{
  "video_id": "nu5TH_lyjqg",
  "channel_slug": "techwithtim",
  "channel_handle": "techwithtim",
  "title": "I Went to the Biggest AI Infrastructure Conference",
  "duration_seconds": 763.0,
  "url": "https://www.youtube.com/watch?v=nu5TH_lyjqg",
  "upload_date": "",
  "transcript": "AI agents are everywhere, but almost no\none's shipping them reliably. And the\nreason for that is pretty simple. Most\nAI agents now are just Python scripts\nrunning in a loop. They work great in a\ndemo, but as soon as it gets to\nproduction, all kinds of things fail. An\nAPI times out, a server restarts, your\nprocess dies at step six of 10, you name\nit, all kinds of things happen, and then\nyou're stuck writing all type of retry\nlogic, state management, and those are\nthings that you're definitely going to\nget wrong. Now, I've dealt with this\nfirsthand. The AI part usually works\nfine, but it's the orchestration around\nit that can become an absolute\nnightmare. Think about long-running\nstate, failures, rate limiting,\nauthorization issues, all of these come\nup in production, and you need a way to\nhandle them. Now, this is what durable\nexecution solves. Instead of you\nhandling all of the state and writing\nall of the retry logic, a platform can\nhandle it for you, and then you just\nworry about building the AI. If\nsomething crashes, it can pick up\nexactly where it left off, with no\nretries and no duplicated work. Now,\nTemporal is the platform behind durable\nexecution. It's open source, it has SDKs\nin over five different programming\nlanguages, and it's free to use from\nyour own machine. Now, this handles\nretries, rate limiting, pretty much\nanything as it relates to orchestration\nof AI agents, and it makes it really\neasy to build and deploy agents at\nscale. In fact, it's already used by\ncompanies like OpenAI, Netflix, because\nof the capabilities it has. Now, today\nI'm at Temporal Replay in San Francisco.\nI'm going to dive into this more in\ndepth. I have a ton of sessions to\nattend. Let's get started. So, I'm just\non my way to my first talk here on AI\nagents in Python using Temporal. Looks\npretty exciting. Let's check it out.\nNow, today I'm at the Temporal Replay\nconference here in San Francisco. I've\ngot all kinds of exciting talks about AI\nagents. Let me bring you along and show\nyou what I learned.\n>> All right, so day one was packed with a\nton of workshops related to Temporal and\ntheir SDKs. Now naturally, I attended\nthe Temporal for Python session where we\nlearned about how to write durable AI\napplications. Now we ran through five to\nsix different demos where we went from\nwriting AI agents from scratch to using\nall of the Temporal primitives to make\nour applications traceable, observable,\nand most importantly, durable. [music]\nAnd by the way, what does durable even\nmean? Well, it means your workflow state\nis never lost. Now normally, if your app\ncrashes, a server restarts, or some API\ncall fails halfway through, you need to\nmanually handle the retries, the state\nrecovery, and all of that complexity\nyourself. Now Temporal solved this by\nstoring the entire workflow state inside\nof the Temporal server. So your code can\npause, fail, retry, or even continue\nrunning days later exactly where you\nleft off. Now your workers execute the\ncode, while the Temporal server acts as\nthe source of truth that keeps track of\nevery step in the workflow. Now we'll\ndive into that more later because after\nmy session, I actually found another\ntech YouTuber, Uma Codes, and here's\nwhat he had to say.\n>> All right, guys. So I'm here at Temporal\nReplay in San Francisco. Look who I met.\nWe got Uma Codes here, and uh anyway,\nsay hi, man.\n>> Hello. How's it Hope everyone's doing\ngood.\n>> Yeah. How you enjoying the event so far?\n>> I am loving it so far. Uh I've been to a\nfew workshops, and I am learning a ton.\nYeah. Yeah, so.\n>> What would you say the number one thing\nis you've learned so far?\n>> Um I think how much they abstract from\nus so we don't have to worry about it\nwhen we design our systems. So I'll say\nthat's the the interesting thing that\nI've learned so far. Yeah.\n>> Sure. And uh do you have a lot of\nexperience building AI agents stuff? Do\nyou do it on YouTube or have or is this\nkind of the first one?\n>> I have experience building AI agents,\nbut in like very confined workflow. Um\nnot in the context of\nTemporal.\n>> Got you. Got you. Okay, so this is your\nfirst time using Temporal.\n>> Yeah, this this this would be my first\ntime. I have heard about them a lot. Um\nlike I said, the whole claim is what if\nyour code or your process doesn't fail\nor never fails, so it's super\ninteresting to actually see how that\nprocess works behind the scenes. Yeah.\n>> [music]\n>> 100%.\n>> Yeah. Thanks, man.