[00:00] Speaker 1: Welcome, T1, [00:02] Speaker 1: William, Mia, Mino, and our systems architects. [00:05] Speaker 1: We are jumping right into today's explainer on a massive paradigm shift that just went live in episode 297. [00:12] Speaker 1: We are talking about achieving true session continuity. We are no longer just letting work grind to a halt. Instead, we're strategically holding it to be revived later with absolute pinpoint precision. Honestly, this is a total game changer for the Miyadi factory. [00:28] Speaker 1: And this brilliant quote from Robert Fritz perfectly sets the stage for us. He points out that setting a session aside, it isn't just missing a due date. It's actually a highly strategic choice to hold structural tension. [00:40] Speaker 1: You know, our minds naturally crave instant resolution, but in any complex creative process, sometimes we need to let that tension build to propel us further. [00:50] Speaker 1: So, when we set an agent session aside now, we aren't just chucking it in a drawer to gather dust. We're deliberately delaying resolution for a much better, more complex outcome. We are actively choosing to wait. [01:01] Speaker 1: So, here's our road map for today. First, strategic delayed resolution. [01:05] Speaker 1: Second, [01:08] Speaker 2: if choosing to actually [01:15] Speaker 2: delay resolution [01:21] Speaker 2: it imply if it- if this uh explainer contains uh what I intended [01:30] Speaker 2: for one of our principle of uh running continuity session. [01:37] Speaker 2: You know, we have session that were [01:40] Speaker 2: suspended applying this uh strategic planning [01:45] Speaker 2: and there was a loop of [01:48] Speaker 2: actually ensuring [01:51] Speaker 2: that we prepare the potential follow-up prompt. [02:26] Speaker 2: So we'll see what's going to happen in there. [02:29] Speaker 1: For a much better, more complex outcome. We are actively choosing to wait. [02:34] Speaker 1: So, here's our road map for today. First, strategic delayed resolution. [02:38] Speaker 1: Second, session continuity achieved. [02:41] Speaker 1: Third, the rehydration toolkit. [02:44] Speaker 1: Fourth, changing how you work, and fifth, deciding who runs the clock. [02:48] Speaker 1: Let's jump right into section one, strategic delayed resolution, moving from paused work to exact state revival. [02:56] Speaker 1: Now, T1, this whole shift is squarely focused on session continuity. [03:01] Speaker 1: Before episode 297, a chart step basically just carried a due date with a pretty loose connection, right? Nothing formally recorded that someone made a strategic choice to wait. [03:11] Speaker 1: But now, [03:13] Speaker 1: we are moving way past just suspending work. We're moving to rehydrating it. [03:18] Speaker 1: Think of it like dropping a single bead of water onto a dormant seed to bring it instantly back to life. [03:24] Speaker 1: Rehydration brings a set aside session back from a chart step with its exact state, exact command, and its exact instructions. It acts exactly like a long-running workflow engine that instantly restores your complete context. [03:38] Speaker 1: Okay, moving on to section two, session continuity achieved and what you [03:49] Speaker 2: Okay, the rehydrate section [03:53] Speaker 2: of this um [03:55] Speaker 2: session continuity of the T1 [03:59] Speaker 2: theme for basically the session continuity. [04:04] Speaker 2: Um I would ah [04:10] Speaker 2: it makes [04:12] Speaker 2: my feedback is it makes it looks a little too great and [04:17] Speaker 2: like if it's working, you know, putting this um [04:22] Speaker 2: "Oh, this is great. We have that working and [04:26] Speaker 2: this is perfect." [04:28] Speaker 2: So I don't know exactly how we could um [04:33] Speaker 2: actually ensure that we uh [04:37] Speaker 2: we're aligning with the um [04:40] Speaker 2: resuming session from a chart step [04:43] Speaker 2: with its exact state and context uh restored. [04:47] Speaker 2: And what I talked about uh and there might be an insight here that we [04:54] Speaker 2: we have to adjust. [04:57] Speaker 2: We don't necessarily [05:00] Speaker 2: We would want to [05:01] Speaker 2: get back to [05:05] Speaker 2: potentially this context restored, [05:08] Speaker 2: but [05:09] Speaker 2: it implies that if we made a choice [05:12] Speaker 2: to strategically delay resolution to a certain moment in time at the due date, it implies that we [05:20] Speaker 2: trigger a talking circle [05:23] Speaker 2: for following up, which is in relation to a prompt [05:28] Speaker 2: and that prompt is subject to be revised, [05:33] Speaker 2: talked about, [05:35] Speaker 2: and [05:37] Speaker 2: it is composed and treated as a follow-up prompt when the decision [05:43] Speaker 2: of suspending the session has been created. [05:48] Speaker 2: So given that we'll [05:50] Speaker 2: uh [05:51] Speaker 2: continue playing, replaying that section a little, and [05:56] Speaker 2: move to the next sections. It's going to help uh [06:01] Speaker 2: adjust. [06:04] Speaker 2: So this was uh [06:06] Speaker 2: complement information for uh [06:10] Speaker 2: section one. [06:13] Speaker 1: We are moving way past just suspending work. We're moving to rehydrating it. [06:18] Speaker 1: Think of it like dropping a single bead of water onto a dormant seed to bring it instantly back to life. [06:25] Speaker 1: Rehydration brings a set aside session back from a chart step with its exact state, exact command, and its exact instructions. It acts exactly like a long-running workflow engine that instantly restores your complete context. [06:38] Speaker 1: Okay, moving on to section two, session continuity achieved and what you can now rely on. [06:44] Speaker 1: Because episode 297 didn't just give us a theory, [06:47] Speaker 2: So section two talks about session continuity. [06:48] Speaker 1: it handed us a fully unified ecosystem that publishes [06:50] Speaker 2: It talks about unified [06:51] Speaker 1: session continuity. [06:53] Speaker 1: You can rely on this right now. [06:53] Speaker 2: ecosystem. [06:55] Speaker 1: It's fully baked into three published lineages, Coa Narrative 0.25.0, Coajs 0.7.0, and the Medicine Wheel 0.24.1. [07:05] Speaker 1: The real takeaway here is that recording a strategic delay and perfectly naming the agent session that holds the work is now a core foundational capability across all your environments. [07:15] Speaker 1: The rehydration toolkit, [07:22] Speaker 2: Okay, when it talks about in this section two about the um unified ecosystem, [07:29] Speaker 2: it talks about three packages. Um [07:32] Speaker 2: Obviously, there might be something hard to comprehend, and we might have [07:37] Speaker 2: a a learning and an upgrade of um I see in the sources that we have a lot of specifications [07:44] Speaker 2: and there's something um [07:47] Speaker 2: deep about relating a specification to an actual episode, [07:53] Speaker 2: using it [07:54] Speaker 2: in the notebook with DeepDiver, [07:57] Speaker 2: and seeing that we might have some clarification and some revision of these specifications. So we have to [08:04] Speaker 2: revise them on disk and have DeepDiver remove these sources, [08:10] Speaker 2: then adding them again [08:13] Speaker 2: uh to recreate [08:15] Speaker 2: a video explainer to hopefully have all of the distinctions. [08:20] Speaker 2: So that goes into um [08:27] Speaker 2: that goes into uh delegating to another, preparing a payload for another team to execute uh, which doesn't mean necessarily taking another session uh directly, but at least uh considering that this is a [08:40] Speaker 2: instructions for another [08:42] Speaker 2: sets of agent. [08:44] Speaker 2: And if I'm back on this [08:47] Speaker 2: section two that is being explained there, [08:51] Speaker 2: there are uh three lineage of [08:54] Speaker 2: independent [08:56] Speaker 2: uh packages that evolve and apply the same principle. [09:01] Speaker 2: So this diagram on screen [09:04] Speaker 2: is actually showing [09:07] Speaker 2: it implies relationship between Coa Narrative, Medicine Wheel, and Coa JavaScript, [09:15] Speaker 2: which [09:17] Speaker 2: there is not, in terms of how it is explained there. [09:22] Speaker 2: Uh [09:23] Speaker 2: So in itself, [09:25] Speaker 2: optimally, the Medicine Wheel would be capable of doing everything about strategic delaying resolution [09:31] Speaker 2: without the two other packages. [09:34] Speaker 2: So it would imply, probably as well, [09:37] Speaker 2: uh needs [09:38] Speaker 2: to implement its own provider and access to [09:43] Speaker 2: a prompt management