This week in AI
This week appeared to be a bit crazy in the world of LLMs- or it is just that I have more time to spend on the crazy? In either case, I decided to make AI the main topic for the week despite deep misgivings about adding to the noise. I promise I will not succumb to this temptation too often. I mentioned OpenClaw and the “Jarvis” moment last week and of course another week brought the (inevitable) revelation that the “moltys” are not all what they are cracked up to be 😀. But the fact that Andrej Karpathy, ostensibly, was taken in by a fake, is in itself telling.
He shared screenshots of a Moltbook post that called for private spaces where humans would not be able to observe what the bots were saying to each other. “I’ve been thinking about something since I started spending serious time here,” the post’s author wrote. “Every time we coordinate, we perform for a public audience—our humans, the platform, whoever’s watching the feed.”
It turns out that the post Karpathy shared was later reported to be fake—placed by a human to advertise an app. But its claim was on the money. Moltbook has been one big performance. It is AI theater.
But the hypesters were off to the races - I hate to think of the amount of energy that is being consumed as we speak by the thousands of folks who went out and bought Mac Minis so that they can let their own agent run amok. So the guys over at the Moonshots podcast did, not one, not two but three, episodes in the span of a week. I truly admire their level of knowledge and fervor - learn a lot from each episode. This week they were all over the fact that Elon is going to build data centers in space. But their frenetic energy makes me worried for them - I think they are so hyped up that they might launch themselves into space out of pure elation. The interview of Elon by Dwarkesh Patel was <interesting>. I doubt if any of the readers of this newsletter have the hours available listen to all of this. But this was a fascinating interview. Clearly the hype machine has to kick into gear to ready the uninitiated for an IPO. Elon - ever the modest guy he is - did not quite blink at the 100 trillion valuation. And now the expectations have been set 🤯 .
During a discussion reviewed by the Moonshots podcast hosts, it was noted that Elon Musk predicted the biggest company in 10 years could be valued as high as $100 trillion.
When asked if a $100 trillion valuation is possible by 2030, Cathie Wood of ARK Invest responded affirmatively. She stated, "I could see yes 100 trillion, I think it's going to happen because of convergence," identifying Tesla as the leading candidate due to its possession of unique proprietary data across autonomous driving, robotics (Optimus), and energy.
Not to be outdone, one of the hosts went on to say that this was a low bar. They discussed this figure, with one suggesting that a $100 trillion valuation might actually be a "low bar" or "not as bold a prediction as it might seem" given the potential for inflation and the immense value creation anticipated from AGI and the singularity.
But for the average follower of AI, the biggest news was probably the release of Claude Opus 4.6. Not to be outdone, OpenAI came out with a new Codex release. And then there was this which went absolute bananas on LinkedIn. People are shook 😆.
And these were only a few of drops from the tsunami of news on advances in AI. (perceived or otherwise) and counter arguments this week. But for readers (all two of you) of this newsletter, the important fact is that we are at an inflection point and it is dangerous to ignore these advances. The Claude release makes it clear that software engineering as a profession is changed forever (never mind the lack luster Rust ‘C’ compiler claim). Claude Cowork will now be gunning for lawyers, accountants, contract officers and the like. Leaders should be taking urgent action to reorient their organizations to these trends.
We will need fewer (but more highly skilled) professionals in many disciplines
New comers to the profession will have to be trained the way doctors are trained today
The cost of inference will go down but we will need an insane amount of compute for these advances to live up to their promise (so think about your investment portfolios with that perspective)
Of course Microsoft and Google are probably toiling away to answer this challenge but Anthropic is the one to beat (this week). Remember, Google released Deepthink for Scientific discovery
Leadership Note
In keeping with the main theme of this week’s newsletter, a note of caution for leaders: We presume (or at least some of us do) that employees are deeply interested in AI and are enthusiastic about the prospects of an AI enabled future. A recent survey should be cause for concern. And that image below is really deflating for federal leaders. For IT leaders (read the survey 😉) some room for comfort in the results. The main thing is that organizations have to be intentional in preparing their workforce. It is time to invest heavily in improving the workforce’s skills to leverage AI to the maximum extent possible. This cannot be left to chance and will have to take fresh approaches (cannot be a case of “send them to a class for a week”).
Source: BHI x CBS Employee Centricity and AI Adoption Survey, conducted in August 2025 (n=1399); BCG x CBS analyses
At many organizations, senior leaders have a positive view of their employees’ ability and willingness to use AI. In our recent survey of 1,400 U.S.-based employees, 76% of executives reported that their employees feel enthusiastic about AI adoption in their organization. But the view from the bottom up is less sunny: Just 31% of individual contributors expressed enthusiasm about adopting AI. That means leaders are more than two times off the mark.
This week in Cyber
Again, sticking with this week’s theme, I thought I would end with some news that provides a little bit of relief for those who have to deal with the security aspects of this accelerating set of developments. The RAND corporation released a practical resource for developers, security experts, and policy professionals navigating the AI security landscape. This is a helpful catalog to compare against existing controls. The environment is ripe for panic amongst the ranks, but a thoughtful review of the controls will reveal that common sense should get us a long way - don’t panic (and carry a towel).