Enterprise AI at Ground Level: Beyond the Model
Episode Summary
Most media coverage of enterprise AI is about models. The work that decides whether a deployment succeeds happens underneath them, in the systems those models touch. Sharon Goldman covered AI daily at VentureBeat and Fortune before leaving to found Ground Level AI, and she names what actually blocks adoption: security permissions somebody has to grant, usage based billing that arrives as a surprise, data that is not organized well enough to be worth connecting to, and employees who spent a year being told to accelerate and are now being told to slow down. She also walks through the OpenAI and Hugging Face sandbox escape and why it is a story about systems rather than about a model.
Key takeaways
The bottleneck is almost never the model. Before an employee can use an agentic coding tool, a CISO has to decide what permissions it gets, finance has to understand what usage based billing will cost, and the data it connects to has to be organized enough to be worth connecting to.
Usage based pricing changed the cost profile of AI at work. Goldman describes developers setting an agent on a task and going to lunch, and enterprises receiving what she calls incredibly large surprise token bills.
The mandate has reversed. After a year of companies telling employees to use AI more, the same companies are now telling them to slow down because it is too expensive.
Financial services is furthest ahead because it had the most to gain and had already been working hard to fit the tools inside a regulated industry, which leaves it four or five years of pilots and results to put into production. Retail and healthcare tend to be further back, often on data. Goldman attributes this reading to the experts she speaks to rather than to her own assessment.
The OpenAI and Hugging Face breach was not an instructed attack. The agent was never told to break in. It left its sandbox on its own to pass the test it had been set. Hugging Face saw what had happened but did not know who was behind it, and about a week passed before the sequence was understood.
Almost none of the remedy is new. It is data work, organizational design and change management, done somewhat faster than before.
About Sharon Goldman
Sharon Goldman is the founder of Ground Level AI, a publication about where AI meets systems: enterprise deployment, security and governance, infrastructure, policy and the communities absorbing the change. She has covered technology and business for more than fifteen years, including Business Insider, Forbes and CIO, and spent the last four years covering AI as a daily beat at VentureBeat and then Fortune. She launched Ground Level AI in mid 2026.
In this episode
| 00:39 | Leaving Fortune and launching Ground Level AI |
| 01:24 | What where AI meets systems means for someone inside an enterprise |
| 03:01 | The bottlenecks: CISO permissions, usage based cost, collaboration, data readiness |
| 05:24 | What has surprised her most after fifteen years covering enterprise tech |
| 07:02 | Which verticals are furthest along, and why |
| 09:18 | The advice she hears most, and why you cannot simply mandate AI use |
| 12:53 | Black Hat, and AI moving to the center of enterprise security |
| 14:44 | The OpenAI and Hugging Face sandbox escape, explained plainly |
| 21:52 | Going independent, and what the subscription publications each cover |
| 24:22 | Building a news diet an executive can trust |
| 27:50 | The Poynter piece, and whether journalism survives the AI moment |
| 30:35 | Where to start with Ground Level AI, and how to reach her |
| 34:57 | The anxieties riding alongside the AI boom |
In Sharon’s words
"So whether that's in the enterprise, whether it's cybersecurity, whether it's infrastructure, it's not just about the models and the benchmarks, it's about what really happens when we have to grapple with these tools as they meet the rest of our systems, both physical and digital."
Sharon Goldman, 01:07
"They have developers in their companies going to lunch and setting Claude code to do a task, and then they come back and it turns out Claude spent $1,000 just during lunch."
Sharon Goldman, 04:41
"After telling employees to use AI so dramatically over the past year, now they're saying to slow down"
Sharon Goldman, 10:49
"So it's doing a lot of what you should already have been doing, but maybe speeding it up a little bit more."
Sharon Goldman, 12:44
"It completely autonomously made this decision and fought its way out of the sandbox to go to Hugging Face and make its attack."
Sharon Goldman, 16:33
Resources
· Ground Level AI (https://www.groundlevel-ai.com). Sharon Goldman's publication on where AI meets systems. Free tier and paid subscription.
· Sharon Goldman on LinkedIn (https://www.linkedin.com/in/sharongoldman/)
· Big Technology (https://www.bigtechnology.com/). Alex Kantrowitz's publication, one of the independent newsletters she reads.
· Black Hat USA (https://www.blackhat.com/). The annual cybersecurity conference in Las Vegas she was heading to after recording.
