~/blog · filtered by category

runtime.

12 entries filed under this signal. ← back to all entries

[blog/category/runtime] · reverse chronological$ ls -la --filter=runtime

2026-07-20runtimeThe run died at 2am: recovery as a first-class workflowThe agent hit a rate limit, or lost the thread, or the context filled and it quietly forgot episode one. You wake up to a half-finished branch. Right now recovery is heroism. It should be a workflow.#recovery#checkpoints#agents#reliability#runtime7 min./read →2026-07-20runtimeRunning eight sessions: worktrees, parallelism and the cost of being the schedulerParallel agents give you throughput and hand you a second job you didn't apply for. Here's how the worktree setup works, and what being the scheduler actually costs.#parallel-sessions#worktrees#agents#runtime#paelladoc8 min./read →2026-07-20runtimeOrchestrating multiple coding agents without them poisoning each otherMore agents is more throughput and more ways to collide. The three things that keep parallel agents from contaminating each other: isolation, contracts, and reconciliation.#multi-agent#orchestration#agents#runtime#paelladoc8 min./read →2026-07-20runtimeMemory for coding agents: what should persist outside the chatNot everything an agent learns deserves to be remembered, and the chat is the worst place to keep the parts that do. A practical taxonomy of what to persist, and in what form.#agent-memory#agents#ai-coding#runtime#paelladoc7 min./read →2026-07-20runtimeYour rules file is lying to your agents: keeping instructions aliveThe rule that pointed at a script you deleted. The convention nobody follows anymore. Your agents read all of it as gospel, and a stale instruction never throws an error. It just quietly steers the work wrong.#claude-md#agents-md#stale-rules#agents#runtime7 min./read →2026-07-20runtimeHarness engineering: the discipline forming around agent operationsThe model writes the code. Everything around the model, the loop that feeds it context, runs its tools, checks its output and recovers its failures, is a system someone has to build. That system has started to get a name.#harness-engineering#agents#tooling#reliability#runtime7 min./read →2026-07-20runtimeCompaction eats your decisions: managing the context window on long workThe window fills, the session compacts, and the constraint you set three hours ago is gone. The agent keeps going, confident, having quietly forgotten the thing that made the work correct. On long jobs, the context window is a budget.#context-window#compaction#agents#context-engineering#runtime7 min./read →2026-07-20runtimeContext loss between sessions: the productivity killer nobody budgets forThe tokens are cheap. Re-explaining the same decision to a cold agent for the fourth time is not. Here's what actually disappears when a session ends, and what to do about it.#context-loss#agents#ai-coding#runtime#paelladoc8 min./read →2026-07-20runtimeBuilding software with AI agents: the operating guideThe code is written. The job that stayed is running the loop around it: define, specify, execute, verify, learn. This is the map of that work, one stop per part.#building-with-agents#ai-coding#agents#paelladoc#runtime#spec-driven-development14 min./read →2026-07-20runtimeAgents will drift: engineering for divergence instead of praying against itDrift isn't a bug you can prompt away. It's what agents do. The useful move is to design a system that expects divergence and catches it, instead of hoping this run behaves.#agents-will-drift#divergence#agents#runtime#paelladoc8 min./read →2026-07-20runtimeAGENTS.md, CLAUDE.md, cursor rules: where your decisions should actually liveEvery tool wants its own rules file. Most of the comparisons only argue about code style. The harder question is where the product decisions go, because that is what actually rots when you get it wrong.#agents-md#claude-md#cursor-rules#agents#runtime6 min./read →2026-07-20runtimeHandoffs: passing work between agents without losing the plotOne agent scopes it, another builds it, a third reviews it, and somewhere in the relay the reason it was being built that way falls on the floor. The handoff is where agent work quietly loses the plot.#handoffs#agents#context#orchestration#runtime7 min./read →

— end of log · 12 entries in runtime —