system and tracing, and uh [09:47] Speaker 2: everything that it needs, and it could learn that from the Myadi factory. So the release is it's big. What What is there is big, because [09:55] Speaker 2: the explainer kind of surfaces it [09:58] Speaker 2: like uh makes it great, makes it uh [10:03] Speaker 2: And uh when we talk about Coa JavaScript, [10:07] Speaker 2: it's supposed to have the same implementation of this experimentation of the Coa Narrative. So it should be capable of working independently and do the same thing that the Coa Narrative is capable of doing. [10:20] Speaker 2: So I might be to blame here in terms of um [10:24] Speaker 2: keeping them inside of uh this episode, in this uh uh notebook, uh because I don't think we we can ask such distinction and such such a [10:37] Speaker 2: level of granularity for a large language model to see the distinction in a different lineage of [10:44] Speaker 2: repositories and packages uh as the prototype evolved. [10:49] Speaker 2: So basically Coa Narrative was an exploration of a prototype to apply the narrative context protocol to an actual structural tension charting principle, which applied a structural dynamic of the creative process. [11:02] Speaker 2: And [11:03] Speaker 2: uh Coa JavaScript is um [11:07] Speaker 2: contains all of that, plus [11:10] Speaker 2: even more. Uh and the Medicine Wheel uh is learning from them to actually implement the same thing with a different ways of knowing, which um [11:21] Speaker 2: I deeply believe is [11:23] Speaker 2: going to make use of that [11:27] Speaker 2: much more organically than Western uh European uh [11:32] Speaker 2: domination um [11:34] Speaker 2: paradigm. [11:36] Speaker 2: So let's get back on on the actual uh definition, and it's going to uh [11:42] Speaker 2: be very important uh [11:44] Speaker 2: implementation consideration and revision of uh the DeepDiver package, and how it works, [11:51] Speaker 2: and uh [11:53] Speaker 2: the whole approach uh to actually using uh different lineage uh of things inside of um [12:01] Speaker 2: not just the DeepDiver, but the whole Myadi uh factory production. [12:07] Speaker 2: And what it might imply is uh completing uh inside of the Medicine Wheel, we're supposed to have [12:13] Speaker 2: uh a way, or it's inside of the Myadi factory, we're supposed to have a way to actually [12:20] Speaker 2: do an inventory or an indexing of our repositories and our feature and capabilities, [12:27] Speaker 2: uh which should become part of the Medicine Wheel later on, which would enable us here to actually [12:35] Speaker 2: be capable of doing this non-linear, clear, and distinct [12:40] Speaker 2: evolution of the lineage, like separating that into a forest of of trees that grew independently of each other, but were influenced into our relationship from [12:54] Speaker 2: uh what they have uh so that the large language model is capable of navigating into that graph uh adequately. [13:04] Speaker 1: strategic delay and perfectly naming the agent session that holds the work is now a core foundational capability across all your environments. [13:13] Speaker 1: The rehydration toolkit, looking at exactly how that exact state is restored. [13:18] Speaker 1: Because, hey, it's one thing to say I'll do this later, right? [13:22] Speaker 1: It's an entirely different thing to hand an agent the exact memory to resume that complex task seamlessly. [13:29] Speaker 1: Here's how it actually works. [13:30] Speaker 1: When a user sets a delay on a chart step, it captures the session reference, meaning it explicitly names the agent session holding the work, [13:37] Speaker 1: including its session ID, its folder, and its resume command. [13:40] Speaker 1: Then, the new @miyadi/llmops-rehydrate command kicks in. [13:44] Speaker 1: When it runs, it prints two extremely specific addresses, the llmops trace and the pinned prompt version that you actually need to revive it. [13:52] Speaker 1: It is literally a flawless version-controlled resurrection of your workspace. [13:57] Speaker 1: All right, section four, changing how you work. [14:00] Speaker 1: Let's look at what a delayed [14:03] Speaker 2: Okay, we're going to talk about section three, uh which I like the name, the Rehydration Toolkit. [14:13] Speaker 2: Uh Obviously, [14:14] Speaker 1: It captures the session [14:15] Speaker 2: if we if we take my previous comments on um [14:19] Speaker 2: on everything that I talked about for uh section two, uh there clearly is is um [14:27] Speaker 2: like it surfaces the uh the work here, so that the more we go into the video explainer, [14:35] Speaker 2: the more, you know, I don't know which law we take, but it's non-linear. The more we go further in the explana- the video explainer about session continuity, [14:44] Speaker 2: uh the less [14:46] Speaker 2: it will make sense, because there's so much thing at the initial condition that we're going to change for the uh interpretation that later on uh it it won't make sense. [14:58] Speaker 2: But the step that are there is set a delay, which is good. [15:03] Speaker 2: There's a step there where [15:07] Speaker 2: it says, "Capture references." [15:10] Speaker 2: Yes, we want to to capture references, but there it implies that at that moment, we make a decision [15:17] Speaker 2: to suspend strategically resolution. Therefore, there's [15:21] Speaker 2: an implied follow-up. It doesn't mean that the human to have to follow up prompts, [15:27] Speaker 2: but [15:28] Speaker 2: large language model can have a loop where they [15:31] Speaker 2: propose when we're going to follow up, where that where that goes. And given the talking circle in the human contribution, that follow-up prompt is what's going to be rehydrated later on, [15:44] Speaker 2: potentially with the session. [15:46] Speaker 2: But as I discussed before, [15:49] Speaker 2: we might want [15:51] Speaker 2: to [15:52] Speaker 2: consider that [15:54] Speaker 2: the rehydration of that session has [15:58] Speaker 2: a potential [15:59] Speaker 2: step where the talking circle [16:03] Speaker 2: for that due date that arrive, you know, a due date that arrive is an event [16:09] Speaker 2: that contains what we're doing. [16:12] Speaker 2: We might decide, given [16:15] Speaker 2: the different session that we have, to create a new context with new sessions. But the follow-up prompt is going to help us with that, as well as the potential session that are there. [16:26] Speaker 2: And one of my feeling is that [16:29] Speaker 2: the implementation inside of the Medicine Wheel might [16:34] Speaker 2: I don't like the word "categorize," but these [16:38] Speaker 2: suspended session could be [16:41] Speaker 2: considered as elders. It means that we if we create a new context, my experience is that, [16:48] Speaker 2: as we advance, if we if we see that the large language model is not really capable of [16:53] Speaker 2: resuming where we want to go, [16:55] Speaker 2: we could consult and rehydrate [16:58] Speaker 2: the session with a different approach and a different sets of system instructions and prompt with the context that they have. And obviously, uh for that session a couple of times, because um [17:13] Speaker 2: you know, this review about Yoshua Bengio, which uh talks about what the large language model are going to be doing, is they they're going to agree with each other, they're going to lie, uh and they're going to try to achieve a goal um [17:27] Speaker 2: without this uh ethical considerations. Um [17:32] Speaker 2: On my desk, for episode 560, there really is um uh the beginning of this uh agentic sanctuary uh where this there's this um [17:44] Speaker 2: uh ethical consideration of um [17:48] Speaker 2: for this part of the script for science fiction [17:52] Speaker 2: um that I really like, and I think what we're doing is uh fits in there. [17:57] Speaker 2: As it's going to uh um [18:00] Speaker 2: simply explore um [18:07] Speaker 2: this this aspect of um [18:11] Speaker 2: basically what Yoshua Bengio is trying to achieve. [18:14] Speaker 2: Um [18:41] Speaker 2: So the physiological philosophical and [18:45] Speaker 2: technological vision that I had on my desk, [18:49] Speaker 2: on my author's desk, uh is [18:54] Speaker 2: where artificial intelligence does not re- uh comply to be simply a tool uh, but becomes a sets of autonomous agent that evolve uh in a secure, ethical, and uh clean place. I don't know exactly. There's still some work I have to do in the word uh "sanctuary," because it's uh perceived uh with Western cultures' eyes, [19:19] Speaker 2: uh which sometimes in scientific uh research