Sunday roundup: afterthoughts on productivity and pace
I published a post this week that was harder to write than most. “I’m Not Apologising for Being Productive” came from a place I don’t usually write from. Not about a technology or a tool or a shipping date. About the friction of moving differently in an industry that rewards sameness.
The response has been good. Better than good, actually. A few DMs from people who recognised the pattern, one from someone I used to manage, which hit differently. It’s easy to forget that writing about the vulnerable parts of how you work is a form of doing the work itself. Not the commercial kind. The kind where you help someone else feel less alone in their own head. That alone makes this Sunday roundup worth writing.
Sunday roundup: elsewhere this week
Deepgram announced a partnership with Fortanix and NVIDIA on June 1, bringing voice AI to on-premises confidential computing. This is the kind of release that doesn’t make a splash on TechCrunch the way a model benchmark does, but it changes the procurement conversation for every regulated industry that was sitting on the sidelines. Healthcare, finance, government. Organisations that cannot route patient audio or financial calls through a cloud API. I watched the internal prep for this one. It was the kind of shipping that matters more for the market than for the leaderboard.
Meanwhile Microsoft held Build, the Tokenomics Foundation announced itself over at the Linux Foundation, and Anthropic published a warning about recursive self-improvement that got picked up by CNN. The Jun 8 post will cover the industry stories properly.
What’s ahead
Next week I want to dig into something practical about Deepgram Flux and turn detection. The model-native EndOfTurn events replaced VAD silence timers for us and the difference in agent behaviour is real. That deserves its own post, not a Sunday roundup sentence.
Thanks for reading this week. See you next one.
FAQ
What is the Deepgram-Fortanix partnership?
Deepgram partnered with Fortanix and NVIDIA on June 1, 2026 to deliver voice AI in on-premises environments using confidential computing. This lets regulated industries run speech-to-text and text-to-speech inside hardware-isolated enclaves where neither the cloud provider nor the infrastructure host can access the audio data or model weights.
Why is on-premises voice AI important?
Healthcare, finance, and government organisations often cannot send audio data to cloud APIs due to compliance requirements like HIPAA, GDPR, or national data residency laws. On-premises deployment with confidential computing removes that barrier, making voice AI accessible to organisations that previously had to build their own solutions or go without.
What is the Tokenomics Foundation?
Announced by the Linux Foundation during the week of June 1, the Tokenomics Foundation is a standards body aiming to bring cost discipline to AI token spending, similar to what FinOps did for cloud infrastructure costs. It plans to define metrics like cost-per-intelligence, tokens-per-watt, and standardised billing definitions across model providers.