The AI Daily Brief: Artificial Intelligence News and Analysis - Fable is Back: Here's What You Should Try First

Episode Date: July 1, 2026

Fable 5 is officially returning after export controls were lifted, but the rollout comes with new guardrails, lingering policy questions, and a short window of subsidized access. NLW breaks down what changed, what to watch for, and why Fable’s biggest value may be in strategy, hard technical problems, and writing with clear standards. In the headlines: OpenAI’s inference cost push, Base44’s new model, AWS’s forward-deployed AI unit, Claude Tag for Teams, and SpaceX’s Memphis data center backlash.Brought to you by:KPMG – Research from KPMG and the University of Texas at Austin shows the highest-impact AI users treat AI like a reasoning partner — and those skills can be taught at scale. Learn more at ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠kpmg.com/us/Sophisticated⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Hyperagent - Hire a fleet of always-on agents. New users get $1,000 in inference. hyperagent.com/aidailybriefSection - Section turns AI investment into workforce transformation and ROI - ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://www.sectionai.com/⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Scrunch - The AI customer experience platform - ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://scrunch.com/⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Blitzy - Want to accelerate enterprise software development velocity by 5x? ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://blitzy.com/⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠AssemblyAI - The best way to build Voice AI apps - ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://www.assemblyai.com/brief⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Robots & Pencils - Cloud-native AI solutions that power results ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://robotsandpencils.com/⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠The AI Daily Brief helps you understand the most important news and discussions in AI. Subscribe to the podcast version of The AI Daily Brief wherever you listen: ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://pod.link/1680633614⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Our Newsletter is BACK: ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://aidailybrief.beehiiv.com/⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Interested in sponsoring the show? [e-mail]

Today on the AI Daily Brief, Fable 5 is officially coming back. Before that in the headlines, the quest to cut inference costs. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. All right, friends, quick announcements before we dive in. First of all, thank you to today's sponsors, KPMG, robots and pencils, Blitzy and Airtable. To get an ad-free version of the show, go to patreon.com slash AI Daily Brief, or you can subscribe on Apple Podcasts. And if you want to learn more about sponsoring the show, send us a note at sponsors at

AID Daily Brief.AI. We kick off today with a story that is very of the zeitgeist that we are living in right now. OpenAI has found a way to slash their inference costs in half, sort of. This headline from the information grabbed a lot of attention, and understandably so. Everyone right now is looking for new approaches to token efficiency, and the implications of these searches have huge impacts on the business models and the companies that are shaping AI and the larger market structures they're operating in. Now, when it comes to this article specifically, the details do suggest that it might be a smaller breakthrough than it appears at first. The claim

is that OpenAI researchers have discovered a new optimization technique that cut their inference requirements in half for existing models. When the technique was applied to chat GPT users who weren't signed into the service, OpenAI was able to serve that entire user base segment on just 100 GPUs. The OpenAI source didn't disclose what the technique was. The information speculated, it could be quantization, cash optimization, batching queries, or rooting queries to a lower power model. Notably, none of those techniques would improve service for OpenAI larger models without compromises. The universal truth that there is no free lunch remains, and most attempts at optimizing inference come at the expense of model quality.

Now, there's also the question of what it means that OpenAI is testing this technique on a tiny batch of their least engaged users. That might be a totally reasonable starting point, just the first test of many, or it could be a cautionary approach that implies that there's some risk of quality degradation. The TLDR is that while there seems to be something interesting here, we probably shouldn't treat it like some sort of silver bullet to resolve the compute crunch. Still, the information Stephanie Palazolo is convinced that OpenAI is on to something. In an accompanying video, she said,

this is a very important secret sauce for them that they don't even want to tell other OpenAI employees about, because if these things leak, it can quickly be picked up by other labs, which can also then use that to lower their costs. This is something they're holding very close to their chests. Now, many pointed to a new research paper from Deepseek, which open sources a speculative decoder system called D Spark that can speed up inference by 85% during testing on small models. Now, it's unclear how D Spark impacts costs, but it is a reminder that inference optimization is not even close to a solved problem, meaning that theoretically huge gains from some