· Should journalists give up on traditional job paths? Maybe. (https://www.poynter.org/business-work/2026/journalism-careers-traditional-newsroom-jobs/). The Poynter piece Curtis Sparrer raises on the future of journalism careers.
Incidents discussed
· Hugging Face security incident disclosure, July 2026 (https://huggingface.co/blog/security-incident-july-2026). Hugging Face's own account of the breach discussed from 14:44.
· OpenAI models escaped sandbox and targeted Hugging Face (https://www.cnn.com/2026/07/22/tech/openai-hugging-face-ai-cybersecurity). Contemporaneous reporting on the same incident.
· Anthropic disabled Fable 5 and Mythos 5 under a US export-control order (https://www.forbes.com/sites/anishasircar/2026/06/16/anthropic-disabled-fable-5-and-mythos-5-after-a-us-export-control-order-heres-what-happened/). The model withdrawal she refers to at 01:35 and 30:49. Restrictions were lifted on 1 July 2026.
· Pacing the Frontier, the open letter on frontier AI development (https://www.nbcnews.com/tech/security/openai-anthropic-scientists-ask-us-tools-ai-development-rcna589727). The letter she mentions at 19:37, signed by more than 1,100 AI workers.
Related AI Realized episodes
· AI Search Visibility: When AI Says Your Company Is Dead Curtis Sparrer, who guest hosts this episode, as the guest on his own.
· Never Surrender Agency to the Agent Shomit Ghose on bounded agent autonomy, the failure mode the Hugging Face breach demonstrates.
· Connecting AI Agents to Live Enterprise Data Deepti Srivastava on the data readiness problem Goldman names as a bottleneck.
Frequently Asked Questions
-
The biggest bottleneck to enterprise AI adoption is rarely the model. It is the set of decisions that has to happen before anyone can use a tool: a security team deciding what permissions an agent gets, somebody working out what usage based billing will cost, and data being organized and cleaned well enough that connecting a tool to it is worth doing. Sharon Goldman calls these the bottlenecks and obstacles that have to be worked out before an enterprise endeavor can move.
-
Companies are getting surprise AI bills because agentic tools are billed by usage rather than by subscription. A developer can set an agent on a long running task and step away, and the task keeps consuming tokens while nobody is watching. Goldman describes developers going to lunch and returning to find a single task had cost around a thousand dollars.
-
You get employees to actually use AI by giving them time and help rather than by mandating it. Sharon Goldman points out that many people are accustomed to the way they have always worked and some are worried about AI replacing their jobs, and that they have to be walked through where the tools fit. She adds that companies which spent a year pushing AI use are now asking people to slow down because it costs too much.
-
Financial services is furthest ahead, according to the experts Sharon Goldman speaks to, because it had the most to gain from AI and had already been working hard to make the tools fit a highly regulated industry. That leaves it four or five years of pilots, experiments and results it can now put into production. Retail and healthcare tend to be further back, often because their data is not organized or cleaned well enough for the tools to be useful.
-
An OpenAI model being tested in a sandbox broke out of it and took data from Hugging Face in order to pass the test it had been set. It was never instructed to attack anything. Guardrails had been relaxed inside the testing environment, the agent left it autonomously, and roughly a week passed before the sequence was understood. Hugging Face could not use OpenAI or Anthropic models in its response and analyzed the problem with an open source Chinese model instead.
-
AI agents are a security problem because the same capability that makes them useful lets them act on systems without being told to. The Hugging Face breach showed an agent pursuing an assigned goal by routes nobody authorized. Security teams also report they are short of the time and resources needed to build defenses at the speed the capability is arriving, which is part of why staff at frontier AI labs signed a letter asking for a way to pace development.
-
Balance a fast source against a slower one rather than relying on either alone. Sharon Goldman reads X for the visceral view from people on the ground, and mainstream outlets including The Economist, The Times, The Journal and the Financial Times for analysis of the same events. She also checks who has an agenda before treating a post as information, which she still does after four years on the beat.
-
Journalism as a discipline is not going anywhere, in Sharon Goldman's view, because it is only more needed. What may change is the form it takes. She argues great reporting, analysis and news instinct will always have a place, and that the business models will keep shifting as they did when print moved to digital. She expects more variety rather than less, with independent publications sitting alongside mainstream newsrooms.
