Your agent's failures are silent: measuring failure modes in production

The failure that hurts is the one that doesn’t throw. A traditional service fails loudly: an exception, a 500, a stack trace, a red line on a dashboard. An agent fails quietly. It runs to completion, returns a confident answer, exits zero — and the answer is wrong, or it spent forty steps and $3 to conclude it couldn’t do the thing, or it looped politely until it hit a cap nobody’s watching. The $200 postmortem was a loud failure I happened to catch. The expensive ones are the quiet failures you never labeled, because you can’t alert on a category you don’t record.

What to measure when your agent works covered the happy path. This is the inverse: what to measure when it doesn’t, and how to know that it didn’t.

Exceptions are the tip of the iceberg

Here’s the trap. Your agent has a try/except at the top of the loop. Exceptions get logged, counted, alerted. Your error rate looks like 0.5% and everyone’s happy. Meanwhile:

• The agent hit its step cap and returned whatever it had — no exception, just a truncated answer.
• A tool returned {"results": []} and the agent treated empty as “done.”
• The model produced a plausible-looking answer that’s factually wrong — a perfect run by every mechanical measure.
• The agent looped between two states for thirty steps, then gave up — loop drift, which raises no error at all.

None of those increment your exception counter. All of them are failures. Your real failure rate isn’t 0.5%; it’s 0.5% that you can see plus an unknown, larger number you can’t. Step one is to make every run end in a labeled outcome, not just “exception or not.”

A run-outcome taxonomy

Every agent run should terminate with an explicit, recorded outcome. Not a boolean — a category. The minimum useful set:

The point of the taxonomy is that these have different fixes. budget_exhausted means your caps are too tight or your task is too big — raise the cap or decompose. gave_up means a capability or tool gap — the agent knew it was stuck, which is the good failure. looped means your loop lacks a progress check. wrong is the dangerous one, because it’s indistinguishable from success at runtime. Collapsing all of these into “error rate” throws away exactly the information that tells you what to do.

Instrument each mode

Terminal-state logging. The single highest-value change: make the loop’s exit path assign an outcome. If you fall out of the loop because a cap fired, that’s budget_exhausted — don’t let it masquerade as success.

def run_agent(task, step_cap=40, token_cap=200_000):
    state = init(task)
    for step in range(step_cap):
        action = model_step(state)
        if action.is_terminal:
            outcome = "gave_up" if action.type == "cannot_complete" else "success"
            return finish(state, outcome, step, tokens(state))
        if tokens(state) > token_cap:
            return finish(state, "budget_exhausted", step, tokens(state), cap="token")
        state = apply(action, state)
    return finish(state, "budget_exhausted", step_cap, tokens(state), cap="step")

def finish(state, outcome, steps, toks, cap=None):
    log.info("agent_run_end", outcome=outcome, steps=steps, tokens=toks, cap=cap)
    return state.result, outcome

Now outcome is a dimension you can group by. “What fraction of runs hit the step cap this week?” becomes a query instead of a mystery.

A progress detector for looped. A cheap one: hash the salient state (open goals, last tool called + args) each step and count repeats. Three visits to the same hash without a new goal closing means no progress — break with looped. This turns an invisible, expensive non-termination into a labeled, bounded event you can alert on.

Post-hoc verification for wrong. This is the hard one, because wrong looks identical to success while the run is happening. You cannot catch it at runtime; you catch it after, on a sample. Run a check on some fraction of “successful” outputs — a schema validation, a re-derivation, a cross-check against ground truth where you have it, or an LLM-as-judge with all the caveats that come with it. The metric that matters is the gap between your mechanical success rate and your verified success rate. If mechanical says 98% and verified-on-sample says 82%, your real failure rate is 18%, and 16 of those points were invisible.

The two numbers that matter

Once outcomes are labeled, two derived metrics tell you almost everything:

1. Silent-failure ratio — (budget_exhausted + gave_up + looped + wrong) / total, i.e. failures that didn’t throw, over all runs. This is the number your exception counter was hiding. Track it as your true failure rate. If it’s an order of magnitude above your exception rate — and it usually is at first — that gap is your observability debt.

2. Cost of failure — tokens (and dollars) spent on runs that ended in anything but success. A wrong run that took forty steps cost you a full run’s tokens and whatever the bad output does downstream. Attribute spend to outcome and you’ll often find a large slice of your bill is being burned by a small slice of runs failing expensively — the same shape as the $200 incident, just spread thin enough that no single night sets off an alarm.

The one-line version

If your agent monitoring only counts exceptions, you’re measuring the failures that were kind enough to crash. The ones that cost you are silent: they exhaust a budget, give up, loop, or return a confident wrong answer with exit code zero. Make every run end in a labeled outcome, add a progress detector and post-hoc sampling, and track the silent-failure ratio as your real failure rate. You can’t fix a failure mode you’ve never named — and the whole reason agents feel unreliable in production is that most teams are naming exactly one of them.