Tagged OpenAI Codex · show all articles
Long tasks fail through state corruption, weak observations, compounding decisions, and false completion—not merely because the model writes a bad line of code.
Coding-agent security is a systems property: untrusted observations become dangerous when they can steer broad tools, credentials, mutable tests, and irreversible actions.
Codex is best understood as a changing stack of models, tools, execution boundaries, state, and review mechanisms—not as a single code-generating model.
A coding-agent score is conditional on its tasks, harness, budget, environment, oracle, and date; useful evaluation preserves those conditions and measures operational value.
Today's Codex is not a continuously scaled 2021 model; it emerged from a sequence of shifts in synthesis, language modeling, evaluation, tools, execution, and human–agent interfaces.
Reliable delegation begins before the prompt: repositories need explicit contracts, reproducible environments, bounded authority, discriminating checks, and reviewable evidence.
Inference can be cheap while accepted software remains expensive; the correct denominator is valuable, reviewed change—not tokens, diffs, or agent hours.
The consequential differences are in trust boundaries, configuration, state, hooks, subagents, and workflow—not whichever model topped a changing leaderboard this week.
The decisive variable is not how much code agents can emit, but whether organizations can specify, verify, trace, and authorize machine-produced changes at comparable speed.
The right Codex surface follows from where state lives, how quickly a human must steer, and which execution boundary can be reproduced and audited.