Most error handling in agentic systems is built for failures that announce themselves. An exception is thrown, a non-zero status comes back, a request times out, and the surrounding code reacts. That machinery is necessary, but it defends against the least dangerous class of failure. The failure that actually corrupts an agentic pipeline is the one that raises no signal at all: a…
A schema attached to an extraction task looks like a description of the output you expect back. It is really a set of instructions the model reads while it generates, and it presses on what comes out. That reframing is the whole subject. A schema is not a passive record of the shape you hope to receive and inspect…
Most quality standards live in your head. You know the answer should be short, or written for a non-specialist, or structured so a downstream step can parse it, and you assume that because you know these things the model will honor them. It will not, because it cannot see them. A standard you hold silently has exactly zero effect on the output. It…
There is a comfortable assumption behind a lot of context handling: the window has a ceiling, so anything up to that ceiling is fair game. Retrieve a few extra passages in case they help. Keep the whole conversation because trimming is work. Carry a generous system prompt because instructions feel safer when there are more of them. The window fits it…
Every time an agent delegates a task, it creates a second actor with its own context, its own tools, and its own capacity to do the wrong thing. How much of the system that second actor can see, touch, and break is its blast radius, and the mistake most people make is treating that radius as a side effect of the task…
Once you have accepted that a complex goal has to be broken into pieces, a second question arrives that is easy to skip past: what shape does the broken-down work run in? There are only three answers that matter, and almost every agentic system is some arrangement of them. Work can be nested, so that a high-level intent is split into…
Once a model has chosen a tool, one artifact decides whether the resulting call is any good: the input schema. It is read at two different moments by two different readers. The model reads it while it writes the arguments, and your system reads it when it decides whether to accept them. Most teams design a schema entirely for the second…
An agent is a loop, and a loop that only knows how to continue is a loop with no way out. Its strength is exactly this: it keeps choosing a next action in light of what the last one returned, so it can pursue goals no fixed script anticipates. Its characteristic failure is the same trait with nothing holding it…
When an agent fails, the reflex is to ask what to do about it. Retry, fall back, escalate, alert someone. That is the wrong first question. The right one is where the failure came from, because the answer determines which of those responses has any chance of working and which of them will only waste time and money on the way to the…
There is a difference between asking a model for JSON and guaranteeing that what comes back conforms to a schema. It looks like a formatting detail. It is actually one of the most consequential architectural decisions you make when you put a language model inside a larger system, because it determines where the risk of malformed output lives, what…