We're sunsetting PodQuest on 2025-07-28. Thank you for your support!
Export Podcast Subscriptions
People
N
NLW
知名播客主持人和分析师,专注于加密货币和宏观经济分析。
Topics
@NLW : OpenAI正在积极推动企业内部AI的采用,通过发布更多以客户和企业为中心的材料来加速这一进程。他们提出了六个核心的AI用例原语,包括内容创作、研究、编码、数据分析、构思与策略以及自动化,这些是AI在企业中最常见的应用领域。我认为,理解这些原语可以帮助企业更好地识别和扩展AI的应用机会,从而在收入增长、股东回报和投资资本回报方面超越竞争对手。未来,这些用例将逐渐演变为由AI代理自主完成,甚至是由多个代理组成的群体协同完成,这将极大地提高工作效率和创新能力。

Deep Dive

Chapters
AI is already used for content creation, but agents will enhance this by creating solo ghostwriter agents, channel-aware repurposing, and eventually entire synthetic creative studios.
  • AI-powered content creation is currently used for marketing campaigns, policy documents, and product releases.
  • Future agents will automate content creation, repurposing, and A/B testing.
  • In the long term, multi-agent teams will create entire marketing campaigns autonomously.

Shownotes Transcript

Translations:
中文

Today on the AI Daily Brief, the six AI use case primitives and how agents are going to change them. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. Thanks to today's sponsors, KPMG, Blitzy.com, and Super Intelligent. And to get an ad-free version of the show, go to patreon.com slash AI Daily Brief.

Hello, friends. Quick note. I'm traveling today, so I was recording this one in advance. So we are doing a main episode only. OpenAI dropped a really cool new report, and it served as a great jumping off point for a broader conversation. So I think you're going to enjoy it. And tomorrow we will be back with our normal style of episode.

Welcome back to the AI Daily Brief. OpenAI has been releasing a lot more customer and enterprise-focused materials recently. Clearly, they are putting their foot on the gas trying to accelerate adoption inside the enterprise. And one of the things that's valuable about it for our purposes is that it gives us a chance to see the big patterns that OpenAI is seeing across lots and lots of customers. The latest resource that they dropped is called Identifying and Scaling AI Use Cases, How Early Adopters Focus Their AI Efforts.

And the thing that I think is most interesting about it is what they call their six use case primitives. These are effectively domains for AI usage under which lots of use cases fall and which by thinking in those terms, people might be able to come up with their own use cases as well.

Now, the setup for this is pretty obvious. The TLDR is that tons of people are using AI. They point to a study that found that 39% of U.S. adults have already used AI, a number which has surely increased since that study was published, and pointed out that that's about double the speed at which the internet was adopted. But they also pointed out a BCG study that found that in the last three years, companies that were considered AI leaders have seen 1.5 times faster revenue growth, 1.6 times higher shareholder returns, and

and 1.4 times better return on invested capital than their peers who were not AI leaders. Now, in terms of what they're aggregating to get to these six AI use case primitives, they write, our insights come from 300 of our most successful implementations, more than 4,000 adoption surveys, and over 2 million business users. Now, these primitives are not actually the substance of the entire report, although they are the biggest part.

The report argues that the three steps to finding and scaling the most impactful AI use cases are one, identifying opportunities to apply AI by understanding what it excels at. Two, teaching employees fundamental use cases that can speed up discovery across every department. And three, collecting and prioritizing use cases that will have the biggest impact on business. We're going to focus on number two because that's where this idea of use case primitives comes in.

Now, these come from over 600 use cases sourced from OpenAI customers. They say that most fall into one of six primitives, which they define as fundamental use case types that apply across all departments and disciplines. The use case primitives include content creation, research, coding, data analysis, ideation and strategy, and automation.

And so what we're going to do now is look at what OpenAI says about the current state of those primitives, and then we're going to spend a little time zooming out into what might be in the future, and specifically how agents are going to change them.

Given ChatGPT's anchor in a large language model, content creation has always been at the top of the heap for people's business use cases. And indeed, a lot of what OpenAI points to in this document will be quite familiar to you. Some of the content creation use cases to get started with, they point out, under marketing, it's creating campaign strategies, headlines, or email campaigns, generating content outlines and first drafts, repurposing content for different audiences or channels. The

The finance team might use it to draft policy documents. A product team might use it to build product requirement documents or release notes. A sales team might use it to generate a script for calls or follow-up emails. The case study they point to is a life sciences company that saved 135 hours in their first six months from first draft email campaigns. Now this is cool, but quite clearly very, very layer one.

So how are agents likely to change this? Well, in the near-term horizon, call it the next 12 to 24 months, we're likely going to start to see solo ghostwriter agents. Think a brand-tuned copy agent that monitors style guidelines, legal rules, campaign OKRs, and drafts copy images and short videos for human sign-off.

