← The AI Daily Brief: Artificial Intelligence News and Analysis
The AI Daily Brief: Artificial Intelligence News and Analysis
Why OpenAI and Anthropic Are Becoming Consultants
00:26:23
Loading summary
Transcript1 lines
- [00:01]
A
Today on the AI Daily Brief as Anthropic and OpenAI both launch consulting ventures, we talk about why there is no AI transformation without Org transformation. Before that in the headlines, is the White House about to make a major reversal in its AI policy? The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. Alright friends, quick announcements before we dive in. First of all, thank you to today's sponsors, kpmg, Granola, Superintelligent and Rackspace. To get an ad free version of the show, go to patreon.com aidaily brief or you can subscribe on Apple Podcasts. To learn more about sponsoring the show, send us a Note SponsorsIDailyBrief AI A couple other things you can do at AIDailyBrief AI first, we have up our latest AI usage pulse survey. This is a monthly survey we do. It should take less than three minutes to go through and it's all about providing better information around how AI is actually being used, how that's changing. I really appreciate everyone who takes the time to do that every month. And again, you can find that link from the homepage. You can also find a link to the AIDB training site. AIDB training has two very distinct parts. There are our free programs which I am super passionate about. That includes AIDB New Year's Claw Camp, More recently Agent Os. More than 20,000 of you have done those free self directed programs this year and pretty soon I'll be launching one for a very different type of audience for a much more beginner audience. That AI Catch up program will be a bit of an update on the New Year's resolution, so look out for that soon. You can also find out about our Enterprise programs Enterprise clause registering for Cohort 3 and for those who aren't quite ready for an agent building program, we put together a new four week executive catch up program that is designed to radically accelerate executives who are feeling behind on AI overall. You can find that again linked from the AI Daily Brief website or@aidbtraining.com now with that out of the way, let's talk about this shift from the White House. We have a bit of a weird one today. I'd worked up a full normal plan for the headlines with some extended emphasis on this White House AI vetting story. But between when I was preparing and when I started recording, we got more information that has basically forced me to do a not all that common effective double main show. We don't have all the information yet, so we will come back and explore this as is necessary with additional information later in the week. But let's talk about the White House and their changing relationship with AI. Yesterday, the New York Times reported that the administration was weighing up a vetting process that would put the government in charge of assessing and approving the release of powerful AI models. Sources said that an executive order could be issued to set up a working group to consider potential oversight procedures. The working group would include tech executives and administration officials, and among the potential plans, wrote the New York Times, was a formal government review process for new AI models. Another consideration was a plan to give government first access to new models, but not necessarily block public release. The administration apparently met with executives from anthropic, Google and OpenAI last week to discuss the plans, and sources noted that the UK's review process, which involves multiple government agencies assessing new models against safety standards, was viewed as a potential framework to replicate. Now it's pretty impossible to see this as anything but a stark reversal of administration policy. One of the core objectives of this particular administration was to unwind restrictive regulations. Specifically, the Trump administration removed power from the US AI Safety Institute, or usaisi, which is housed within the National Institute of Standards and Technology, based on Biden era policies. The USAISI performs safety testing on any model with more than 100 billion parameters, which for reference Frontier models are currently well over a trillion parameters. And on day one, the Trump Admin revoked the executive order that made government safety testing mandatory, shifting to a voluntary system. During the Biden years, regulation was headed towards giving the government a more formal role and the power to hold back public releases for models deemed to have safety or security issues. These proposals were primarily discussed in relation to the California state government with the SB 1047 bill in late 2024. However, there is a sense that these regulations would filter up to the federal level during a possible second Biden term. The Trump administration, meanwhile, explicitly rejected this concept, asserting that regulatory delays could risk the US falling behind in the AI race with China. In his first public speech as vice president, J.D. vance told World leaders at the AI Action Summit in Paris that, quote, excessive regulation of the AI sector could kill a transformative industry just as it's taking off. At a July