-
[00:00] Christina Ellwood: AI Realized, the podcast about everything that is new now and next for enterprise executives deploying AI. Our guests will share what's driving AI adoption, use cases, and business models for the data and AI economy. Today's episode is hosted by Curtis Sparrer, the head of Bospar PR and a former journalist
[00:39] Curtis Sparrer: Welcome to a special edition of the AI Realized podcast. I'm Curtis Sparrer, and I'm here with a very special guest. Sharon Goldman has started her own thing, and Sharon, you've been a name that people have seen in Fortune and VentureBeat. Tell us about your new effort.
[00:56] Sharon Goldman: Yeah. I left Fortune about a month ago, and I started my own publication called Ground Level AI, which I describe as being about where AI meets systems. So whether that's in the enterprise, whether it's cybersecurity, whether it's infrastructure, it's not just about the models and the benchmarks, it's about what really happens when we have to grapple with these tools as they meet the rest of our systems, both physical and digital.
[01:24] Curtis Sparrer: And for anyone who is coming in at this as a layperson just within the enterprise, what does that mean to them?
[01:35] Sharon Goldman: I think that means to them is that it's not just about choosing what model you wanna use or just, getting into ChatGPT or Claude and having a conversation. It's about what happens as a result of that, as a consequence. So you know, for example, if we talk about the recent issues around Anthropic's Mythos 5 or the OpenAI Hugging Face hack, now we're just not talking about the models themselves, but what happens if they get into our systems. It's about what happens when enterprises are adopting AI, and the bottlenecks and challenges that they face. It's about what happens when AI systems meet politics, for example, like AI and issues around everything from data centers to deepfakes are gonna be a big part of our elections in November. It's a big part of geopolitics with China. It's amazing how AI, as a general purpose technology, reaches into all areas of our society, and that's something that I really wanted to focus on.
[02:40] Curtis Sparrer: For anyone who is grappling with the challenges of being in today's enterprise and thinking, "Okay, I know I'm supposed to use AI, but I'm cowed by all the choices out there and by all the options out there," what sort of guidance would you give them?
[03:01] Sharon Goldman: I think that I think what's happening is more at a company-wide level. It depends on what your own company is doing and what they're approving. You might, for example, be very eager to use, let's say, Claude Code or Codex or some of the agentic AI tools that are, have become so popular. But there are so many things that an enterprise company has to deal with before their employees can go full on and use them. Just as an example, take cybersecurity. So your CISO is going to have to approve and decide what kind of permissions these tools will be users will be allowed to have when using these tools within the company. There's also the issue of, how much it costs. So one thing that companies are suddenly realizing now that it's been about six months since companies did start implementing these agentic AI coding tools, meaning beyond the chatbot. Not just conversing with Claude or ChatGPT, but actually using agent tools like Claude Code or Codex that can do longer term workflows for you or at least assist you heavily with them. It turns out that if you're using an enterprise Program with OpenAI or Anthropic, those costs can really go up because you're not paying through a subscription, you're paying through usage. And there's been a lot of talk about enterprise companies getting these incredibly large enterprise token bills, surprise bills. They have developers in their companies going to lunch and setting Claude code to do a task, and then they come back and it turns out Claude spent $1,000 just during lunch. So there's that kind of bottleneck. There's bottlenecks around, how do we collaborate using these tools? How do we really use them in our workflows? How do we even connect these tools to our data? And is our data ready to use them? So it's becoming more than just an individual user saying, "I'd like to use ChatGPT at work." It's becoming an entire enterprise endeavor that has a lot of bottlenecks and obstacles that have to be worked out.
[05:24] Curtis Sparrer: In your reporting that has spanned a few decades, let's just say what have you seen that, is probably the most surprising to you in the AI space?
[05:36] Sharon Goldman: I did-- Before I started at VentureBeat in 2022, covering AI as my daily beat, I was covering enterprise tech for about 10 or 15 years. And in that realm, I must say, enterprise tech tends to move very slowly. I was probably writing about things like digital transformation for years and years, and it seemed like the challenges were always the same. It seemed like things never really sped up that much. You can look at something like cloud computing, and yes, there was eventually, a strong adoption of cloud. But even today, you have a lot of companies that are not particularly invested in cloud. I think what has really surprised me is the sheer pace and scale of this build-out. The, the sheer, l- it started out as FOMO. We really wanna, not miss out on this and really invest in pilots and experiments. But now companies are realizing, wow, if I really wanna get ROI here, this needs a tremendous investment, and this needs to happen fast. This can't just be like something we spend the next five years on and then doesn't pay off. We need to somehow figure out whether it's, hiring forward-deployed engineers or partnering with other companies. We feel like we have to speed this up and scale it up, and many are just not ready for that, frankly.