You might also see channel-aware repurposing, where an agent schedules optimized versions across CMS, social, and ad platforms. Then as you start to zoom out, that agent might get more sophisticated, so think two to four years down the line. And obviously all of these timelines could be radically compressed at the speed that we're going. But in the second horizon, however far away it is, you might start to see that context-aware ghostwriter start to integrate audience feedback loops.

The agent could watch engagement metrics in real time, run A/B tests, and revise creative continuously like a 24/7 growth hacker. But you also might see the beginning of a coordinated creative pod, where that copy agent takes that context-aware information and the data-informed information that it's getting and now cues tasks to a small team. Think a tone tuner, a translator, a thumbnail designer, with a scheduler agent posting variants across every channel, with that new dedicated analytics agent reporting back.

Finally, in Horizon 3, our farthest out horizon, call it four plus years, you're going to have entire synthetic creative studios, multi-agent teams, think writer, designer, voice actor, producer, that storyboard, shoot, edit, localize, and place ads end-to-end.

even interacting with finance agents to set budgets. And what's more, the studio might take the form of a swarm, where it's not a writer or designer or voice actor agent, but numerous that work in parallel, working with finance, compliance, and review agents to prune options until the swarm converges on the most cost-effective on-brand campaign.

The next use case primitive is research. OpenAI writes about research today. AI is widely used for research across industries, to searching the web for relevant articles or competitive data, to more comprehensive multi-step research projects that scan the web for articles, data points, and insights. We see teams uploading long internal documents for quick insights too. One of the biggest advantages of using AI for research is that you can specify the format and structure of how the analysis is presented to you, in table format, bullet points, organized in specific sections, or cross-referenced.

Some of the research use cases they point to as good for getting started with are in the context of sales and marketing, investigating new industries or understanding competitors better, in finance, searching for benchmarks from publicly listed companies, in product, sizing new markets or researching competitors or identifying trends, in IT, searching for new vendors and rating their product strength and weaknesses, or in software engineering, reviewing API endpoints and external documentation.

But what about in the era of agents? Well, to some extent, that era has already come to research. One of OpenAI's first consumer agents, depending on how you consider or define them, is deep research. Already with a very simple prompt, deep research, whether it's from OpenAI or you're using Gemini or Grok's version of this, is going to autonomously plan, browse, triage, and synthesize hundreds of sources into analyst-level reports. This is available right now, not some far-off future state.

What's likely just a little ways off, in horizon two again, whenever that is, whether it's 12 months or 24 months or 36 months, might be something like continuous intelligence agents. Think deep research but always on. Subscribing to data feeds, patents, earnings calls, spotting weak signals, generating briefings. Maybe even pinging domain experts when their confidence is low and they're trying to figure things out. You might also start to see the first swarmification of research agents.

Imagine, for example, that you have a persistent Intel cell, not just a single agent, where a planner agent seeds sub-agents like a news crawler, a patent watcher, and an expert interviewer, and then you have a synthesis agent to merge those updates into a live dossier that pings you on threshold events. Now, the most sophisticated version of this is swarms that can interact with experts and data in totally different ways.

Imagine, for example, negotiation agents that automatically reach out to subject matter peers or their agents, schedule interviews, purchase reports, update private knowledge graphs, and maybe even debate with each other on different interpretations to come up with both consensus and conflicting views.

Use case primitive number three is coding. One of the ones we talk about most on this show, of course. AI at this point is absolutely ubiquitous across coding. And as OpenAI points out, this is both existing software engineers using AI for things like debugging, generating first draft code in unfamiliar languages, or porting code from one language to another, but also non-coders starting to build with code for the first time.

Interestingly, OpenAI points to coding use cases that aren't just for software engineers. For example, marketing might use it to build interactive charts or data visualizations. Finance is doing things like creating Python scripts to automate parts of the monthly close. And product is, of course, doing the thing that we talk about a lot here with vibe coding, which is building interactive prototypes to flesh out new product ideas.

And once again, I think you could argue that the era of agents is actually starting to impact this particular primitive. You already have dev pair agents that are watching IDE events, running tests, filing PRs, operating alongside coders in much more autonomous and comprehensive ways. Also at this point, the vibe coding tools are basically agents that are taking natural language input from non-coders and turning it into process and code that can then be used.

What's next in Horizon 2 might be something like a composable software factory. Think a spec-to-prod pipeline where planner agents break features into tasks, junior dev agents code, senior agents review, and DevOps agents ship every day. All of this is, of course, orchestrated through a shared memory.

And even farther out, Horizon 3 might be a complete self-healing system. Think monitoring agents that detect anomalies, which can then spawn repair agents that roll back, patch, or spin up new microservices with minimal human intervention.