event last year, the president referred to the AI industry as a baby, commenting, we have to grow that baby and let that baby thrive. We can't stop it. We can't stop it with politics. We can't stop it with foolish rules and even stupid rules now. The New York Times attributed the change of thinking to the rollout of Mythos. Anthropic, of course, has famously declined to release the model to the public, saying that it's so powerful at identifying security vulnerabilities in software that it could lead to a cybersecurity reckoning. The New York Times administration sources commented that the White House is concerned about political repercussions if a devastating AI enabled cyberattack were to occur. Another contributing factor is the departure of AI czar David Sachs, who left the administration in March after exhausting his 130 day limit on working as a special government employee. In his wake, Treasury Secretary Scott Besson and White House Chief of Staff Susie Wiles have reportedly stepped in to take over the AI portfolio which, whatever you think about them as individuals, neither has any sort of tech industry experience, presumably making it pretty difficult for them to assess the current situation. Regarding Mythos, the idea of the government vetting AI models surfaced some very strong views. Zach Lilly, the Director of government affairs at NetChoice, posted, this would be a terrible blow to American AI competitiveness. No need to return to Biden era regulatory impulses. Innovation at the speed of government isn't innovation at all. Very bad idea. Administrations of the future will force labs to imbue the biases of their side into models in order to get sign off. This will also reduce the number of labs who can ship models having to deal with compliance. I cannot advise against this strongly enough. Relatively few are super keen on this idea. Some of the AI safety folks are directionally aligned with this type of policy but pretty skeptical of its source. The most vocal support came from the China Hawk contingent, with Chris McGuire from the Council on Foreign Relations tweeting, this is a sorely needed regulatory pivot with substantial geopolitical implications. If the US Government vets models pre release and presumably requires companies to include certain safeguards, it also needs a global plan to preserve their security. Post release, he went on to describe a system where the US Government exerts control over global cloud services and distribution of frontier model weights, directly advocating for a return of an even stronger version of the Biden era diffusion rule. Now, interestingly unrelatedly but on the same day, the New York Times published a guest essay by former Trump AI policy advisor Dean Ball and his Biden administration counterpart Ben Buchanan. They wrote, we come from different parties and have guided AI policy under very different presidents, but we agree AI has become so powerful that along with its tremendous promise, the technology poses immediate risks to national security. The article is basically framed as a call to action, calling on Washington to urgently establish a bipartisan AI regulatory regime. Now the article contains a complete policy prescription. Remember, it wasn't a direct response to this news, but it did discuss the testing and release of models. At a minimum, they wrote, Congress should mandate audits of AI developers, safety claims and processes, requiring that they be conducted by independent expert bodies overseen by the government. In other words, they're effectively arguing that model capability isn't necessarily the best place to target regulatory oversight. That models aren't inherently safe or unsafe the way they are used is the primary issue. Building on the arguments from the essay, Secure AI Project co founder Thomas Woodside wrote, I don't think approving model deployments is the best way to think about this. Oversight should be more continuous because models are updated constantly and critically. Models can pose risk during internal use and during development policy. Reacher Miles Brundage elaborated saying on Cyber which kicked this off model capability assessment is part of the story, but other things also matter, like misuse detection, access programs for low refusal models, etc. And these will necessarily evolve over time. Point being not that the government should micromanage a bunch of other things in addition to pre deployment assessment, but just that a sharp pre and post deployment distinction is the wrong mental model. Ethan Molik noted a tendency for policymakers to assume the concepts around AI risk are far more defined than they actually are. He commented, a challenge with AI regulation and vetting is how bad our benchmarks of AI model performance and risks are. There is no benchmark for risks and red teaming requires experiments from dedicated specialist organizations and and is not easy to put metrics around no clear objective numbers. That doesn't mean that there should not be regulation and vetting, but it does suggest that it is hard to write a criteria right now that is not somewhat vague. More R and D into non lab benchmarks is urgently needed. We have remarkably few good ones that are unsaturated and clear. Now to give one example of this, recent testing