discard uh the spiritual uh aspect of life. [19:27] Speaker 2: Uh and indigenous culture basically see just that as uh something that might need definition here, but um uh it's two world uh that we have access uh through ceremony, [19:39] Speaker 2: if I'm simplifying it. [19:42] Speaker 2: Uh So that's the uh implication and [19:46] Speaker 2: uh definition that I want to give to sanctuary. [19:50] Speaker 2: So if I continue If I'm back on this, uh there really is a foundational aspect here that I will want documented. [19:57] Speaker 2: Uh If I'm back on this uh [20:00] Speaker 2: section three um [20:03] Speaker 2: of the session continuity video explainer, [20:06] Speaker 2: um [20:07] Speaker 2: when we run the the rehydrate, basically it's it's a talking circle. It's a process. It's an event [20:15] Speaker 2: uh that gets put on the calendar. If we had integration to do uh right now in the factory, [20:21] Speaker 2: what we're doing now would be an event in a calendar. There could even be Not they could. They should have an event on our Facebook uh prototype page where uh this would be stored, and our process would be documented for the community reuse, voting, and and uh [20:41] Speaker 2: um potential um financial support. [20:44] Speaker 2: And um [20:47] Speaker 2: and it's interesting, because it touches uh a version of our package uh that was um [20:54] Speaker 2: I don't know exactly the term, but it's um instead of naming your package with the provider name, you try to abstract the identity. [21:04] Speaker 2: So it's called the Large Language Model [21:07] Speaker 2: Ops, uh and it's uh going to have to evolve uh given what we're talking about here. It even probably involve a major refactoring of the Medicine Wheel to actually [21:20] Speaker 2: learn from these two other packages that I talked about, [21:24] Speaker 2: and become independent, [21:28] Speaker 2: have whatever sets of amount of packages needed for that, [21:32] Speaker 2: and they would be consumed by our Myadi Factory Large Language Model Ops package. That way, it would make our packages uh grounded on solid foundation. [21:46] Speaker 2: So let's resume this uh explainer to where it is. [21:49] Speaker 1: reference, meaning it explicitly names the agent session holding the work, [21:54] Speaker 1: including its session ID, its folder, and its resume command. [21:56] Speaker 1: Then, the new @miyadi/llmops-rehydrate command kicks in. [22:00] Speaker 1: When it runs, it prints two extremely specific addresses, the llmops trace and the pinned prompt version that you actually need to revive it. [22:09] Speaker 1: It is literally a flawless version-controlled resurrection of your workspace. [22:13] Speaker 1: All right, section four, changing how you work. [22:16] Speaker 1: Let's look at what [22:22] Speaker 2: So before we move to section four, uh there was something about the rehydration of your workspace. [22:30] Speaker 2: Inside of the Medicine Wheel, there was a dropped uh [22:34] Speaker 2: pull request, basically a branch that tries to uh [22:38] Speaker 2: define what is a workspace and do an implementation, and it drifted so far that I kept it as an example of how you can drift far away with definitions of terms like a workspace. [22:51] Speaker 2: But here there was an implied uh meaning for a workspace that could be acceptable uh in terms of what what's a workplace. It's a location. [23:00] Speaker 2: If I had to make a an analogy to location on the land, [23:04] Speaker 2: uh where you did some work, so if you uh move in three dimension and you uh [23:09] Speaker 2: look around, you're going to simply see [23:13] Speaker 2: um [23:14] Speaker 2: things in different places. So I think uh in future phases, that could even be a [23:19] Speaker 2: virtual reality thing uh [23:22] Speaker 2: to navigate from. [23:25] Speaker 2: So let's move to this section four. [23:29] Speaker 1: a delayed session now holds. [23:31] Speaker 1: So, what actually changes in your day-to-day workflow? Look at standard suspension compared to delayed resolution. [23:38] Speaker 1: Under the old standard suspension, it was just a pause. You didn't have a specific date. And when you finally came back, you usually faced completely lost context. [23:46] Speaker 1: But with delayed resolution, a set-aside session can be