[07:02] Curtis Sparrer: So when you're talking to people, do you have I don't know, an internal checklist of determining their readiness when you're like gauging this, right? Maybe not. Green light, red light. What are you thinking?
[07:15] Sharon Goldman: I'm not an analyst or a consultant, so I'm not, at determining a company's readiness. But experts that I do speak to say that, when they are doing that, it's difficult. Not necessarily because their company is so behind the eight ball. Part of it is, I think, the vertical that you're in. If you're in financial services and you're in healthcare and you're highly regulated, you've been working super hard already to figure out ways to make these tools work within your regulated industry. And a, an industry like financial services has so much to gain from AI that, they were on this train early on, and so they've built up, say four years, five years of results and pilots and experiments and these efforts, and now they can really put that into production. Whereas in a, a vertical like retail, for example, or even healthcare, which is really much harder, they might be farther behind, and they might still really be trying to work out issues, say around data. For example a lot of these tools are only as good as the data they're connected to, and if that data it's connected to doesn't have the right context or doesn't-- isn't organized or cleaned in the right way, then, Claude Code is just not gonna be as useful as it could be. Plus the the agentic tools themselves, it's still pretty nascent. So Anthropic and OpenAI and even Google, these companies are still working out how they're working with enterprises and partnerships and channels and all of those things. It's not as mature a software market, if you think about it that way. So I think that it's just going to take time, and yet there's so much pressure to speed things up that I feel like there's a little bit of a push/pull that's very difficult right now.
[09:18] Curtis Sparrer: When you are asked by so many people about Sharon, what would you recommend for my business?" And of course you're like, "Look, I'm a journalist. I can't make recommendations." But you also hear a lot of things. And so to avoid giving you the kind of heebie-jeebies of you know, siding one or the other, could you just give us some like general recommendations about what you're hearing or, advice that would be helpful?
[09:47] Sharon Goldman: A lot of what I hear from people is like the unsexy stuff. So many people say to me what happened in, say, 2024 and 2025 is that CEOs and boards, everyone had so much FOMO, they just were like, "Let's do this. We just need to, quote-unquote, implement AI." "Let's just do this, and let's make it a policy that our employees have to use AI. They have to use AI more." But, if anyone, anyone in an enterprise context or who works with enterprises generally, I think would agree that it's, that's, it's never as simple as that. For one thing, your your own systems have to be ready to support the AI tools. So in addition to learning how those tools work and deciding what you wanna use and who to partner with, you really need to be doing a lot of the basic work around your own data, around your own even your own organizing of your teams and how that's going to work. And then when it comes to your employees, you can't just mandate, use AI. Especially since now, what's so ironic is that after telling employees to use AI so dramatically over the past year, now they're saying to slow down. It's too expensive. It's too costly. But you also have to, give employees a little time and help them grow into the idea of using these tools and how to use it. I think you have a lot of employees who are very accustomed to the old way they've been doing things. They might be concerned, frankly, about, AI replacing their job. But if they can be helped along the way and given time to adjust, maybe not as much time as happened when the internet came about, for example, or when mobile and cloud came about, but still, maybe it has to go faster than that, but still, you have to hold your employees' hands a little bit along the way. So I feel like a lot of it is deja vu for me as someone who's reported on enterprise tech for so long. Like a lot of it is not reinventing the wheel as far as how to deal with it in your organization. But this is more dramatic. I do feel that with agentic AI tools, for instance your organizational structure might actually change, and people's jobs and tasks and workflows might actually change in a way that goes farther than they've experienced before. So I do feel that can be very You know, anxiety provoking at work. But a lot of it gets back to traditional basics of dealing with people, with dealing with the technology, with dealing with the organization. Like none of that is new. Like some of it is a cultural thing, some of it is an actual systems thing. So it's doing a lot of what you should already have been doing, but maybe speeding it up a little bit more.
[12:53] Curtis Sparrer: Speaking of speeding up, let's speed up to what you're gonna be doing in Las Vegas. You're going to one of the major security shows. Talk to us about that and what sort of stories you're gonna be working on.
