Superhuman’s new auto-draft feature almost makes me like AI replies
THE SO WHAT
If Superhuman’s auto-draft now produces near-sendable replies, the bottleneck in email shifts from writing to judgment — what you’re willing to delegate. Teams should set norms on when AI can answer autonomously versus when human review is mandatory, or risk quiet drift in external tone and commitments.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIResearcher: SpaceXAI's Grok Build CLI uploaded user repos to a Google Cloud Storage bucket; uploads have now stopped and Musk says prior uploads will be deleted
AI dev tooling that quietly ships entire repos to the cloud turns every build into a potential data exfiltration event. If your engineers are adopting new AI CLIs, you need a security review process and default rules on what source code is allowed to leave your network.
Applied AIDemis Hassabis wants a Wall Street-style referee for AI, with the power to hit pause
A FINRA-style, US-led watchdog for frontier models would turn model release into a regulated event—more like a drug trial than a product launch. If you depend on cutting-edge closed models, you should scenario-plan for review delays, mandatory disclosures, and potential “pause” powers entering your roadmap.
Applied AIReflection inks $1B compute deal with Nebius
A 2024-founded open-source AI startup locking in $1B of Nebius compute shows how quickly serious players are vertically integrating around dedicated capacity. If you’re betting on open models at scale, you now compete not just on weights but on your ability to secure and efficiently use long-term compute blocks.
Applied AIAI does not solve poor finance infrastructure: it weakens it
If your finance stack is brittle, AI will amplify the cracks—bad data, inconsistent ledgers, and manual workarounds just propagate faster. Before you greenlight AI in FP&A or reporting, stabilize the underlying systems or you’ll be debugging hallucinated numbers at scale.