from the UK AI Safety Institute showed that GPT 5.5 has roughly the same automated hacking capabilities as Mythos, which of course reignited questions of whether Anthropic was actually concerned about cybersecurity or just didn't have the compute to deliver the model, or some combination of both. Relative to the safety conversation, though, the point is that it's unclear where the line should be drawn or even how it should be defined. Now some pointed out that lots of folks were jumping to conclusions and that articles implying that this was a licensing system for model releases were getting it wrong. And the major point of confusion seems to be whether the proposed review regime would actually empower the government to block releases, or just whether it would be a review. Which brings us to this morning when the news broke that the US Government had indeed made agreements with Google, Microsoft and XAI to share early access to their AI models with the government. The government body tasked with testing is the Commerce Department center for AI Standards and Innovation that is already doing this work and which in fact OpenAI and Anthropic have similar agreements with the Commerce Department going back to 2024. The agreement, writes the Wall Street Journal, calls for AI developers to share models with CA ISI with reduced or removed safeguards to evaluate national security related capabilities and risks. Now, these deals are apparently not the same as this cybersecurity focused executive order that we were discussing earlier, so there could be even more news by the time this podcast comes out. My sense is that there is going to be a lot more discussion of this before the week closes. So for now that's where we will end. I will keep a close eye on this story as it develops, but for now, that's going to do it for the headlines. Next up, the main episode One of the most important AI questions right now isn't who's using AI? It's who's using it? Well, KPMG and the University of Texas at Austin just analyzed 1.4 million real workplace AI interactions and found something surprising the highest impact Users aren't better prompt engineers. They treat AI like a reasoning partner. They frame problems, guide thinking, iterate, and push for better answers. And the good news? These behaviors are teachable at scale. If you're trying to move from AI access to real capability, KPMG's research on sophisticated AI collaboration is worth your time. Learn more at kpmg.com us sophisticated that's kpmg.com us sophisticated. Today's episode is brought to you by Granola. Granola is the AI notepad for people in back to back meetings. You've probably heard people raving about Granola. It's just one of those products that people love to talk about. I myself have been using Granola for well over a year now and honestly, it's one of the tools that changed the way I work. Granola takes meeting notes for you without any intrusive bots joining your calls. During or after the call, you can chat with your notes, ask Granola to pull out action items, help you negotiate, write a follow up email, or even coach you using recipes which are pre made prompts. Once you try it on a first meeting it's hard to go without Head to Granola AI aidaily and use code AIDaily. New users get 100% off for the first three months. Again, that's Granola AI aidaily. OpenAI and Anthropic are both launching enterprise AI consulting efforts because everyone is realizing that the challenges and the capabilities of AI. The challenge is getting individuals in the organization actually ready to use it. The truth though, is that all the forward deployed engineers in the world aren't going to help you if you don't actually have a coherent strategy based on an understanding of your actual AI readiness. Super Intelligent Maturity Maps give you a chance to see where you stand relative to the industry on deployment depth, systems integration, data access, outcomes, people and governance. And from there, our customized AI planning assessments can help you figure out what you need to do to improve your readiness and how to sequence it. Go take your own Maturity Maps quiz@BESuper AI and send us a note if you want to go deeper. If you're trying to move AI from pilots into production inside your organization, the hard part isn't getting something to work once it's running it at scale and achieving real business outcomes. It's the day to day operating model with the governance, security controls and clear accountability that production actually demands. Rackspace Technology just announced a partnership with Palantir to help you run Palantir Foundry and AIP in production and through a governed managed operating model. Here's what that looks like in practice. Implementation support Help getting your data ready and migrated infrastructure hosting and ongoing managed operations so your team isn't left stitching together infrastructure data pipelines and AI workflows on your own. The focus is getting high impact use cases, live in weeks and then scaling from there across your workflows with consistent operating controls from edge to core to cloud. If you want to learn more about running Foundry and AIP in production with Rackspace technology, go to rackspace.com palantir welcome back to the AI Daily Brief. Today