{
  "video_id": "fAxlEcXiSts",
  "channel_slug": "openai",
  "channel_handle": "openai",
  "title": "Operationalizing AI in workflows: Lee Spacagna, Solutions Engineer, OpenAI",
  "duration_seconds": 701.0,
  "url": "https://www.youtube.com/watch?v=fAxlEcXiSts",
  "upload_date": "",
  "transcript": "[music]\n>> Hi everyone. My name's Lee and I'm a\nlead solution engineer here at Open AI,\nworking very closely with financial\nservices customers across EMEA.\nToday, we've got some really exciting\nthings to show you, some of which we\nonly launched last week.\nEvery day when I work with financial\nservices institutions, they always have\none question, where can AI actually\nchange how my business runs?\nToday, there are two paths for AI\nadoption. First, giving ChatGPT and\nCodex so that employees can use AI in\ntheir daily work.\nSecond, there's systems. This is where\ncompanies are building entirely new\nproducts. They enhancing their customer\nservice, they're improving client\nadvisory, and they're working on\noperational support.\nSo, it looks like this. We've got\nChatGPT and Codex working from the\nbottom up using employees as they get\nmore AI literate.\nAnd then we've got AI systems from\ntop-down, and these are those major\ntransformational initiatives.\nBut, there's a missing layer in the\nmiddle.\nThe automation at the team and the\ndepartment level. And this is a gap that\nthe new ChatGPT workspace agents is\ndesigned to close, and that's what I'll\nbe demoing for you today.\nWhen we say agents, we mean AI systems\nthat we can delegate meaningful tasks\nto, not just ask questions of.\nAnd we can do that by using tools that\nwe already rely on, like email,\ncalendar, and those productivity apps\nthat we're using every day.\nAnd we want these agents to complete\nwork in the same way that people do.\nAnd in the last few months, there's been\na huge jump in capabilities. And we had\nanother leap last week with GPT 5.5.\nToday, agents take on complex work that\nused to take hours or days, and they can\nhandle them from start to finish.\nBut, what happens when you need to build\nsomething custom, and you need to\ndelegate something that your team or\nyour departments are currently working\non?\nMany of you have already built custom\nGPTs.\nWith workspace agents, we have evolved\nthat into something much more powerful.\nWe've got a new agent builder, which\nbrings shared applications, skills, and\ndeployment all to one platform. And this\nallows these agents to work in the same\nplace that work already happens.\nSo now let's jump into the demo.\nThis is the is the standard chat GPT\ninterface that I'm sure you're all\nfamiliar with.\nBut down the left-hand side, you can now\nsee we've got an agents option that we\ncan start with.\nAnd now for many teams, the challenge\nisn't a lack of work to automate, it's\nthat the work is spread across meetings,\ndocuments, emails, and other systems.\nAnd the decisions all depend on specific\ncontext.\nSo today, I'm going to show you how I\ncan quickly spin up an Agenty co-worker.\nIn this case, I want to build my own\nchief of staff agent.\nI want it to help me coordinate work,\ntrack priorities, prepare meetings, and\nhelp keep the team moving. Every\nfunction can delegate meaningful work to\nagents, and these can understand the\nrole, use the right tools, and they can\noperate with how the team already works.\nHere, I'm going to use one of the\ntemplates we've got already for the\nchief of staff agent here, and you can\nsee this already has a set of\ninstructions. It already has a set of\ntools that it's able to work with, and\nit already has capabilities that I can\nstart using.\nNext, you can see here I want to start\nconnecting some of those tools that I\nmentioned to make sure it's correct for\nmy workflow. In this case, I'm going to\nuse the Microsoft set of tools here. So\nwe've got things like Outlook Calendar,\nTeams, and then my Outlook email.\nNow, we can see that the instructions\nare going to be automatically written by\nanother agent. So you don't need prompt\nengineering skills, you don't need any\ntechnical skills at all.\nBut what it means is that as a business\nuser, you can you now use an agent to\nbuild another agent for you just with\nnatural language.\nAnd that's it. That's the initial\nversion. I haven't written any code, and\nI've got the first version of the agent\nready to go.