Voice Agents Are Finally Real. Your Architecture Isn't.The launch of GPT-Realtime-2 landed quietly for a supposedly major release. No benchmark fireworks, no "state of the art" claims. Just three streaming audio models and a developer guide about reasoning effort, preambles, and long-session state manage...May 11 · 4 min readmehaisi.hashnode.dev
A Million Tokens Changes Nothing If Your Agent Can't Remember YesterdayA Million Tokens Changes Nothing If Your Agent Can't Remember Yesterday Context windows are having their megapixel moment. DeepSeek-V4 ships with a million-token capacity and suddenly everyone's building "unlimited memory" agents that still forget wh...May 10 · 4 min readmehaisi.hashnode.dev
The Line Between Vibe Coding and Production Is DissolvingThe Line Between Vibe Coding and Production Is Dissolving Simon Willison recently admitted something that should alarm anyone building production AI systems: the distinction between "vibe coding" and responsible agentic engineering has started to blu...May 9 · 6 min readmehaisi.hashnode.dev
Correctness Before Corrections: What vLLM's RL Migration Teaches Us About Agent ReliabilityCorrectness Before Corrections: What vLLM's RL Migration Teaches Us About Agent Reliability The vLLM team just shipped V1. The migration guide isn't about new features. It's about a fundamental inversion: they stopped trying to patch correctness issu...May 8 · 3 min readmehaisi.hashnode.dev
The Line Between Vibe Coding and Production Is DissolvingThe Line Between Vibe Coding and Production Is Dissolving The distinction between "vibe coding" and "agentic engineering" was never about the tools. It was about accountability. And that line is dissolving faster than most teams are prepared to handl...May 7 · 4 min readmehaisi.hashnode.dev
