This week, the Elon Musk / OpenAI trial gets going, Air taxi demos in New York, and someone created a powerful LLM based only on text available pre 1930 - vintage AI.
On to the main stories:
- Pentagon inks deals with seven AI companies for classified military work. This news follows on from the public fight over AI for military applications a few months ago between Anthropic and the US Department of War. SpaceX, OpenAI, Google, Nvidia, Microsoft, and Amazon Web Services have now all signed contracts with the US Department of War that permit "any legal use" of the technology and would not be subject to control by the company providing the AI capability. Anthropic is not on this list, but I would be surprised if that doesn't happen in the next 12-24months. Not necessarily because Anthropic goes back on its stance, but because they may be forced into it by government demands. If Claude appears to be the strongest model for certain use-cases, it's unlikely the US military will want to use it.
- Thoma Bravo’s $5.1B Medallia wipeout deepens software, private credit reckoning. If you need more evidence that Enterprise SaaS Software is trouble, this is it. One of the world's largest private equity funds is essentially writing down the value of its giant software assets. Medallia is a high-end solution for customer surveys, and it falls into the category of things that are perhaps going to be much easier to do with AI. Part of the failure is due to leverage, but part is surely due to the fact that there are dim prospects for growth. Medallia is no doubt a well-developed product, but... the business has no network effects, and it will become easier and easier to replicate the functionality itself. Pricing power evaporates in that situation.
- Amazon Bedrock now offers OpenAI models, Codex, and Managed Agents. OpenAI models are finally coming to Amazon (in agent form). After amending its agreement with Microsoft, OpenAI jointly announced a new service with Amazon. The service essentially allows you to run a contained version of OpenAI's Codex Agentic system within Amazon's Bedrock AI system. I think this will be significant as companies look for ways to run powerful, long-running agents, but in controlled ways. All the cloud providers will likely offer something like this. For those who prefer to run the agents in their own infrastructure, there are platforms like Jozu that allow you to do that.
- Anthropic joins the Blender Development Fund as Corporate Patron. In parallel to their Claude for creative work announcement, Anthropic has joined the governance of the blender foundation and integrates tightly with a number of other creative tools via python APIs. I'm not sure how easy it will be initially to do work via Claude but I think this is a singificant acceleration of the trend of using models to drive other software. By enabling frontier models to manipulate data structures such as blender files. This tool use will rapidly accelerate what models can do. It will be interesting how human and AI workflows blend. On the other hand it will likely mean that nacent AI features being added to products from Abelton to Adobe and others will probably wither on the vine.
- Humanoid robots to become baggage handlers in Japan airport experiment. Moving even closer to the physical world, JAL and Tokyo's Haneda airport are partnering to trial Unitree humanoid robots for luggage handling. If this announcement had been made in the UK or the US one might begin to hear grumblings about worker rights. In Japan though there is an acute labor shortage so adoption might meet a lot less resistance. Judging from the video, the robots might take some time to be ready for prime time.
Wishing you a great weekend.