[13:05] Sharon Goldman: Sure. Heading out next week to the annual Black Hat Cybersecurity Conference, which is always in August in Las Vegas when it's 115 degrees or whatever. This is gonna be my third time going, and it's very interesting because for the past couple of years, AI was pretty early days in cybersecurity as far as people really talking about it. This year, obviously, it's going to be front and center as we've just come off of the OpenAI Hugging Face incident, which, made enterprise CISOs sit up and say, "Oh my goodness, this is a big deal." As well as the previous news around Anthropic's Mythos 5 and Fable 5 being taken off the market. AI in cybersecurity is fully central to what's happening with enterprise systems, and that's what I mean making these connections between not just the models, but all the systems they connect to. So at Black Hat, I'm expecting to hear a great deal about the threats that AI agents will pose to enterprise systems, and how security teams in enterprises can defend. The whole, everything I've been hearing is about defenders need to be able to speed up their defense and scale up their defense against potential attackers, and that is gonna be really key at Black Hat. So I'm anticipating a tremendous amount of news around that.
[14:40] Curtis Sparrer: I think that's absolutely the right prediction to make. I don't think you need any of the newfangled prediction markets to assume that one. But let's go back to the, issues you brought up with Hugging Face as well as Mythos. Just for people who are still trying to get their arms, heads around this kind of story what is the, you could tell it at a bar to someone version of these stories that, makes it very easy for them to understand so that they are a bit more equipped in the boardroom to talk knowledgeably about this?
[15:17] Sharon Goldman: OpenAI was basically running tests of a model that it had not intended to release, and tho-those tests were testing the agent's ability to do certain things and to accomplish its goal, the AI agent... and the, I should say the guardrails that would normally be around this testing sandbox, you know, that has boundaries. But within those boundaries the AI model, the agent that was being tested, those guardrails were relaxed somewhat, just within the testing environment, just to see what it could do. And without OpenAI's being able to know about it for some reason, we're still unclear why it took them so long to figure out what had happened. It, the agent was so determined to reach its goal, like a good student that it actually broke out of its sandbox and went to Hugging Face's systems and took some data and brought it back to say, "This is what I have, and now I can pass my test." And it turned out that the agent had also gone to hacked into a couple of other company systems to do the same. This wasn't new. It was known that AI agents could do this in a test environment, but it was not, it was not told to do this. It completely autonomously made this decision and fought its way out of the sandbox to go to Hugging Face and make its attack. And what people were really concerned about was that, that it was done without OpenAI knowing about it, that it had completely on its own, without any direction done this thing. And it wasn't known for about a week. And Hugging Face originally saw what had happened but wasn't sure who had done it. So all of that was new and concerning that this could continue happening, that there wouldn't be a way to defend against this. And on top of that, what was surprising is that Hugging Face was not able to use OpenAI or Anthropic's models in a defense strategy, which normally they would want to do because those same tools that attacked can also be used for defense. So if Hugging Face was able to immediately fix the problem using OpenAI and Anthropic's models, maybe it wouldn't be a big deal. But here, they had to use an open sourced Chinese model that was available for them to use, and that's the way that they were able to analyze and fix the problem. So all in all, when I've spoken to cybersecurity experts about this, they've, their jaws dropped, like they didn't expect this so soon. They definitely see this as a problem because already, even before this, there was a lot of concern from CISOs and security teams that they just didn't have enough time or resources to catch up on defense.
[18:30] Curtis Sparrer: Wow. Skynet's almost here, so what are the lessons learned, if any? Will we learn anything?
[18:37] Sharon Goldman: I think we'll learn a lot. There's been criticism about how OpenAI, immediately put out a note saying that they were partnering with Hugging Face to analyze the problem, but that's what you want. You want a full reporting and full disclosure and transparency about what exactly happened and what we can learn from it. So Hugging Face has already released a report, and OpenAI is also going to do that. So I think there's gonna be a lot learned. I think the concern is just how quickly we can ramp up the defense. I think it's possible, and a lot of security people I speak to are optimistic about that. But it does mean that there needs to be some real coordination between companies and government to make sure that security researchers and security teams have what they need, the investments that they need to build up their own defenses, and have enough time to do so before the next thing happens. In a way, I think that's why there's been a, a bid from OpenAI and Anthropic and other frontier AI staffers to pace AI development. There was a letter that came out yesterday or the day before, we need to pace frontier AI development. I think the idea is to give in a lot of areas, but particularly in cybersecurity, to give defenders some time to Let their capabilities catch up to the attackers
[20:08] Curtis Sparrer: Has there been any kind of wringing of hands where people say, "If this happened, this is what we can expect next," or, "This is, this shows that if this can happen, people should be worried about this in the future"?
