surface
cli, ide & app tradeoffs
terminal, desktop app, ide, weba guide to coding agents in production software (projects, startups & big tech)
practical guidance for choosing, operating, and reviewing coding-agent systems when the work has real users, real constraints, and consequences.
01 / taxonomy
understand the layers of every agentic system
it's important to know which layer holds the solution space to your bottlenecks
harness
tools, runtime & memory
codex, claude code, qwen codemodel
valuing inference
hosted, open-weight, localorchestration
how parallel work is planned & coordinated
worktrees, agents, control planes02 / operating models
compare setups most commonly used
terminal + first-party appdeep local control with one supervision surfaceextra app and execution-location complexityide-integrated agentselection context, diagnostics, inline revieweditor coupling and local resource pressureterminal-only agentlow interface overhead and scriptabilitymore manual coordination across concurrent worklocal or open stackmodel choice, privacy, and infrastructure controlhardware, serving, and evaluation ownership
03 / current guides
field notes with visible limits
codex
hands-on + source-verifiedcli, desktop, ide, cloud, worktrees, permissions
claude code
source-verified, run pendingterminal, desktop, ide, agents, hooks, permissions
shared operating system
hands-on + inferenceauthority, worktrees, verification, handoffs
04 / reproducible field lab
the claims should survive inspection
the protocol records the task, base commit, runtime, interventions, checks, artifacts, redactions, and limitations. missing measurements stay missing.
inspect the protocol and runs05 / latest material change
the repository becomes the publication
introduced the layered market model, evidence contract, field lab, and paired codex and claude code guides.
read changes