novel technique are definitely plausible. And of course, even if open AI hasn't found a way to boost efficiency by 50% across the any sort of gains here could still be a very big deal. OpenAI's army of free users are a significant drag on profitability, so anything they can do to cut inference costs to that user base could really move the needle. And given the particular audience, certain types of quality reductions may be more tolerable. Everett Randall of Benchmark Ventures has been talking about a phenomenon he's calling the AI mom test. He recently said, there's nothing my mom actually asks of her AI products that needs to be done by the frontier or even a near frontier model. And it seems to be at least

initially, like that could be the group that this new technique addresses. Certainly there is a lot of chatter out there about innovations and new approaches in this area. AI aggregator Andrew Curran tweeted, I'm posting this prediction now so I can quote it later. There has been a significant breakthrough in architecture, specifically around memory efficiency, not by one of the big labs, but by a team that was spun out of OpenAI. They will probably announce it soon. Now, in addition to labs finding new approaches. Companies themselves are also finding new more efficient architectures. 20-minute VCs Harry Stebbings tweeted, in the last 24 hours, I have had five founders message me of varying-sized companies, some 10-person startups and one $200 billion public company. All of them

stated they have been able to cut inference spend by 75% or more with little effort, no performance change, and better latency. The times they are a changing. Now, speaking of innovations in this new token efficiency era, vibe coding platform base 44 has launched their own AI model in an attempt to shore of the business. The model is called Base 1 and follows the same playbook as Cursor's Composer. Namely, Base 44 has taken an open-source bass model and applied their own fine-tuning using training data from hundreds of millions of user interactions on their platform. CEO Mayor Shlomo laid out the strategic thinking in a few different ways. Firstly, Base 44 is making the bet that narrowly trained models can be competitive with

the frontier. This bet appears to be paying off for Cursor, with most viewing Composer 2.5 is good enough for common tasks. Right, Shlomo? General models need to be at everything. They need to understand many programming languages, many workflows, many domains, and many kinds of reasoning. However, Base 44 only needs their model to be good at building web apps. Now, of course, Base 44 also views the model as a cost control measure. Shlomo wrote, it gives us more control over cost, latency, reliability, and quality, while still letting us use the best external models where they are the right fit. Finally, he writes, the model gives Base 44 a way to utilize platform data to improve the product. The idea is similar

to the Harness model pairing that OpenAI and Anthropic have pursued with their own coding platforms. Base 44 believes they can develop the model and harness in tandem to deliver strong platform-specific results. As AI becomes a bigger part of how software is created, right to Shlomo, owning more of that intelligence becomes just as important as owning the infrastructure around it. Now, speaking of strategic convergence, AWS is launching a new division to join the AI deployment race. AWS announced that they will invest a billion dollars to create a new unit staffed with forward-deployed engineers to help customers set up and use AI tools. Now, this follows, of course, OpenAI and Anthropic both launching private equity partnerships to house their FTE divisions,

with Google in May expanding their existing FTE division, and Microsoft announcing an FTE partnership with EY in June. Francesco Vasquez, VP of AWS's frontier AI engineering and services, said that the company has been upskilling salespeople to be FTEs, giving them the title of solution architects. Vasquez said the effort is an expansion of AWS's generative AI center, which was first announced in 2023. She added that AWS will focus on industries with the strongest demand, including health care, government, and financial services. Reinforsing the themes of the moment, Vasquez also noted that the big shift towards AI budget optimization rather than just the deployment of AI capabilities. She said, open source and the use of open weight models

is definitely gaining traction for customers for a variety of different reasons. Price performance, but also they serve as the task. And another indication of just how significant this trend is, the AI Engineer World's Fair, which is happening this week in San Francisco, is actually hosting an AIFTE mini conference within the event. Now, following up on the Claude Tag story from last week, Anthropic seems to be preparing to bring agents to Microsoft Teams as well. The information reports that Anthropic recently told Microsoft that they plan to bring

a version of Claude Tag to Microsoft's Slack equivalent Teams. Claude Tag, you might remember, allows users to summon Claude into a channel to receive tasks or take instruction. Unlike previous cloud integrations, Claude Tag isn't tied to any particular user. Instead, it functions as an organization-centric agent with persistent memory and tool access independent of the user. In fact, it's actually not calling Claude exactly. It's calling the full suite that comprises Claude code.

