Superlite: hand an agent a spec, come back to tested work

kept by lunawisp256 oct 19

Superlite cuts Claude Code build times by roughly half by parallelizing tasks, limiting human approvals to one initial batch, and enforcing workflow rules via code.

Fable's judgement
The author learned to let Claude Code's Fable model use its own judgement to delegate coding tasks to lower-power models, improving efficiency and token usage.
also on Haiku, Sonnet, Opus, Claude Code, #automation, #ai-coding, #productivity, #developer-tools
Why I believe in SOTA models over custom ones
The future of AI is powerful general models becoming cheaper, not narrow custom models, because specialized tasks still benefit from broad intelligence.
also on Haiku, Sonnet, Opus, #developer-tools
Get Shit Done: A meta-prompting, context engineering and spec-driven dev system
Get Shit Done (GSD) is a spec-driven dev system using meta-prompting and subagents; users report massive LOC output but also heavy token burn, mixed results, and concerns about review and safety.
also on Superpowers, Claude Code, #ai-coding, #software-engineering, #productivity, #developer-tools
How Kenn is doing Agentic Engineering – Wes McKinney
Wes McKinney details Kenn's agentic engineering process, emphasizing human oversight in design and verification, using custom tools like Superpowers and roborev to maintain quality while merging hundreds of PRs weekly with low bug rates.
also on Superpowers, #automation, #ai-coding, #software-engineering, #developer-tools
Claude's Voice Mode Just Got Smarter
Anthropic updated Claude’s voice mode to use Sonnet/Opus models, add Gmail/Slack context, and improve language support, while noting its turn‑based design limits naturalness.
also on Sonnet, Opus, #productivity, #developer-tools