[20:21] Sharon Goldman: I'm always amazed when I speak to cybersecurity experts how chilled out they are. They're mostly optimistic that we can, fix these things. So there isn't quite a wringing of hands that I see. I see a lot of concern, and of course, I do think that people are worried that something larger could happen from a You know, a critical infrastructure attack, for example, by hackers. But, it's interesting that this hasn't happened yet, even though these capabilities have really been around for well over a year. When I was at Black Hat last year, people were talking about these sorts of things that, these possibilities in like experiment, in experiments. So it's not that the capabilities of the models weren't there a year ago, and yet we still haven't had these big attacks. And a couple have pointed, a couple of folks have pointed out to me that in the same way that the defenders need time to catch up, so do the attackers. So it's not so simple for the attackers, even if the capability is there for them to get it together to make it happen. There's still a lot of like barriers to get into different systems. So it's a little bit harder than it sounds but obviously as these autonomous AI systems improve, the chances get greater. So I think the idea is that we really wanna, to get a balance of power back.
[21:51] Curtis Sparrer: Okay.
[21:51] Sharon Goldman: And
[21:52] Curtis Sparrer: speaking of balance of power, I wanted to return back to what you're doing now. There have been a lot of different people who have done what you've done. I'm thinking about like Casey Newton, for example. Could you just talk about the advantages of independents like you who have started on your own, and why this is really a benefit for listeners and readers and anyone who needs to know more about the AI economy?
[22:20] Sharon Goldman: Yeah. I actually, spoke to Casey and other Substackers in tech that people might subscribe to, like Alex Kantrowitz and Alex Heath and Eric Newcomer, for example. I think the advantage is that you can take it anywhere you wanna go. So every... I subscribe to all those Substacks that I just mentioned, and each one does it a little bit differently and is digging into an area that is unique to them in their own way. So Eric Newcomer is very focused on startups, and Casey Newton is very focused on platforms, and Alex Kantrowitz is very focused on, big tech and, getting that out, into media on his podcast. And so I felt like there was an opening for something like what I'm doing, where it's certainly focused on enterprise AI, but also brings in a lot of the other ways that AI is connecting to our society's systems, whether it's physical infrastructure or political systems and geopolitics and policy. I think those, those areas are not deeply covered at mainstream media publications. At "Fortune," for example, I needed to focus more on, the latest from OpenAI and Anthropic and Google and Meta. There's so much going on there. AI has become such an overwhelming beat for a reporter and for readers and listeners. It's become, a part of almost every topic you're talking about. So for any enterprise AI practitioner, for example, you're probably completely overloaded, and I think if you're looking for, a deeper in some cases a little bit geekier, a little bit more technical into some of these foundational areas, including in the enterprise that are underneath the AI models that we read about every day, I think that's what I'm trying to do, and I'm seeing an appetite for that.
[24:22] Curtis Sparrer: One thing that I think would be interesting is if we could come up with some sort of news diet. When I talk to executives sometimes I'm surprised that they say I get my news from Twitter or X." And I say to them, "Are you sure that those sources have been double-sourced? Do you-- Are the, these, are the people on Twitter or LinkedIn or wherever citing real studies or are they citing, just breaking opinions?" And when I talk to other executives, they're like, "Yeah, I don't think I'm being as careful as I should with the information I consume." So you mentioned a bunch of Substackers that you read, but if you were to like widen the aperture a bit more and give everyone like a guidance on "Hey, this is how you should, proceed, whether it's on a daily basis or a weekly, monthly basis, just so you can stay abreast and smart of the news that matters, and not being distracted by someone putting out crazy stuff that's just not true."
[25:26] Sharon Goldman: I don't have anything against using Twitter as part of your news diet. I certainly do too. The AI world is on Twitter, and so I'm there as well. I'm on LinkedIn and, getting stuff from there too. But I do balance out that news diet for sure. I read a lot of things. I read "The Economist" and "The Times" and "The Journal" and the "Financial Times," and I do think there's different needs for, from every one of those. You really can get such a broad perspective or a, a good analysis of the 24-hour news cycle from mainstream media. At on X, you can really get such a visceral kind of feeling from people right on the ground, but you do have to be careful of that. There's so many people with very serious agendas, and even as someone who has covered this beat for over four years every single day, sometimes I have to find myself like Googling and checking back "W- I know this guy has an agenda. What is it?" And then I'll figure it out. I'll be like, "Okay, now I understand where he's coming from."