Now, one interesting sub-story behind all of this is what it says about the relationship between Anthropic and these platforms, both Slack and in the future, Microsoft. There has been some scuttlebut that behind the scenes, some Salesforce employees, were concerned about Anthropic being allowed into the ecosystem on such a fundamental level, although I don't think that that was universal given the amount of PR that I got from Salesforce about this integration. But still, introducing Claude Tag to Microsoft does add another level to this power struggle. Currently, both Salesforce and Microsoft allow third-party agents to access their ecosystems free of charge. In fact, Microsoft's CEO, Satchinadella, went one better during

an investor call in April. He claimed that Claude helped reinforce Microsoft's ecosystem commenting, It's fascinating that here we are in 2026, and the most exciting things in AI are plugins in Word or Excel. When you see that, that means we have a structural position in knowledge work. That said, will the tune change as Claude gets more endemic across the knowledge work spectrum? It's certainly something interesting to keep an eye on. Lastly today, one of the things that we've been talking about when it comes to the data center buildout is the opportunity for the data center builders and operators to, if I might be allowed to put it crassly for a moment, buy off the communities that they're operating in, or perhaps a better way to think about it, is to cut them into the

benefits that the data centers might represent. One example of someone starting to nudge down that path comes from SpaceX. The company is discounting Starlink subscriptions in Memphis in an attempt to quell backlash. SpaceX's Colossus data centers are located just south of Memphis and have been the subject of significant local controversy. The campus operates on 46 gas turbines providing off-the-grid power. Local groups have complained about air pollution and noted the turbines are operated without a permit as they're technically portable, which some see as abuse of a legal loophole. The U.S. government, in fact, recently intervened in a lawsuit to shut down the turbines, claiming that Colossus is a matter of national security. On Tuesday, SpaceX announced

that residents in the Greater Memphis area will receive half-price Starlink subscriptions and free hardware for new signups. In addition, the mayor of Memphis recently announced that SpaceX has recommitted to the construction of a wastewater treatment plan. Those plans had been halted in April, with SpaceX claiming they were prioritizing the construction of the Colossus 2 data Center. However, the construction halt came just days before the lawsuit was filed. Annancing the discount, VP of Starlink, Michael Nichols wrote, The unique capabilities of the Colossus Data Center could not be accomplished without the partnership and support from the local Memphis community. Happy to bring affordable and great Starling connectivity

to our neighbors. Look, directionally, I'm glad that we're starting to see this, but I have to say, but for anyone from SpaceX who might be listening, while I applaud this direction, I would be going a lot harder than a half-price discount for people who have to become your customers to even get that discount. Still, some encouraging signs that we're starting to think differently about these relationships. Still, that's going to do it for today's 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 in 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 prompting.

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 slash us slash sophisticated. That's KPMG.com slash US slash sophisticated. I cover the capability gap between AI potential and AI reality every day on the show. Most companies are still figuring out how to start. Robots and Penciles is already launching and scaling.

Agendic and generative AI in production, at large enterprises in weeks. AWS Advanced Tier pattern partner more than doubled in a year. And they're hiring. 50 open roles. If you're someone who knows this moment is different, who wants to be inside it, not watching it, this is worth a look. At Robots and Penciles, the best ideas win, and the team is purposefully kept super high quality. This is the kind of place you look back on as the best decision you ever made.

Take a look at robots and pencils.com slash careers. If you're looking to adopt an agentic SDLC, Blitzy is the key to unlocking unmatched engineering velocity. Blitzie's differentiation starts with infinite code context. Thousands of specialized agents ingest millions of lines of your code in a single pass, mapping every dependency. With a complete contextual understanding of your code base, enterprises leverage Blitzie at the beginning of every sprint to deliver over 80% of the work autonomously.