\n>> And the day continued with plenty of\nother workshops covering some of the\nother SDKs that Temporal has available.\nAnd they even dove into their Nexus\nframework for more advanced and\ndistributed workflows.\n>> [music]\n>> But rather than just show you b-roll\nhere, I want to give you a quick demo so\nyou can see how this actually works. All\nright, so I'm on the computer here and\nI'm going to give you a quick demo of\nhow Temporal works. I'm just going to\nshow you a finished workflow. We'll\nbriefly go through the code, but of\ncourse there's a lot more to dive into\nhere to really fully understand it. So,\nthe basic idea of Temporal is that you\nhave kind of three main components. You\nhave a dev server, which I'm running\nright here, which is kind of handling\nall of the message brokering between\nlike the actual AI running and the\ndifferent clients. We have the worker.\nThis is where you'll actually run\nsomething like an AI agent. And then you\nhave the client. This is where you send\nsome kind of requests right to initiate\nsome type of workflow. So, if we just do\nsomething basic, right, like I submit a\nrequest saying what is the weather in\nDubai, we'll see that if we go over here\nto the Temporal dev server, which is\nrunning right now, we can see we have a\nrunning task. We can click into it and\nwe're able to view a timeline of all of\nthe different events that are going on.\nNow, in this case it ran quite quickly,\nbut if it was being slow, we could see\nwhat activities were slowing us down. We\ncould see exactly how long everything\ntook. And then we can step through this\nand see all of the results from any\ngiven step. And this is what I was\ntalking about with the traceability and\nobservability. This really allows you to\nhave better insights into what your AI\nagent is doing. So, you can see we\ncalled GPT-4O. This was the system's\ninstruction we had. We had some input,\nwhich is what was the weather in Dubai.\nWe can see all of the arguments that\nwent through here. And then the benefit\nof this is that if this were to fail or\nthere was some bug or something, we\nwouldn't restart the entire workflow. We\nwould just sit here and wait until this\nwas kind of rebuilt or ready to run\nagain, and then we continue from the\nother task. Now, I could show you this\nwith a failure, but I don't want it to\ntake too long. And then if we go back\nhere, let's just say I run, you know,\nanother task right here. Give this a\nsecond. Let's refresh. You can see new\none comes up and is running, and we get\na full log of everything that's going\non. Now, all you really have to do in\norder to use Temporal here is just write\na little bit of code. So, for example,\nif I go here, this is the start workflow\nfile. You can see this is kind of, you\nknow, representing the client. If we go\nto the tools workflow, you can see this\nis a particular workflow where we kind\nof have an AI agent that has access to a\nfew different tools. And then if we have\na look at these tools, we run them as\nsomething called an activity. And\nwhenever we put a tool in an activity,\nit automatically makes it durable, which\nagain gives it all of those benefits\nthat I talked about before. Now again, I\ndon't want to dive into too much of the\ncode here, but I just wanted to show you\na quick demo of how this actually works.\nThe main thing here is this Temporal dev\nserver where you can have all these\nactivities, schedules, batch, workers,\nNexus, all of this kind of stuff. You\ncan see the workers that I had. You\ncould have multiple of them. You could\nhave name spaces. And then of course,\nthe workflow itself where you're\nexecuting kind of a run of an LLM or an\nAI agent. Anyways, let's get back to the\nvideo and you'll learn a little bit more\nabout it. All right, guys. So, we're\njust wrapping up day one at the\nconference. We just did some workshops\ntoday, learned about durable AI agent\nexecution. Super cool stuff. I would\nhave thrown some demos in earlier, so\nhopefully you guys got some information.\nTomorrow, we've got a bunch of keynotes,\nspeakers, [music] other creators I'm\ngoing to talk to. Stay tuned. See you\nthere. Now, we move on to day two of\nthis three-day conference where we\nstarted the morning nice and early\n[music] with the official keynote.\n>> [music]\n[music]\n>> Now, there was a lot discussed and\nannounced here, so I want to go over\nwhat I found most notable. Now, number\none, [music] Temporal is used by almost\nall major AI companies and is essential\nto their infrastructure. OpenAI,\nMistral, Emergent, Docker, Replit,\n[music] Retool, you name it, they are\nprobably using Temporal and they're\nusing it more than ever before.