\nBut now I want to customize this to my\nown requirements. I want it at 9:00\na.m., I want it to run every day. I want\nit to look at all of my meetings, look\nat all of the applications, look at my\nemails that came in overnight, and\ngenerate a daily brief so that I could\narrive and be prepared for all of the\nmeetings for the day.\nSo, once again, here I just give the\ninstructions again in natural language,\ntelling it I want it to run at 9:00 a.m.\nNo technical skills needed here at all.\nAnd within a matter of seconds, we can\nsee that it's able to customize my\nagents to my team's requirements and\nwork exactly how I like to work every\nday.\nSo, once this is done, we can go and\ntest our agent.\nAnd we can now see there's two starter\nprompts underneath to get me started. If\nI want to, I can give it its own set of\ninstructions to perform for me, but you\ncan see that there's two already there\nto go. And you can think of these as\ncapabilities that have already been\nbuilt into this agent.\nSo, let's ask it to prepare the today's\nbrief.\nHere, you you can see I'm asking it to\ndo a concise brief using all of the\navailable information that I mentioned\nearlier, highlight priorities,\ndecisions, blockers, and follow-ups, and\nthen post these in the CFO team channel\nin the daily prep channel.\nAnd now we can see the agent spinning\nup.\nIt's going to start grabbing all of\nthose details, connecting into my email,\nconnecting to my calendar, and all those\nother sources that I mentioned. It's\ngoing to check all of those meetings\nthat I've got for the day. It's going to\ncross-reference that with information\nthat might be in my emails. It's going\nto pull all of the contacts needed from\nall of these sources.\nAnd the first time it runs, it's going\nto ask me to uh give permission to post\nto the Teams channel, so it's just going\nto set that up now. And we'll go and\ngive it the the approval and see how\nthat's worked.\nAnd that's it. That's now posted to\nTeams. So, let's now go and have a look\nat what it was able to generate for me.\nSo, now over in Teams, we can see in the\ndaily prep channel, we can see that\nthere's an update, so let's go and have\na look what it posted.\nAnd we can see our Chief of Staff agent\nfrom ChatGPT has gone and collected all\nthat information and then it has posted\nthat daily brief for me inside Teams,\nexactly where I want the information to\nbe for my daily work.\nSo within a couple of minutes, we've\nbuilt an agent from scratch. We've\nconnected it to tools that I use every\nday. We've given it some customized\nguidance and we now have a running chief\nof staff agent for my whole team.\nBut let's go back to the agent and take\nit a step further.\nThis This week my team have been burning\nthemselves out running from meeting to\nmeeting and they haven't had any time to\nprep in between.\nSo now let's add a new capability. I\nwant the agent to proactively research\nbefore every meeting, like having an\nexpert chief of staff who's the telling\nme who's there, what's the latest and\nthen what's the goal of that meeting.\nSo for this we need some additional\ncontext for some other tools. So now\nlet's go and add some more that are\navailable.\nI'm going to start off with SharePoint.\nThis is where I'm storing all of the\ncompany information, all the information\nthat I've taken as notes that's shared\nacross the organization. So we're going\nto add that.\nAnd next I want to add Salesforce for\nall of that CRM and all that kind of\nrich information about all the contacts\nfrom that customer.\nSo we're going to add Salesforce as\nwell.\nAnd now that's done. We've got those two\napps connected.\nYou can also add other apps that you use\nevery day here as well or even custom\napplications that you just you have\ninside your business. Next is skills.\nSkills are a way of capturing snippets\nof information instructions to perform\ncritical tasks.\nThink of these as an amazing way to\ncapture all of that those tribal\nknowledge and conventions that are\ncurrently trapped in people's heads and\nwe can turn those into repeatable\nworkflows.\nYou can see that there's two skills\nalready in use here. We've got the chief\nof staff skill and we've got a final\nbrief formatting skill.\nBut now let's go and add another one\nthat I've already been using across my\nteam. So I've already got a skill here\nfor meeting prep. This tells chat GPT\nthe way that I want this information to\nbe structured, the key information\nthat's needed, the source of this\ninformation and where I want that\ninformation to be posted. So let's go\nand add that to my agent as well.