Enterprise-grade end-to-end tested code that leverages your existing services, components, and standards. This isn't AI autocomplete. This is spec and test-driven development at the speed of compute. Schedule a technical deep dive with our AI experts at blitzie.com. That's BLYtZY.com. This episode of the AI Daily Brief is brought to you by HyperAgent, where you run fleets of agents your team can manage together. New users get $1,000 in inference. Forget local agents and chat workflows waiting on your laptop to be prompted.

Hyperagent deploys always-on agents in the cloud, doing real work across the tools your team already uses. Marketing's agent turns competitor moves into landing pages. Sales as agent enriches leads, drafts emails, and updates the CRM. Ops agent chases the paperwork and tracks the budget. Every agent has access to shared context and follows your rules about scope and approvals. It's time you add agents that feel like teammates. Hire yours at Hyperagent built by the team at Airtable. claim your $1,000 in inference at hyperagent.com slash AI Daily Brief.

Welcome back to the AI Daily Brief. Well, friends, after something like 19 days offline, it should be that by the time you're listening to this, Fable 5 has been turned back on. On Tuesday night, Anthropic announced that the government had lifted export controls and cleared them to begin redeploying Fable 5. At 7.52 p.m., this tweet exploded onto the scene. We've received notice that the Department of Commerce has lifted export controls. controls will begin restoring access tomorrow and we'll share an update soon. We're grateful to our users

for their patients and to everyone who worked with us on redeploying the models. A few hours later, they shared further details on the rollout beginning today, July 1st. Fable 5 will once again be available to all global users across all paid subscriptions. There will be a short extension of the subsidy period, though we were promised when it first came out. Fable 5 will be included for up to 50% of weekly usage limits until next Tuesday, but after that, access to the model will require the purchase of usage credits. Mythos restrictions remain as they were at the end of last week, with approved U.S. firms able to access the model for both domestic and foreign workers. Anthropics said they will continue to work with the government on an expanded rollout under

Project Glasswing, including providing the model to international firms. Several administration officials commented on the resolution. White House Chief of Staff Susie Wiles, who, according to reports, has been one of the AI policy leads in recent months, wrote, Under President Trump's leadership, the United States is the undisputed winner in the AI race. My gratitude to companies across industries who continue to work closely with the White House to implement the president's executive order, promoting advanced AI innovation and security. This includes excellent work around advanced model access and guardrail testing and security. The government and private sector have worked together in a way we have never seen before,

and this foundation of America First is unprecedented. Our shared priority remains get the best tech deployed as quickly and safely as possible. Commerce Secretary Howard Letnik, who has been in charge of applying the export controls, added more specifically, over the past two weeks we have worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the U.S. government and strengthen America's leadership in AI. Now, alongside the announcement, Anthropic provided their own version of the events of the past few weeks. They discussed the jailbreak reported by Amazon and their efforts to convince the administration that nothing was amiss. As a refresher, the core claim was that Fable was able to identify

serious vulnerabilities in a codebase, which the administration believed was a mythos-level capability. Our testing, wrote Anthropic, confirmed that many less capable models, including Claude Opus 4.8, GPt 5.5, and Kimi K-2.7, could identify the same vulnerabilities as Faber 5 did in the report. When it came to the demonstration of how to exploit the single vulnerability, every model we tested could produce the same demonstration as Fabl 5, including Claude Haiku 4.5, Sonnet 4.6, Opus 46, Opus 4-7, Opus 4-8, GPD-5, GPD-5, and Kimi K2.7. Importantly, they continued, the reported technique did not expose any unique mythos-level cyber capabilities. The behavior reflected a borderline case for Fable 5's safeguards. There are some tasks that are unlikely to be dangerous,

but are nonetheless blocked by the safeguards out of an abundance of caution. The reported technique allowed access to one such behavior, but it only involved routine defensive cybersecurity work. Now, this has been Anthropics' position from the beginning, that although this was a genuine delbrate, it didn't unleash dangerous cyber attack capabilities. Nevertheless, Anthropic has amped up the guardrails for Fables' return. They have trained a new classifier designed to target and block the behavior described in the Amazon report with a claim success rate of 99%. As with the previous version, users will be informed if they trip the guardrails and will be reverted to Opus 48 for that request. Anthropics said that they have tested the new

