Bartłomiej Krupa engineers the harness around LLMs — agentic systems built for reliability and low cost. Deep dives, atomic notes, and a curated radar on LLM optimization, AI cost reduction, harness engineering, and agentic infrastructure.https://bartlomiejkrupa.dev/https://bartlomiejkrupa.dev/articles/claude-model-tier-comparison/https://bartlomiejkrupa.dev/articles/claude-model-tier-comparison/Opus 4.8 orchestrates, Haiku 4.5 does the parallel grunt work, Sonnet 4.6 is the default in between. Choose by the job and the token bill, not the benchmark.Tue, 30 Jun 2026 00:00:00 GMTclaudeopussonnethaikupricingagentsmodel-selectionhttps://bartlomiejkrupa.dev/articles/context-engineering-beats-a-bigger-window/https://bartlomiejkrupa.dev/articles/context-engineering-beats-a-bigger-window/Claude Code runs up to a 1M-token window on flagship models, but a bigger window doesn't fix a cluttered one — every turn replays the whole history and quality drops before the cap.Tue, 30 Jun 2026 00:00:00 GMTclaude-codecontext-windowcontext-engineeringtokenssubagentshttps://bartlomiejkrupa.dev/articles/why-agents-ignore-your-claude-md/https://bartlomiejkrupa.dev/articles/why-agents-ignore-your-claude-md/A bloated CLAUDE.md gets filtered out as noise. Keep it to universal instructions and move everything else behind progressive disclosure.Tue, 30 Jun 2026 00:00:00 GMTclaude-codeclaude-mdcontext-engineeringagents-mdprogressive-disclosurehttps://bartlomiejkrupa.dev/garden/claude-md-universal-only/https://bartlomiejkrupa.dev/garden/claude-md-universal-only/Put only universally applicable instructions in CLAUDE.md; task-specific content raises the odds an agent treats the whole file as noise and ignores it.Tue, 30 Jun 2026 00:00:00 GMTclaude-mdclaude-codecontext-engineeringhttps://bartlomiejkrupa.dev/garden/context-engineering/https://bartlomiejkrupa.dev/garden/context-engineering/Context engineering is curating what enters the model's context window each turn, rather than relying on a bigger window to absorb the clutter.Tue, 30 Jun 2026 00:00:00 GMTcontextclaude-codetokenshttps://bartlomiejkrupa.dev/garden/lost-in-the-middle/https://bartlomiejkrupa.dev/garden/lost-in-the-middle/Language models use information at the start and end of a long context far more reliably than the middle — recall drops well before the token limit.Tue, 30 Jun 2026 00:00:00 GMTcontextretrievaltokenshttps://bartlomiejkrupa.dev/garden/subagent-context-isolation/https://bartlomiejkrupa.dev/garden/subagent-context-isolation/A subagent runs in its own context window and returns only a summary, keeping heavy intermediate tokens out of the parent conversation.Tue, 30 Jun 2026 00:00:00 GMTsubagentsclaude-codecontextmcp