On top of that, you likely have other related agents like governance agents recording every steps, notifying humans only when absolutely necessary, but documenting the whole thing as it happens. Today's episode is brought to you by KPMG. In today's fiercely competitive market, unlocking AI's potential could help give you a competitive edge, foster growth, and drive new value. But here's the key. You don't need an AI strategy. You need to embed AI into your overall business strategy to truly power it up.

KPMG can show you how to integrate AI and AI agents into your business strategy in a way that truly works and is built on trusted AI principles and platforms. Check out real stories from KPMG to hear how AI is driving success with its clients, and

at www.kpmg.us slash AI. Again, that's www.kpmg.us slash AI. Today's episode is brought to you by Blitzy, the enterprise autonomous software development platform with infinite code context.

Which, if you don't know exactly what that means yet, do not worry, we're going to explain, and it's awesome. So Blitze is used alongside your favorite coding copilot as your batch software development platform for the enterprise, and it's meant for those who are seeking dramatic development acceleration on large-scale codebases. Traditional copilots help developers with line-by-line completions and snippets, but

But Blitze works ahead of the IDE, first documenting your entire codebase, then deploying more than 3,000 coordinated AI agents working in parallel to batch build millions of lines of high-quality code for large-scale software projects. So then whether it's codebase refactors, modernizations, or bulk development of your product roadmap, the whole idea of Blitze is to provide enterprises dramatic velocity improvement.

To put it in simpler terms, for every line of code eventually provided to the human engineering team, Blitze will have written it hundreds of times, validating the output with different agents to get the highest quality code to the enterprise and batch. Projects then that would normally require dozens of developers working for months can now be completed with a fraction of the team in weeks, empowering organizations to dramatically shorten development cycles and bring products to market faster than ever.

If your enterprise is looking to accelerate software development, whether it's large-scale modernization, refactoring, or just increasing the rate of your STLC, contact Blitzy at blitzy.com, that's B-L-I-T-Z-Y dot com, to book a custom demo, or just press get started and start using the product right away.

Today's episode is brought to you by Super Intelligent and more specifically, Super's Agent Readiness Audits. If you've been listening for a while, you have probably heard me talk about this, but basically the idea of the Agent Readiness Audit is that this is a system that we've created to help you benchmark and map opportunities for

in your organizations where agents could specifically help you solve your problems, create new opportunities in a way that, again, is completely customized to you. When you do one of these audits, what you're going to do is a voice-based agent interview where we work with some number of your leadership and employees.

to map what's going on inside the organization and to figure out where you are in your agent journey. That's going to produce an agent readiness score that comes with a deep set of explanations, strength, weaknesses, key findings, and of course, a set of very specific recommendations that then we have the ability to help you go find the right partners to actually fulfill.

So if you are looking for a way to jumpstart your agent strategy, send us an email at agent at bsuper.ai and let's get you plugged into the agentic era. A fourth AI use case primitive is data analysis. OpenAI writes, AI helps anyone harmonize data from different sources, identify insights and trends, and work with complex spreadsheet data without needing advanced Excel, SQL, or Python skills.

You can provide AI with multiple spreadsheets or screenshots of dashboards to support quick analysis. It can interpret spreadsheet data, understand visual charts, and even help format your output for reporting. Some use cases that they recommend to get started with. For marketing, it's something like uploading webinar attendance data and visualizing it. For product, it's analyzing trends and social media feedback. For sales, it's mapping leads to accounts and scoring them for intent signals. And for finance, it's things like analyzing expense data and looking for trends.

So what might be the agentic extension of this? Well, imagine something that supports scheduled tasks in a more ongoing and automated way. For example, imagine a notebook agent that would chain SQL and Python, generating charts, write narrative insights, and attach citations, producing things like a Monday KPI digest without having to be prompted.

A second horizon 12, 24, 36 months out might be something like automodelers, where an agent selects an ML technique, trains, validates, and deploys predictive models, and then feeds predictions back to ops agents, basically starting to do more of the work of an entire data team.

When you get to four plus years out, you're talking about a complete data mesh swarm where you have everything from schema agents that propose changes and simulate downstream breakage to privacy agents that can veto or redact columns if personal data risk exceeds policy, lineage agents that can update catalogs and notify affected teams, all with extremely minimal human involvement.

And so you're probably seeing a pattern here, where our current paradigm and the use cases that OpenAI is talking about is humans using assistants to do their job, the most immediate agent paradigm being agents autonomously doing big chunks of that job, and the more farther out paradigm is swarms and teams of agents collaborating to actually do entire categories of work altogether, with only broad human oversight and leadership from a high-level strategic standpoint.

