 Saturday Curation AI moved into the room where work happensSun 3 May 2026 · 6 items Six things worth your time from the week ending 2 May 2026.
OpenAI comes to AWS, and AI choice becomes a buyer question OpenAI models, Codex, and managed agents are now coming to Amazon Bedrock in limited preview. The useful point is not cloud drama. It is procurement. More companies will be able to use OpenAI inside the systems their IT, security, and finance teams already approve. For professionals, the best AI tool may increasingly be the one your organisation can actually deploy. OpenAI + AWSClaude moves into the creative tools where the work lives Anthropic added Claude connections for Adobe, Blender, Ableton, Autodesk Fusion, SketchUp, Splice, Affinity by Canva, and Resolume. The important move is simple: less copying work into chat, more asking for help inside the actual file. If your work lives in specialist software, the AI assistant that understands the working file will beat the one waiting in a blank chat box. AnthropicDeal teams get AI without exporting the sensitive room Affinity and Datasite both launched ways for AI assistants to work against live relationship and deal-room data. This matters beyond private capital. The old pattern was copy sensitive material into a chat and hope the permissions held. The better pattern is for the AI to come to the governed system, read only what it is allowed to read, and leave an audit trail. Affinity + DatasiteClaude Security turns AI review into a standing control Claude Security is now in public beta for Claude Enterprise customers. It scans software, checks whether the finding is real, and drafts a fix for review. Even if you never touch code, watch the shape of this product. AI is moving from one-off help to standing controls: regular reviews, proposed fixes, and evidence that someone can approve or reject. AnthropicCodex gets goals, and the management pattern becomes visible Simon Willison's short note on Codex's new /goal command is worth reading even if you do not code. The pattern is the point: give the AI a goal, give it a budget, and let it keep working until it believes the job is done. That is a management contract, not a prompt trick. The human role shifts to setting the goal, the budget, and the review bar. Simon WillisonAI values are now procurement terms The Pentagon signed classified-network AI agreements with SpaceX, OpenAI, Google, Nvidia, Reflection, Microsoft, AWS, and Oracle. Anthropic was left out after a dispute over military-use terms. The point for every buyer: a vendor's values are no longer brochure copy. They can decide whether the tool is deployable inside your organisation at all. DefenseScoop + TechCrunchWant to build a system where your AI actually knows how you work? Take the audit |