:29 necessary if these systems are going to
5:29 necessary if these systems are going to become part of real life. I've spent
5:32 become part of real life. I've spent
5:32 become part of real life. I've spent most of my career building systems,
5:34 most of my career building systems,
5:34 most of my career building systems, integrations, automation, real world
5:36 integrations, automation, real world
5:36 integrations, automation, real world software. And what I've learned is that
5:39 software. And what I've learned is that
5:39 software. And what I've learned is that the hardest problems usually aren't the
5:41 the hardest problems usually aren't the
5:41 the hardest problems usually aren't the flashy ones. They're they're kind of the
5:43 flashy ones. They're they're kind of the
5:43 flashy ones. They're they're kind of the boring ones. They're the operational
5:44 boring ones. They're the operational
5:44 boring ones. They're the operational ones. That's reliability. That's trust.
5:48 ones. That's reliability. That's trust.
5:48 ones. That's reliability. That's trust. That's coordination. It's sustainability
5:51 That's coordination. It's sustainability
5:51 That's coordination. It's sustainability over time. Your uptime, that's what
5:54 over time. Your uptime, that's what
5:54 over time. Your uptime, that's what really matters. Coffee Break is about
5:56 really matters. Coffee Break is about
5:56 really matters. Coffee Break is about orchestration,
5:57 orchestration,
5:58 orchestration, coordination, long-term memory, context,
6:01 coordination, long-term memory, context,
6:01 coordination, long-term memory, context, reliability, keeping systems aligned
6:02 reliability, keeping systems aligned
6:02 reliability, keeping systems aligned over time instead of just producing one
6:05 over time instead of just producing one
6:05 over time instead of just producing one impressive,
6:07 impressive,
6:07 impressive, magical, mysterious response. Honestly,
6:10 magical, mysterious response. Honestly,
6:10 magical, mysterious response. Honestly, a lot of this comes from building real
6:12 a lot of this comes from building real
6:12 a lot of this comes from building real systems and seeing where things actually
6:14 systems and seeing where things actually
6:14 systems and seeing where things actually fail is part of why I've been working on
6:16 fail is part of why I've been working on
6:16 fail is part of why I've been working on like the Jibo project. Rebuilding
6:18 like the Jibo project. Rebuilding
6:18 like the Jibo project. Rebuilding something like that reminds you quickly
6:20 something like that reminds you quickly
6:20 something like that reminds you quickly that personality and intelligence are
6:22 that personality and intelligence are
6:22 that personality and intelligence are only part of the equation. Timing
6:24 only part of the equation. Timing
6:24 only part of the equation. Timing matters, consistency matters, memory
6:26 matters, consistency matters, memory
6:26 matters, consistency matters, memory matters, trust matters. The magic
6:29 matters, trust matters. The magic
6:29 matters, trust matters. The magic usually isn't this one feature and this
6:32 usually isn't this one feature and this
6:32 usually isn't this one feature and this one cool thing. It's how everything
6:34 one cool thing. It's how everything
6:34 one cool thing. It's how everything works together. That's what I'm trying
6:36 works together. That's what I'm trying
6:36 works together. That's what I'm trying to build toward with coffee break.
6:39 to build toward with coffee break.
6:39 to build toward with coffee break. Still, I'm early. I'm still building.
6:42 Still, I'm early. I'm still building.
6:42 Still, I'm early. I'm still building. I'm still figuring things out. But I
6:44 I'm still figuring things out. But I
6:44 I'm still figuring things out. But I think this is where AI is heading over
6:46 think this is where AI is heading over
6:46 think this is where AI is heading over the next several years. And I I I think
6:48 the next several years. And I I I think
6:48 the next several years. And I I I think it's not just smarter models. It's
6:50 it's not just smarter models. It's
6:50 it's not just smarter models. It's smarter systems. Anyway, th those are
6:53 smarter systems. Anyway, th those are
6:53 smarter systems. Anyway, th those are some of the things I've been thinking
6:54 some of the things I've been thinking
6:54 some of the things I've been thinking about lately. I just want had to get
6:56 about lately. I just want had to get
6:56 about lately. I just want had to get this out there. I want to make AI
6:58 this out there. I want to make AI
6:58 this out there. I want to make AI smarter, but I want it to be useful,
7:00 smarter, but I want it to be useful,
7:00 smarter, but I want it to be useful, responsible, affordable, and sustainable
7:03 responsible, affordable, and sustainable
7:03 responsible, affordable, and sustainable in the real world. And I think we're
7:05 in the real world. And I think we're
7:05 in the real world. And I think we're heading in a really interesting period
7:06 heading in a really interesting period
7:06 heading in a really interesting period over the next few years. Some of it's
7:08 over the next few years. Some of it's
7:08 over the next few years. Some of it's really exciting. Some of it's going to
7:10 really exciting. Some of it's going to
7:10 really exciting. Some of it's going to get weird. Some of it's probably going
7:12 get weird. Some of it's probably going
7:12 get weird. Some of it's probably going to force us to rethink how we interact
7:14 to force us to rethink how we interact
7:14 to force us to rethink how we interact with technology entirely.
7:17 with technology entirely.
7:17 with technology entirely. Quite honestly, I I don't think anyone
7:19 Quite honestly, I I don't think anyone
7:19 Quite honestly, I I don't think anyone has all the answers yet. I know. I
7:21 has all the answers yet. I know. I
7:21 has all the answers yet. I know. I definitely don't. I'm just a builder.
7:23 definitely don't. I'm just a builder.
7:23 definitely don't. I'm just a builder. I'm experimenting. I'm testing. I'm
7:26 I'm experimenting. I'm testing. I'm
7:26 I'm experimenting. I'm testing. I'm trying to break things. I'm rebuilding
7:29 trying to break things. I'm rebuilding
7:29 trying to break things. I'm rebuilding them. I'm trying to understand where all
7:31 them. I'm trying to understand where all
7:31 them. I'm trying to understand where all this is actually heading. That's really
7:34 this is actually heading. That's really
7:34 this is actually heading. That's really what Coffee Break is about. So, if this
7:36 what Coffee Break is about. So, if this
7:36 what Coffee Break is about. So, if this kind of stuff, if it interests you,
7:38 kind of stuff, if it interests you,
7:38 kind of stuff, if it interests you, stick around. I'm going to be sharing
7:40 stick around. I'm going to be sharing
7:40 stick around. I'm going to be sharing more about Coffee Break. I'll put more
7:42 more about Coffee Break. I'll put more
7:42 more about Coffee Break. I'll put more Jibo videos out. I know everyone likes
7:44 Jibo videos out. I know everyone likes
7:44 Jibo videos out. I know everyone likes watching the dancing robot, AI systems,
7:47 watching the dancing robot, AI systems,
7:47 watching the dancing robot, AI systems, orchestration, memory, long running
7:49 orchestration, memory, long running
7:49 orchestration, memory, long running agents, probably a few experiments along
7:51 agents, probably a few experiments along
7:51 agents, probably a few experiments along the way, too. These are definitely
7:54 the way, too. These are definitely
7:54 the way, too. These are definitely interesting times to be a builder, and
7:56 interesting times to be a builder, and
7:56 interesting times to be a builder, and I'm having a blast building things. I
7:58 I'm having a blast building things. I
7:58 I'm having a blast building things. I hope you'll come along on this ride with
8:00 hope you'll come along on this ride with
8:00 hope you'll come along on this ride with me, and uh thank you so much for your
8:02 me, and uh thank you so much for your
8:02 me, and uh thank you so much for your time and watching me on this video.