Speaking of strategy, the fifth use case primitive is ideation and strategy. A use case they say is popular across all teams, from brainstorming a new blog post to helping structure a document. And they point out, as models become more capable of thinking through complex problems, we're seeing many teams build strategic plans with them, taking into account their data, goals, context, constraints, and dependencies.

So some of the ideation and strategy use cases they highlight in marketing, brainstorming campaign ideas, uploading a marketing brief and asking what's missing, in finance, building a market expansion plan for a new geography, for product, uploading your PRD and identifying areas of weakness before an executive review. For sales, they point to practicing your pitch or discovery skills with voice mode.

This is definitely an area where we've seen huge improvements based on model updates. The O3 reasoning model, for example, is a massive improvement over 4.0 and 4.5 when it comes to this sort of strategic ideation. But it also feels like we're just scratching the surface. So what might be next in agent world? Well, in the immediate term, I think we'll start to see things like scenario planner agents.

Imagine agents that can run Monte Carlo simulations over market, cost, and competitor data, producing options trees with risk and ROI heatmaps for executives. Agentic Horizon 2, more like a year or two down the line, might be something like synthetic focus groups, where you have persona agents recreate target customer segments with fine-grained psychographics, creative agents that test messaging, pricing, or feature bundles against those personas, and inside agents that surface emotion curves and recommend go-to-market tweaks.

This is the area where you might really see some of the most important and powerful synthetic employees. Imagine, for example, a chief of staff agent that attends every meeting, through both voice and vision, tracks OKRs, nudges owners, reallocates budget, escalates things when strategy drifts. An AI COO is not out of the question. And of course, a single AI COO might actually be a swarm of agents managed by a single coordinator that has that function.

Lastly, AI primitive number six, automations. This is where I think a lot of enterprises start when they think about AI. OpenAI writes, automations can be simple like generating weekly competitive updates or more complex like creating a finance report for weekly executive briefings. Some of the other automation use cases they focus on, in marketing, it's things like building Slack update summaries for meeting notes.

In product, it's summarizing and sharing weekly customer insights. In finance, it's turning weekly financial data into an executive overview. In IT, it's things like uploading your software architecture as a screenshot and asking for key dependencies, risks, and opportunities for optimization. So where does this go in the world of agents?

Well, once again, this is an area where agents are really starting to come online now. I don't think they're quite as proficient as something like deep research, which really just kicks butt at what it does. But you are starting to see web use agents like Operator imitate human clicks and keystrokes to execute multi-step workflows.

The places you see organizations playing around with this are in areas like procurement, travel booking, CRM updates. And while those things are, as of this recording, still sort of nascent, I think that over the next 12 months, a lot of that is going to become completely de rigueur.

And once again, I wouldn't be surprised if part of the way that this happens is not single agents that can do a bunch of stuff, but individual agents that are good at very specific things working together in concert. So instead of a web actor agent, think a web actor pod. A form fill agent for handling invoices and expense reports. A CRM update agent for syncing meeting notes and follow-up tasks. A coordinator agent that can resolve collisions, ask for clarification on ambiguous fields, and does things like timestamp every action for later audit.

Now, where this leads to, I think, in Horizon 2 is an even more extensive orchestration layer. Imagine a fleet manager that spawns specialized agents to monitor SLAs, handing off edge cases to humans, leading ultimately all the way, in the farthest out horizon, to entire autonomous business units. Finance agents that can close books, supply chain agents that can negotiate contracts —

HR agents that run continuous pulse surveys and personalized L&D. Again, the steady pathway here is just like we discussed with Microsoft's Work Trend Index this year. From where we are now, which is humans partnering with co-pilots and assistants, to in the future, everyone being an agent boss managing swarms or armies of agents that function together in complex ways to execute comprehensive strategic priorities.

So what is going to enable all this acceleration? One is improvements in memory. The more agents can remember preferences and past contexts, the more capable these agents are going to get. A second is improvement in tool use frameworks, function calling to thousands of SaaS endpoints, or even IoT devices, robotics. These are going to greatly expand capabilities. You're going to see tons of what feel like infrastructure agents, things like built-in task schedulers, policy engines that can review and audit things or look at them in terms of safety or cost.

and allow organizations to spin up agents more confidently. And then, of course, we are going to have coordination protocols, standards that allow specialists to delegate subtasks to peers in a way mirroring real teams. Now, this isn't all going to happen overnight, but it is happening. And I think that if I have a TLDR...

It's that as valuable as it is to teach your team these six use case primitives, you need to be thinking about it not just in terms of how they use assistants and LLMs today, but how they're going to manage agents in roles like theirs in the future. For now, though, that is going to do it for today's AI Daily Brief. Appreciate you listening or watching as always. And until next time, peace.

We're sunsetting PodQuest on 2025-07-28. Thank you for your support!

Export Podcast Subscriptions