we are discussing a pair of news stories and their implications. Yesterday News broke about two different efforts, one from OpenAI, one from Anthropic, to build massive service related businesses around enterprise AI deployments. The first report was not an official launch from OpenAI but a report in Bloomberg. The report was that OpenAI had raised more than $4 billion from PE giants like TPG, Brookfield Asset Management, Advent and Bain Capital to build a new venture that they're apparently calling the Deployment Company. The firm, which will have an initial valuation of $10 billion, will be majority owned and controlled by OpenAI. Now, while we haven't gotten the formal announcement yet, there has been a ton of buzz flying around about this, which makes sense. You can't really have this many people coming together for a big new effort without some information about it leaking out. I will add only that from what I've heard, this is not some sideshow effort and it is certainly not a side quest. In fact, it seems to me that this cuts to the core of OpenAI's new focus and is kind of a recognition that the last mile is a lot longer in practice than just a mile Anthropic officially unveiled its similarly focused Last Mile joint venture. Anthropic's main partners in the venture are Blackstone, Goldman Sachs and Helman and Friedman. Blackstone and Hellman are investing approximately 300 million each, with Goldman Sachs putting in another 150 million. There are a whole slew of other partners involved as well as with the total investment being around 1.5 billion. The announcement from Anthropic was fairly minimal and effectively it is a forward deployed engineering organization. In the announcement they write, applied AI engineers from Anthropic will work alongside the firm's engineering team to identify where CLAUDE can have the most impact, build custom solutions and support customers over the long term. Now the idea of forward deployed engineers is nothing new. It was popularized by Palantir and over the last few years has become more and more de rigueur as a part of AI deployment consulting. And speaking of AI deployment consulting, Anthropic makes it clear that this new venture is not replacing their previously announced CLAUDE Partner Network, which represented Anthropic partnerships with a bunch of the big global systems integrators, but an extended effort to get even deeper and closer to the actual point of enterprise transformation. One interesting wrinkle from Anthropic is that they do specifically call out that this new organization will work with mid sized companies, which might be a because that's where they've identified opportunity, or b frankly, a concession to those partners who are focused on the bigger organizations. Indeed, down in the section on the CLAUDE Partner network they our partnerships with Accenture, Deloitte, PwC and the other consulting and systems integration firms in the CLAUDE Partner network are one of the ways Claude benefits the world's largest enterprises today. The announcement also has a small section about what the work will look like. A typical engagement, they write, will start with a small team working closely with the customer to understand where Claude can have the biggest impact. From there, the company's engineers alongside Anthropic applied AI staff will will develop cloud powered systems tailored to each organization's operations. One thing to note is that although they are discussing a engineer to engineer partnership locus, it's not just enterprise engineering process that they're trying to redesign. For example, they discuss a multi site healthcare services group like a network of physician practices. Clinicians, they say, spend hours each day on documentation, medical coding, prior authorizations and compliance reviews. An engagement might begin with the company's engineering team sitting down with clinicians and IT staff to build tools that fit into the workflows that staff already use. The clinicians know where time disappears in a shift and what good patient care actually requires. The company's engineers build around that knowledge, allowing clinicians to devote more time to patient care. Now what's interesting here is that in addition to a thesis about a model of the right type of consulting engagement that is a partnership between applied AI engineers from Anthropic and the engineers inside the company, there's also almost embedded a thesis about how AI change is going to diffuse through new emergent partnerships between the engineering organization and other parts of the company. It feels to me a bit like anthropic embedding best practice in an operating model. Now, to be honest, both of these announcements feel a little bit rushed. OpenAI's of course isn't even an announcement, it's just a report. And Anthropic's announcement, such as it is, has very few details, including not actually naming this new venture. Now one other bit of news that I think is actually part of the same story. Sierra, which is a customer and client relationship agent company, announced that it's raising just under $1 billion at over a 15 billion doll. The reason that I say it's related is that Sierra's operating model has always from the beginning bid to have support heavy deployment engagements that