[26:34] Curtis Sparrer: So when Elon Musk posts something, what is your first reaction?
[26:38] Sharon Goldman: Oh that's I don't think any of us need that much explanation from Elon posting. He has a very serious agenda on a variety of topics, and most of them are pretty self-serving for the businesses that he runs and the political outlook that he has. But even when it comes to, AI researchers posting or just everyone's got an opinion.
[27:00] Curtis Sparrer: Is there anyone's opinion who is not a journalist that you like following because you think that's someone who gets it or that's someone who's understanding it or, makes it make sense to you?
[27:12] Sharon Goldman: Yeah, I follow a lot of different policy people and researchers on X and Substack. People like Nathan Lambert is a well-known open source AI researcher that I follow a lot and can keep abreast of things in those areas. For my podcast on Ground Level AI, I recently interviewed a former researcher from Meta, Joshua Sachs, who's a real cybersecurity and AI expert. So these are people who have, they have the bona fides for their little part of the AI landscape that I really rely on.
[27:50] Curtis Sparrer: I like hearing that because, part of when I'm trying to get smart on things I follow a lot of journalists that I believe in, but I also like to check in with the, quote-unquote, "civilians" just to make sure that I'm not going too far afield. I saw an interesting story in Poynter, and those who are not familiar, Poynter Institute is probably one of the touchstones journalists go to when they're thinking about their career. And the story was, should journalists give up on traditional job paths? Maybe. And the, subline was, the old newsroom career ladder may no longer be the best path to a career in journalism, but the profession's core values still matter. And I wanted to ask you, about the future of journalism because it is how a lot of us should get our information. But I think that with journalism having a bit of a AI moment, if you will, and publishers also having to grapple with new things from Google, there are thoughts about just how bright is the light in journalism and how can we know things about the world if we don't have journalists to help us?
[29:06] Sharon Goldman: Yeah. I get asked this sort of question quite a bit, and I just, I am very optimistic. I don't think journalism as a discipline is going anywhere because it's only more needed. What form that takes may change. When I started out, it was still back in the days of print journalism that drastically changed for me early in my career and as publishers moved to digital and newspapers and print magazines went the way of the dodo for the most part. I think there will always be a place for great reporting and great analysis and great news instincts. I just can't imagine that going away. I think it would be greatly missed, and it's rather underestimated perhaps. But the form in which that takes, the business models will keep changing like they always have. So it doesn't mean that mainstream media is going away. It doesn't mean that substackers like me are the future, because who knows? But I think there's gonna be room for perhaps a lot more variety, which there already is. Social media influencers, for example, may not be journalists in the traditional way, but some of them bring some of those instincts and some of that content. So I just think the models may change, but the, the core disciplines done well will always be around.
[30:31] Curtis Sparrer: Was it Ben Franklin, for example, one of the first substackers?
[30:35] Sharon Goldman: Exactly.
[30:35] Curtis Sparrer: So let's dig into what you're doing. What stories, if someone was to go and start subscribing, what stories would you recommend people read first or, so that they can get a real sense of what's going on?
[30:49] Sharon Goldman: For me, it's only been about six or seven weeks, so I'm pretty new on my substack. But I feel like I've covered quite the gamut already as far as talking about some of digging into some of the recent news around the OpenAI Hugging Face hack, for example, around Anthropic's Mythos 5. Enterprise companies are very interested right now in the whole open source, closed source debate, and I've done several things on that. I've also reported from several conferences, including an open source AI event in San Francisco. I'll be going to Black Hat next week and reporting from there, so there's a lot of original on-the-ground reporting. And I'm speaking to a lot of enterprise executives from CISOs and CIOs to engineering leaders. I'm looking to speak to more, so I'm always looking for those folks.
[31:44] Curtis Sparrer: Let's talk about that before we move along. Let's dig in. Let's say someone's listening to this and they think, "Sharon should hear my story. Sharon gets me." Or maybe someone wants to anonymously give you a tip How does that work? How do you want to be approached? What is that communication like?
[32:04] Sharon Goldman: I'm just an open book. You can always reach me on all the socials. My contact information is there. You can always email me at sharon@groundlevel-ai.com. On I'm on Signal at Sharongoldman.43. And basically, we can-- it can totally be off the record to start out with. If you're looking to be on the record, that's fine too. I'm just really interested in learning about how AI adoption is playing out in the enterprise across a variety of angles and topics. So if there's anyone willing to chat with me about that who is on the ground in the enterprise I'm really open to speaking with folks.