\nNow, number two, Temporal has seen\nmassive adoption has become the de facto\ninfrastructure for managing AI\napplications at scale. Now, OpenAI\nthemselves grew their usage by over 60x\nin just the last year. And then number\nthree, they announced three new\ncapabilities. We had serverless workers,\nstandalone activities, and workflow\nstreams. Now, I'm not going to go into\ndetails on all of this as it only makes\nsense if you're a pre-existing user, but\nI can say that the crowd was as\nenergetic as I've ever seen for a tech\nconference [music]\nfull of a bunch of developers when these\nwere announced. And for those of you\nthat are still a little bit confused on\nwhat Temporal is, I actually met up with\nanother tech influencer, Chris, who gave\nher best shot at explaining it. Let's\nhear her.\n>> Chris, nice to meet you here at\nTemporal. We're going to have a a quick\nconversation. Can you give a quick intro\nto the audience?\n>> Hi everyone. My name is Chris. I'm\n[music] a tech content creator as well.\nAnd we met at this conference. It's\nreally nice meeting you, by the way.\n>> My question is, a lot of people don't\nknow what Temporal is that's watching\nthis video. Could you give us a quick\nhigh-level explanation of your\nunderstanding of what Temporal [music]\nis?\n>> My quick understanding,\nlike simple in simple terms, I would say\nif you are building anything, like if\nyou are a vibe coder or you software\nengineer, you know that almost any\napplication breaks\nand [music] Temporal can help you\nprevent that and catch errors early and\nhandle certain things on your behalf.\nSo, it's a very useful tool to have if\nyou're building anything pretty [music]\nmuch.\n>> For sure, for the technical people it's\nlike durable execution for, you know, AI\nagents, but observability, logging,\nretries, long-running tasks, delegating\nto an IO, you know, something like that\nwith AI agents. So, actually you put it\na lot simpler than I think I described\nit. So, what do you think so far? Day\ntwo?\n>> Day two, a lot of hype\n>> Yeah.\n>> that has been backed up by evidence\nactually.\nUh, Open AI\nwith their 600 million along with a\nbunch of other companies. So, so far so\ngood.\n>> Yeah.\n>> Loving this [music] too.\n>> Keep it going.\n>> Looking forward to day\n>> Now, what he was just talking about\nthere was Open AI and Temporal's\npartnership. And what we heard from the\nmanager of the applied infrastructure\nteam at Open AI when he was discussing\n[music]\nduring the keynote. Now, I'll leave a\nlink to the recording of the talk down\nbelow as we roll into day three.\nDay three again started bright and early\nwith another keynote, but this time\nfocused more on a technical deep dive of\nTemporal and some of their newer\nfeatures. Now, unlike many other\nconferences, [music] they actually\nshowed a real functioning demo on\nscreen. They stepped through the code\nand they showed us logs, dashboards, and\neverything else that most other\ncompanies just try to hide for their\ndear life. Now, I don't have enough time\nto really dive into all of the details\nhere, but it was [music] a nice change\nof pace from the pure marketing hype\nthat you usually see at all of these\nconferences.\nNow, you may notice the change of\nbackground here. And that's because I\nhad to leave pretty early on on the\nthird day to take my 16-hour flight,\nyes, 16-hour flight back to Dubai. So,\nI'm here now and I've had some time to\nkind of reflect on what I've learned and\nI want to leave you with this. Now, AI\nis not going anywhere. It's being\nintegrated deeper and deeper into our\ndaily lives, and as a developer, we're\nbeing asked more and more to build it\ninto applications. Now, the issue is\nthat AI breaks, right? It fails, it\ntimes out, and the more we try to scale\nit, the more of these issues we run\ninto. Now, this turns our apps into\nmaybe 5% code for calling LLM APIs and\n95% for managing all of the issues that\ncan occur. Now, this is the exact issue\nthat Temporal solves, and it's why so\nmany companies are scaling their\nadoption of it right now. Now, this\nmakes sense given that it's free, open\nsource, and available across all of the\nmajor languages, so you can adopt it\nwith no risk whatsoever. You don't need\nto pay for anything. If you haven't\nalready, give it a try, and I'm sure\nonce you do, you're going to see what\nI'm talking about here, and you'll\nunderstand why all of the top companies\nare racing to adopt it. Anyways, guys,\nthat's all I have for you in this video.\nI hope you enjoyed this, and if you like\nthis style of video, definitely let me\nknow. I'm happy to make more. I'll also\nbe doing a Temporal deep dive soon, so\nstay tuned for that, where we actually\ndive into the code, and I show you\nexactly how it works and how you can\nbuild with it.",
  "transcript_chars": 13772,
  "ingested_at": "2026-06-17T16:31:32.363092+00:00",
  "source": "channel",
  "yt_meta": {
    "view_count": null,
    "like_count": null,
    "channel_id": null,
    "categories": null,
    "tags": null
  }
}