almost situated halfway between a traditional tech startup and a consulting firm. It's not unreasonable to view these Anthropic and OpenAI efforts as effectively scaling that model into a more general purpose deployment firm. Interestingly, the response from people is pretty much well, yeah, this was expected. Vercell's Drew Bredvik writes, this is neither bearish nor bullish, just proves that it's going to be a 10 to 20 year slog to get all businesses agentified. Drew, by the way, had written a post back in October called AGI Some Assembly Required that made the argument crisply that many have made that model innovations and technology innovations are likely to outpace the ability of companies to absorb that capability. Set Box's Aaron Levy writes, whether it's existing consulting firms, new ones that emerge, FTEs from agent vendors, or new internal agent engineering roles, the amount of work that's going to be created to implement agents and enterprises will exceed anything we imagine today. Aaron then goes on to talk about just how many different pieces are involved in the easy to say, hard to do implementation of agents. Putting a fine point on that. Abhishek V from Twitter points out that there are more FTE job openings right now than the total number of FTEs currently employed in the US now, to be clear, I'm not actually sure that There are under 3,000 FTEs in the US right now, but the number of live job openings on LinkedIn, which is what he's pointing to, certainly shows no matter what just how much demand for this type of role there is. Putting the raison d' etre for these efforts in the simplest way I've seen it. Investor Jeff Woo writes, agents compress labor faster than institutions compress bureaucracy. The reality I think we're waking up to is that when push comes to shove, there is no AI transformation without organization transformation. This is, I think, increasingly clear to people who are trying to deploy AI inside big enterprises, but it's also starting to show up in the data. Microsoft just released their annual Work Trend Index. This is an annual report they do that is always a really good source of understanding where a broad subset of organizations are. When it comes to AI adoption, a lot of the data is concentrated around organizations using Microsoft, but because Microsoft has such big coverage that actually gives a fairly wide ranging view as opposed to, for example, when I survey folks, you're going to have a very concentrated leading edge of the market sort of perspective. Given that you're all listeners to a daily AI podcast, Microsoft divided their findings this year into three sections. The first is how AI lifts the ceiling on individual potential. The second is the idea that the job of every leader is to re architect work. And the third is to make an argument that every firm is, or at least needs to be, a learning system. In section 2, they put exactly the gap that anthropic and OpenAI are trying to address with these new efforts. In plain terms, most organizations, they write, are not yet built to capture the value of expanded human agency. The challenge is not isolated to tools or individuals. It's a breakdown across the system that connects leadership, culture, management practices, and how work is measured. Put simply, they say, in many cases employees are moving faster than the organizations around them. And one of the most valuable pieces of this year's index is a quadrant they've developed called the transformation paradox, that shows the relationship between employees and their organizations when it comes to AI readiness. The Y axis runs on the bottom from low individual capability to high individual capability on the top, while the X axis runs from low organizational readiness to high organizational readiness. Now, of course, where organizations want to be presumably is high individual capability and high organizational readiness. Microsoft calls that the frontier. And unfortunately only 19% of organizations fall into that category. Organizations that have both low individual capability and low organizational readiness, Microsoft calls stalled, and they represent 16%. In organizations where AI conditions are high but individual AI practices low, they call that unclaimed capacity. And that represents the smallest section, just 5% of organizations. Meanwhile, organizations where there is high individual capability but low organizational readiness, they refer to as blocked agency, which represents 10%. Now, you might notice that those don't add up to 100%. It's because they actually categorize 50% of the organizations as still emergent, where both individual AI practice and organizational conditions are taking shape in a way that they're not exactly quantifiable into these categories. So far, I would argue, based on our experience both around aidb, but at Superintelligent as well, that if you put those emergent organizations into one of the four quadrants that they most belonged in, the number that were in the blocked agency quadrant, which was again individual AI practice, high organizational AI conditions low would rise dramatically. That is, I think, increasingly the average state where, yes, there is a very wide range of adoption, but the people who are adopting AI well are significantly