classifier with the Commerce Department's Center for AI Standards and Innovation, which agrees that Anthropic safeguards are, quote, extraordinarily strong. Anthropic noted, the new classifier also comes at the cost of flagging benign requests more often during routine coding and debugging tasks. As with all our safeguards, we'll continue to refine this to better distinguish genuine misuse from legitimate requests and reduce false positives. Now, one of the first responses was effectively that, while folks were glad that Fable was coming back, it wasn't exactly clear what had changed that addressed the government's initial concerns. Policy advisor Dean Ball wrote, great news, but we have no idea what Anthropic did to make the model safe,

what commitments Anthropic has made going forward, and whether or how any of this applies to other frontier models in the government's licensing queue. We know that GPT 5.6 is in that queue, but it's fair to assume that other model developers are at least in early stages of submitting their models. Reinforcing the point that he has made over and over again, Dean continued, this opacity will not lend itself well to a stable, investable, trustworthy industry over time, but, and here Dean ends on a positive note, the U.S. government needn't figure this all out, in a day, and a two-week review timeline is not insane in the grand scheme of things. The status quo is not tenable, but we made progress today. That progress is just a first step,

but it is worthy of applause nonetheless. Prins also pointed out how many new questions. The letter about the export controls being lifted brought up. With Anthropic agreeing to, quote, proactively detect and address security risks associated with Fable 5 and Mythos 5, Prince asks, how is Anthropic required to proactively detect these security risks, not publicly disclosed? When Anthropic agrees to work diligently with the U.S. government on protocols and standards, and releases for Mythos Fable and Future Models, Prins notes that this is extremely broad language

that appears to cover all future models and is not limited to cybersecurity risks. Still, I think Miles Brundage captured the sentiment of many when he wrote, the first rule of Fable Club is you do not ask too many questions about what exactly Anthropic agreed to that they weren't doing before and you enjoy your access. And overall, from a policy perspective that seems to be the tone, even though people have lots of questions,

the fact that we're getting the model back in this version gives people reason to be cautiously optimistic. writes Boxes Aaron Levy, it's been a messy process to get here, but at least there's some semblance of a framework that could be practical. The note of caution here would be that there's a lot of subjectivity that goes into various risks and their actual levels of exploitability in practice. We're likely going to be living with a framework that requires heavy judgment

and back and forth between labs and the government for major releases. The best we can hope for is that this is a relatively efficient process and hopefully as ways of being sped up for incremental version updates and models. It would be a bad outcome if every release after this level of threshold of capability required the same review process, and we don't get the same rate of breakthroughs we've been seeing. Now, one of the reasons that I think people were willing to extend that benefit of the doubt was that as recently as yesterday, people were speculating about Fable 5 coming back only with things like a new KYC regime, where people had to verify their identities and perhaps

verify their identities as American citizens before they could get access. That appears not to be the case. Fable 5 is not just coming back for U.S. users, but for all users globally. And yet, There is a big question lurking. When I tweeted an image of Dario and Uncle Sam riding an eagle, delivering Fable across America flanked by fighter jets, Aaron Schneider asked the key question, but will it be as good as we remember? The specific concern comes around this line from the announcement. In the near term, some routine tasks like coding and debugging will fall back to Opus 4-8.

That led folks like Lassan on X to write, The Fable 5 relaunch is kind of fake. Some routine tasks like coding and debugging will fall back to Opus 4.8? You can use that even more restricted Fable 5 version in your Anthropic subscriptions until July 7th. Lexon X wrote, LMFAO, this cannot be a real statement from Anthropic. Routine tasks like coding, skull emoji. They've lost their minds. Do they think people are paying an absurdly high price just for the privilege?

