AI digest: Code agents and China's robot ambitions
New coding workflows, open-source speech breakthroughs, and China's push for AI-powered one-person companies.
This week brought practical tools for developers and a glimpse of where AI automation might head next.
Garry Tan ships gstack for structured AI coding
Y Combinator’s Garry Tan released gstack, an open-source toolkit that wraps Claude Code into eight distinct workflow skills for planning, review, QA, and shipping. The approach separates concerns rather than throwing everything at one massive prompt. Smart move given how messy AI coding gets when you skip the structure.
Hume AI drops TADA speech model under MIT licence
Hume AI open-sourced TADA, claiming it’s five times faster than competing speech models with zero hallucinated words in testing. The model processes text and audio in sync, which could fix the annoying delays we see in current voice AI. MIT licence means we’ll actually see this used in production, unlike the research-only releases that gather dust.
China backs “one-person companies” with AI agent subsidies
Seven Chinese local governments launched million-dollar funding programmes for OpenClaw projects, targeting businesses run by one founder with AI agents as employees. This isn’t just research funding, it’s an industrial policy bet on AI replacing traditional teams. Whether it works or creates expensive ghost companies remains to be seen.
Meta plans massive layoffs to fund AI spending
Meta reportedly considers cutting up to 20% of its workforce to offset its aggressive AI infrastructure investments. The $600 billion AI bet needs funding from somewhere. Shows how seriously big tech is taking the infrastructure race, even if it means gutting existing teams.