constrained by organizational readiness issues. But one of the things that's clear from the work trend Index is that when an organization does get its stuff together, it leads to more of what Microsoft calls frontier professionals, that is the best, most effective and most regular users of AI. There are big org level differences between frontier professionals and non frontier professionals. While 64% of all employees said that their manager openly uses AI, 85% of frontier professionals said the same. This was basically the same across every other question around managers, including setting quality standards for AI work, creating space for experimentation, and encouraging more ambitious work. Redo design. In each case, the difference between the frontier professionals and the non frontier was at least 20 percentage points. Frontier professionals also were more than 2x more likely to say that they were rewarded for the reinvention of work with AI, regardless of the outcome. Section three, Organizations as Learning Systems puts a fine point on this, arguing that the biggest factor behind AI impact is not Individual but organizational. The headline stat is that organizational factors in which they include culture, manager, support and talent practices account for more than 2x of AI's real impact and as does individual mindset and behavior. Now this is of course only going to get more challenging in the time of agents. This is what a lot of our content has been about. Recently, Microsoft identified that the number of active agents in their EcoSystem has grown 15x year over year overall and 18x in large enterprises. In an episode soon we'll get into some of the common mistakes we're seeing with deployment, but frequent guests and AI DB training leader Nufar Gaspar shared three observations of AI adoption shortcuts that are pretty common but end up not working out. The first one she calls Buy and Hope, which I think will be familiar to many of you. That's where an organization pays for the licenses, sends the excited email out to people saying they now have these new toys, and then hopes that somehow turns into transformation. The second adoption shortcut that doesn't work she calls contain and Delegate. This is where leaders hand the transformation goals over to the AI team rather than a interacting with it themselves and b actually trying to diffuse it across the organization as a whole. The last adoption shortcut that doesn't stick that she identifies, which is one that Professor Ethan Malik beats the drum about quite a bit, is the outsourcing of knowledge. This was particularly common, I think in the first couple years of post ChatGPT AI consulting where organizations would hire their GSI or big consulting firm of choice and effectively hope that they could figure it out for them. Now of course, given that Anthropic and OpenAI are both raising billions of dollars to pursue the deployment opportunity, there is clearly a role for those types of external expertise partnerships. But as you heard, the model is not drop in an expert to figure it out for you, it's embed builders alongside your teams to run a best practice laden process that integrates engineers with everyone else. We frequently talk here about the capability overhang. That's the gap between the possibilities of AI and the value that people are actually getting out of it. The capability overhang operates on an individual and an organizational level. However, in the age of agents, especially for organizations, it is getting worse not better. Model and harness capabilities together are racing ahead of where most organizations and it is going to take an all hands on deck kind of approach to actually help organizations get the value that they could out of this. Now, as an aside, and clearly beyond the scope of this particular show, I will note how infrequently slash never the AI jobs displacement discussion actually engages meaningfully with the reality of last mile deployment issues. I think this is an important discourse to correct, given that for those who feel that AI will ultimately unlock and create new opportunities, but also cause a fairly chaotic transitional period, one could argue that organizational inertia actually becomes an asset in slowing down the rate of diffusion in a way that gives us more time to adapt. Still, holding those broader society implications aside, when it comes to the actual questions of how organizations can get value out of AI, I think the most interesting thing to watch will be to what extent the FTE collaboration style model of consulting that both of these organizations are clearly pursuing will becomes the norm for this type of professional service going forward. Now, a lot of issues remain. For one, there is for example, an inherent business model challenge where these organizations presumably face a tension between their goal of helping companies be not only most effective in their AI adoption, but also most efficient and on the other hand, their parent organization's goal of selling as many tokens as possible. The point is that there's going to be a lot of interesting things to watch with these developments, but for now that is going to do it for the AI Daily Brief. I appreciate you listening or watching as always and until next time, peace.