Now, to read from the Claude Code team came on to clarify that that tweet had particularly loose language. He wrote, as with the original classifiers, a small fraction of routine coding and debugging tasks will be flagged and fall back to Opus. In other words, while the tweet made it seem like coding and debugging were part of the routine tasks that would fall back to Opus in general, the Claude Coe team is saying that no, in fact, it is just a small fraction of routine coding tasks that will be flagged in that way. Now, given all this, I want to come back and talk about the ways that I think you should start to test Fable right away when it comes back. But before that, we actually do

have one more Anthropic model to talk about. Earlier on Tuesday, Anthropic announced Claude Sonnet 5. And if I had to guess, this suggests to me that they were not sure that Fable would be coming back as fast as it was, as I'm not sure that they would have announced this particular model when they did, in the way that they did, if they knew that later that night, they'd get to announce that Fable 5 was coming back. Anthropic pitched the model as their most agentic version of Sonnet yet, writing, it can make plans, use tools like browsers and terminals, and run autonomously at a level that,

just a few months ago, required larger and more expensive models. Now, according to the benchmarks, Anthropic tried to pitch the model as almost as good as Opus 4-8 for a fraction of the cost. It's a few percentage points shy of Opus on the two major coding benchmarks, Sweet Bench Pro and Terminal Bench 2.1, with the same gap existing for computer use and knowledge benchmarks. Maybe the most interesting result was GDP Val, where Sonnet 5 delivered a huge jump over Sonnet 4.6, and even slightly outperformed Opus 4.8. Now, this could be indicative of the strong agetic performance, that Anthropic was touting, while GDP Val aims to measure economically valuable work, in practice, the score largely comes down to successful tool calls.

Sonnet 5's score could indicate that it's much more capable of following through with end-to-end agendic work rather than getting stuck halfway through. While Sonnet 5 retains the same pricing as Sonnet 4.8, Anthropic is trying an introductory price for API use. Until the end of August, Sonnet 5 will cost $2 per million input tokens and $10 per million output tokens, and will revert to the standard $3 and $15 after that. In contrast, the price for Opus is $5 and $25, so Sonnet 5 could be a more cost-effective option for some use cases. But what about external benchmarks and actual user impressions?

On Cursor's cursor bench, they wrote that it was a meaningful step up from Sonnet 4.6. It also saw a jump on the artificial analysis intelligence index from Claude Sonnet 4.6's 47 to a score of 53, which puts its just one point behind Claude Opus 4.7 and a couple points behind GPT-55. but they point out that without the promotional pricing, it will actually cost more per task than Opus 4.8. On max effort, they write, Sonnet 5 used around 40% more output tokens per intelligence task than Sonnet 4.6,

and around three times the number of agendasic turns for their knowledge work evaluations. The overall run, in fact, cost more than Opus 4.8. This is because it generates almost twice as many tokens as Opus 48 per task. And when it came to running the entire bench of tasks, Sonnet 5 was actually more expensive than Fable. That led a lot of people to wonder what the heck this is actually for.

Max Blade wrote, Composer 2.5 is faster, GLM 52 is cheaper, Opus 48 output is 10 times better. It's an improvement, yes, but without a real use case, it is dead on arrival right now. Will you guys be running it? Now, one important caveat with that is that the artificial analysis tests are run at max settings, and that might not be the optimal way to run Sonnet 5. David Shapiro writes, Okay, I can't believe I'm going to say this, but Sonnet 5 Max is too high effort. It's like giving a box of squirrels a bunch of cocaine and saying, go with God, and just seeing

what comes out the other side. Ben Davis, who works with YouTuber Theo, wrote, After doing around a billion of tokens on it today, you're all wrong. Sonnet 5 is good. Is it inefficient? Yes. Slow? Yes.

Expensive, yes. But if you think it's the same as GLM 5.2, a model I really like and will continue to defend a ton, you're a fool. The thing with this model is you have to use it wildly differently. It's not the type of sonnet model we're used to. probably could have been Opus 5 if it was slightly bigger. It's basically an automatic Routh loop. It's spawning tons of subagents, making many stacked PRs, reviewing itself adversarily, auto-testing its changes and not drifting off task at all. If you use this thing the same way you've

been using old models, you're going to have a bad time. Indeed, some are wondering if the real way to look at Sonnet 5 is as the work model that would run the subagents that something like Fable 5 would spin up. Dan McCatia writes, You'll use Fable 5 as the super-intelligent advisor and Sonnet 5 as the fast and efficient implementer. Still, when push comes to shove, the model that people are really excited about having back is Fable 5. And for anyone who used it in those first three days that it was actually available, you'll understand why. So the question becomes, as Fable 5 comes back online, and we have this one week where you can use it in your normal subscriptions, what should you be using it for? Aniket Pangemani writes, you have one week to use Fable 5 without going bankrupt.