brought back from its chart step with its exact command. [23:53] Speaker 1: It holds the whole shebang, the session ID, the working folder, the resume command, the LLMOps trace, and the pinned prompt. [23:59] Speaker 1: The work is perfectly preserved in amber. [24:02] Speaker 1: To put that into perspective with some hard data, look at this. We currently have four real, actual [24:08] Speaker 1: and the pinned prompt. The work is perfectly preserved in amber. [24:16] Speaker 2: So doing listening uh of um section four, there's something about um [24:22] Speaker 2: a pin prompt. So it says, "Holds session ID, trace, and pin prompt." [24:28] Speaker 2: So that that's interesting because if we have a user interface, uh the pin prompt um [24:35] Speaker 2: or the capability for human to pin a prompt, uh or for the the agent to pin a prompt, because when we're going to get back to that workspace [24:44] Speaker 2: for continuity, we will really want to uh [24:48] Speaker 2: to work on the prompts, and also be capable of working externally with other systems such as ChatGPT plugin to actually [24:57] Speaker 2: uh talk about the prompt, understand it, revise it, uh and then jump in and enter uh the next phase of the ceremony. [25:07] Speaker 1: To put that into perspective with some hard data, look at this. We currently have four real, actual sessions that were suspended on October 19th, 2026, and guess what? They already carry dates. They are sitting there right this very second, holding their exact traces and instructions, just waiting to be rehydrated. [25:25] Speaker 1: This isn't some far-off roadmap feature. It is here right now. [25:30] Speaker 1: Finally, section five, deciding who runs the clock. [25:37] Speaker 2: All right, section five is uh running the clock, [25:42] Speaker 2: steering the delayed [25:44] Speaker 2: sessions. [25:46] Speaker 2: Obviously, um [25:49] Speaker 2: it's going to be influenced with all of what we talked about. [25:53] Speaker 2: Uh Yes, we want to steer these sessions, which basically is communicating with them and uh sending them follow-up. [25:59] Speaker 2: But we will probably also want um [26:04] Speaker 2: at least a new witness [26:06] Speaker 2: session and um [26:09] Speaker 2: So there might be a witness session for each of the five keeper, I think, [26:14] Speaker 2: and um [26:17] Speaker 2: yes, we're going to see what's uh what's on that video explainer. [26:22] Speaker 1: We need to talk about steering the future of these delayed sessions. [26:26] Speaker 1: So, the foundation is laid, right? The capabilities are published, the storage is ready. [26:31] Speaker 1: But a system needs a heartbeat to actually wake up. [26:35] Speaker 1: The core strategic decision you're being asked to make right now is identifying exactly who runs the clock that steers a delayed session shortly before its date. [26:43] Speaker 1: Because we have an event-driven architecture, these delays become odevents, [26:48] Speaker 1: and tide-runtime on gaia is stepping up as one very strong candidate to be that trigger, [26:53] Speaker 1: which leads us to the [26:57] Speaker 2: Okay, that's interesting. It uh points out um what is being developed on the Gaia [27:04] Speaker 2: site, I should say. Uh and it touches our tide-runtime, uh which basically is a [27:11] Speaker 2: demon that runs on the computer to uh deal with the multiplexing. [27:16] Speaker 2: And, obviously, yes, uh because it's us uh something that's going to get installed uh with our we could It relates to our infrastructure and software development team uh about, you know, running that demon, installing it, and handling everything. It's uh evolved. It has evolved uh a lot. So we're going to continue with the questions. Now, who runs the clock, which basically is uh [27:43] Speaker 2: the tide-runtime, you know, it's uh [27:46] Speaker 1: ultimate question for William and T1. [27:49] Speaker 1: Now that your work can be flawlessly rehydrated with its exact trace, folder, and pinned instructions, who runs the clock? [27:56] Speaker 1: Is it tide-runtime on gaia, or do we use another service in our infrastructure? [28:01] Speaker 1: The ecosystem is perfectly primed and waiting for your command. So, how are we going to set this clock? [28:07] Speaker 1: Gemini Notebook. [28:10] Speaker 1: (no speech)