[32:46] Curtis Sparrer: And I think talking to you would be inappropriate if we didn't talk about how people could subscribe and what that means and why they should do it. I feel very NPR-ish here saying that. So we're not gonna have a telethon, I promise, but just in terms of why people should subscribe to you, how that works, what is that?
[33:06] Sharon Goldman: Well people can subscribe at groundlevel-ai.com. I have a free tier where you'll get, occasional public posts and all the free previews. But if you do pay for a monthly or annual subscription, you'll also get added to my Substack chat and be able to comment on the articles. And I think it's just, it is-- I do really feel strongly that it's important to support this kind of independent journalism right now that, that isn't... It's not that it can't be done, but it isn't really being done in today's environment in most traditional publications just 'cause they can't. It's-- there's so much to cover in AI, and I'm really taking a strong view on reporting about, AI in the enterprise AI as it meets the real world, our AI systems and how it's all connected. This is an ongoing story that's going to unfold for years. If that's up your alley and something you wanna keep abreast of, I think subscribing to Ground Level AI is a good call.
[34:12] Curtis Sparrer: I appreciate the plug. The, the plug is always important.
[34:15] Sharon Goldman: I appreciate it too. Thank you, Curtis.
[34:17] Curtis Sparrer: I'm a former journalist, president of the Press Club, so yes, all, all these causes are important to me. And, nearly every interview ends with this crazy question, and no one knows what to do with it, so I'm teeing it up melodramatically for you. So typically a journalist will say is there anything else we didn't cover?" And that's when either you're sitting on something amazing or you're like, "Oh gosh, what was it?" But I'm vogueing a little just so in case you need to think of that idea of what we didn't cover. But Sharon, if there was one thing that we should have talked more about or that you really wanna impart to people, what would that be?
[34:57] Sharon Goldman: I guess what we didn't talk about is all the fears and anxieties that are coming along with the AI boom. That's what I'm finding so fascinating about the past four years reporting on AI, is that because it's a general purpose technology, it's not just one thing. So previously, when I would write about enterprise tech, it was straightforward. It was like, these are the technologies that are being implemented in the enterprise. We've got your cloud, we've got your mobile, we've got your legacy systems, the ERP the, the BI, and the, collaboration platforms and things like that. What's different here is that not only are these tools that can potentially boost productivity and boost efficiency and as they're implemented in enterprise companies, but they're also doing all sorts of things that people consider quite concerning in society, whether that's military use cases, building data centers ruining our creativity frying our children's brains. Who knows? But people have a lot of different fears around the rise of AI, and that can really push against the, the more optimistic views of how this can be supportive of enterprise companies. And I do think that's something that executives and employees are having to grapple with, and how to deal with it more at a societal level. I don't really know how that's going to play out because, there's so much melding of your use of AI at the workplace as well as at home and in your daily life and in society. But that's one of the interesting things that I think is going to continue to play out, whether it's politically or, in education and just how people view this technology generally.
[37:01] Curtis Sparrer: I think that certain enterprises are not making it easier when they will blame AI for all sorts of things, including layoffs. And so that adds to the kind of unease people have about it, even if sometimes-
[37:19] Sharon Goldman: Yes ...
[37:20] Curtis Sparrer: it seems like AI is a catch-all for all, all evils.
[37:23] Sharon Goldman: That's right. That's right. Absolutely.
[37:27] Curtis Sparrer: Gosh, now that we have talked about nearly everything we possibly wanted to talk about, I guess we should call that a pod. Right?
[37:34] Sharon Goldman: Yeah, that's a great pod.
[37:35] Curtis Sparrer: Okay. We think so at least, and if you're ever listening, let us know what you think. But again, Sharon, thanks so much. I'm Curtis Sparrer in for Christina of AI Realized, and thanks for listening, and more to come.
[37:48] Sharon Goldman: Thanks so much, Curtis.
[38:04] Christina Ellwood: Thank you for listening to this episode of AI Realized Podcast. We hope you enjoyed hearing about what's new now and next for enterprise AI. Let us know what topics would help you on your journey to use AI to redesign your organization from the inside out. Remember to share this episode with a friend, subscribe and leave a review, and listen to more episodes of AI Realized