LLM Wrappers vs Harnesses: why the system around the model is where value lives
MartinGTobias · x · 2026-10-30
MartinGTobias explains his shift from dismissing "LLM Wrappers" to going all-in on "Harnesses." Wrappers are just elaborate prompts telling the brain how to think — thin, low incremental value. Harnesses give the model a full body: tools, memory, context control, integrations (think Claude Code for coding), delivering value that persists independently of which model sits inside. His test: if the model changes tomorrow, what's left?
More from coding & agent
- Vicem: Open-Source Tool Hands Off Task State Between Coding Agents — NoConclusion8361 · 2026-10-30
- Subagents Are the Only Claude Code Tasks Worth Downgrading to Cheaper Models — Interesting-Bite316 · 2026-10-30
- Gemini 4 tops public coding benchmarks but drops to bottom on private tasks — Hesamation · 2026-10-30
- User builds 20-bot Grok suite to track wins, prep 1:1s and close promises — NicoVerderosa · 2026-10-30
- Why 30-year-old BM25 beats vector embeddings for agentic search, per search vet Jo Kristian Bergum — jobergum · 2026-10-30
- Self-hosted ComfyUI film studio hits 100+ features: crowd removal, lip-synced dialogue, per-render workflows — Key_Cap_8111 · 2026-10-30