Substack

Cut Coding Costs with GLM-5.3 in Claude Code & Codex

The author explains how moving expensive coding workloads from high-cost models to GLM-5.3 can significantly reduce API bills while keeping existing toolchains intact. GLM-5.3 can be used inside Claude Code and Codex; Z.AI’s GLM Coding Plan starts at $18/month and offers lower metered usage (cheaper outside Z.AI peak hours in Singapore). The post outlines a short setup (including a six-line handoff and a cost-per-accepted-result scorecard), shows an example overnight Codex run that exceeded $300, and argues that routing suitable jobs to GLM-5.3 can pay for the GLM plan quickly. The full technical guide and configuration are available to paid subscribers.

AI Writing Vibe Shift Accelerates

The newsletter analyzes recent signals showing a shift in attitudes toward AI-generated writing: Substack is partnering with Pangram to detect AI-written posts, LinkedIn introduced a 'Seems Like AI Slop' feedback button, Anthropic announced an invisible watermark (and a watermark API) for Claude outputs to meet EU transparency rules, publishers and agents are drawing lines on suspected AI-authored manuscripts, and several tech companies and marketing teams are instituting internal guidelines discouraging AI-written copy. The piece frames these moves as part of a broader industry move toward transparency and protecting brand voice.

Author Warns of Speculative Risks in Generative AI

An opinion essay by Gary Marcus (published on Substack) likens certain financial behaviors in the generative AI industry to check-kiting. The author notes circular financial elements—for example, borrowing against OpenAI shares to buy more of the same—and questions whether the large amounts of debt being issued in the sector can ultimately be repaid, calling the investments highly speculative. The piece frames this as a structural risk but stops short of calling it outright fraud, urging a different analytical perspective on the industry's financial dynamics.