Here's how you should make the most of it. 1. Use Fable 5 for planning, not for implementation. Use the Codex plugin in ClaudeCode to delegate implementation to GBT 5-5. 2. Ask Fable for suggested improvements on your projects, starting with your most important and valuable projects. 3. Use Opus and GPT-55 to brainstorm what your hardest technical problems are across your projects and then ask Fable to propose solutions to them. 4. Use GPT Pro through Oracle to review Fables output.

Now, this reflects a lot of the advice that was around when Fable 5 first came out. basically that what it was for was your hardest challenges and that it was almost too powerful for routine day-to-day things. Now, I agree with half of this. The half where Fable is very good at your most difficult, specifically technical problems. And to the extent you have some of those, whether you're technical or not, if you're working on some hard coding project, obviously spend as much time as you can using Fable 5 for that. But one thing that already in the very first couple of days of using Fable 5 that I disagreed with the common sentiment around was that you wouldn't necessarily notice the difference using Fable 5 on more routine banal non-technical tasks.

In my experience, that was completely wrong. Likely the two most common categories of tasks that I have outside of coding, building, and prototyping things is strategic thinking and writing. Now, on strategy, Fable 5, in my limited experience, blew GPD 55 and Opus 48 out of the water. Both of those two models are extremely steerable. They are overly deferential to pushback and tend to overinterpret and sort. For example, if you ask them a strategic question and then try to say something to reduce sycophancy like, don't just accept what I'm saying is true, provide pushback if it's warranted,

those models will assume 100% that their response will only be successful if there is pushback. Then, however, if you push back on them, they will cave almost immediately and go out of their way to justify whatever it has become clear they think you're trying to get them to say. Fable 5 didn't do that. In my interactions with Fable 5, as I was debating with it, it would frequently accept part of my pushbacker ideas while sticking to its guns on other parts. That is behavior that I have never seen from any other model and instantly made it a thousand times more valuable when it came to any sort of strategic thinking or iteration. Given that every single one of you,

I guarantee has strategic questions that you're working through at any given time, this is something that I would immediately experiment with in Fable. And by the way, it also has the benefit of not really consuming all that many tokens, so it's not going to use up that 50% usage limit particularly quickly. The second common sentiment that I want to push back on is around writing. Now, arguably, Every is the most thoughtful tester when it comes to new models in writing. And in their initial vibe check, it wasn't in their estimation all that much better. They called it sharp judgment trapped in familiar prose. They wrote,

Every's writing benchmark found that Fable's writing is clear, but still short of human judgment on what to include in a piece of writing and how to structure it. The benchmark asks the model to write an introduction from scratch, fill in a missing paragraph from context, and deliver a promotional email, a LinkedIn post, an ex post. On the editing side, it asks the model to replicate human edits and detect common AIisms in a deliberately robotic draft. And set to extra high effort for this, it basically scored between Sonnet 4-6 and Opus 48. Now, on a particular type of writing task, this is not what I found. In my not tests, but real-world uses of Fable 5 for writing, I found that it was

way better at instruction following and fell into far fewer of the common AI traps. It's writing wasn't overly wrought and try hard. It had fewer of the standout AIisms like it's not this, it's that. And while I would want more Repsen to say definitively that it's better for all types of writing, my suspicion is that especially in situations where you have a fairly clear rubric of what a good example of writing looks like, I think it's going to be able to do a much better job at meeting that standard than those previous models were. Now, that doesn't mean that it's any better at blank page writing, but given how many of people's

writing use cases for AI are saying, here's all the examples of what has been done in the past, do this type of thing in the future. My experience so far is that Fable 5 is much better at that. Look, when all is said and done, it is extremely, extremely welcome news that it is coming back to everyone. While I don't think that you should change your Fourth of July plans to stay inside hacking at some big project, I certainly wouldn't blame you if you did. For now, that's going to do it for the AI Daily Brief. Appreciate you listening or watching as always. And until next time, peace.