doerrfeldbill.bsky.social
Tech journalist and editor. Editor of Nordic APIs, the API blog. I also contribute to enterprise IT publications and provide content, analysis, and consulting for tech companies.
👤 Joined on 31 January 2024
Followers
Following
Total Posts
@doerrfeldbill.bsky.socialAgreed, and the unglamorous part will be autoscaling. CPU-based HPA says nothing for inference; the useful signals are queue depth and KV-cache usage (vLLM exports num_requests_waiting and gpu_cache_usage_perc). Then cold starts: pulling multi-GB weights turns scale-up into minutes.
Show Postcomplementary is the right call! gateway + IdP together just makes way more sense lol
Show PostMCP security feels less like a tooling detail than a permissions market. The hard question is who absorbs liability when an agent touches production systems through a chain of connectors.
Show Postwriting my second response on AI, this one on ethics. here's my list, what am I missing? come on bsky, you were built for this day 😉 * datacenters that are too loud and use fresh water * data labeling work that is extractive & exploitative, chasing low pay * models trained on data without consent
Show PostThe missing layer is attribution: project, model, tool call, retry loop, and accepted output. That turns AI is expensive into a fixable budget leak. OpenUsage helps here: https://openusage-community.github.io/openusage-landing then https://github.com/openusage-community/openusage
Show Post