\nFinally, let's save our changes.\nAnd now I want to give the agent some\nmore instructions about what to do with\nthese new applications that I've gone\nand connected. So, again, we'll use the\nagent on the side to have a natural\nlanguage conversation and give it this\nadditional context to go and update the\nagent.\nSo, here there's all the information.\nI've just added Salesforce and\nSharePoint, add a new capability. I want\nit to be able to generate these quick\nmeeting briefs in\nin ChatGPT. And all the information I\nwant to give it is just give me the\ninformation for the next meeting.\nSo, it needs to go through here and make\nthe updates. Um and then it's it will\nalso um add a new starter prompt there\nfor me to use in a second.\nSo, now again, let's go and update this.\nAnd now we can go and deploy this and\nit's now available for the whole team.\nSo, let's go and use it.\nAnd here is my completed agent, deployed\nand ready. And now you can see we've now\ngot a third starter prompt underneath as\nwell to prepare me for my next meeting.\nNow it's going to run, pull all of those\ncontacts from those different sources\nincluding Salesforce\num and SharePoint that I went and added.\nIt's going to go and um to check all of\nthe information, put that together into\na concise brief in the way that the\nskill gave the instructions to go and\nrepresent that. And personally, I have\none of these running every day.\nUm I have my own agent that checks all\nof my emails that come in overnight, the\nimportant updates from across the\nbusiness, the things that I said I would\ndo on Slack yesterday or on calls and\nall of the contacts from the\ntranscripts. And now it means I get that\nfirst hour of my day back cuz I come\ninto the into work in the morning and\nall of my emails have a draft ready to\ngo with all of the context from across\nthe business. And it means that I can\njust go through all of those emails and\nclick send and just approve those and\nget those out to my customers. And it's\ncompletely transformed the way that I\nwork.\nSo, So we can see here we're all ready\nto go for that next meeting.\nBefore the team was understaffed and\nthey couldn't prep for those meetings,\nbut now we've enabled everyone to turn\nup as if they've been prepped by their\nown chief of staff agent.\nBut that was just one example there, but\nthis is a pattern that you can apply\nacross the business.\nWe've seen examples of agents like KYC\nonboarding, AML investigations,\nrelationship management, and more.\nThe opportunity here isn't about one\nsingle automation project. It's actually\nabout a brand new operating model. Every\nteam can spin up a role-specific agent\nto take manual work off their plate and\nhelp the business move faster.\nBut the next question is is what if\nwe've got thousands of these agents? How\ndo we manage them?\nAnd that's where Frontier comes in.\nFrontier is our platform for deploying\nand managing agents at scale.\nIt connects to systems usually in silos,\nthings like data warehouses, CRM, and\ninternal applications.\nIt gives AI co-workers the same shared\ncontext that the teams currently rely\non.\nAnd from there, agents can reason over\ndata, they can run code, they can use\ntools, and they can take actions all in\na governed environment.\nAnd the key thing here as well is as\nthey work, the system improves. They\nwill learn from interactions, they will\nevaluate their performance over time,\nand it means that the more they do, the\nbetter they get, just like human workers\nin the business right now.\nSo today it's possible to build in\nChatGPT, Codex, and the API, and we want\nto make it easier to deploy\nout-of-the-box agents, plugins, and\nskills all specific for financial\nservices workflows.\nThis matters because it moves the system\ntowards much more automation.\nWe can use purpose-built agents that\nplug directly into work, and they can\nhandle all of the repeatable processes\nwith even less lift and customization.\nWith all those foundations in place,\nagents become incredibly powerful,\nallowing you to delegate more workflows\nto AI over time. And next, Stephanie is\ngoing to show you how teams are using\nthem to create transformative impacts\nacross the workforce. Thank you.\n>> [applause]",
  "transcript_chars": 12728,
  "ingested_at": "2026-06-18T10:32:02.546658+00:00",
  "source": "channel",
  "yt_meta": {
    "view_count": null,
    "like_count": null,
    "channel_id": null,
    "categories": null,
    "tags": null
  }
}