all answers

General agent concepts

Failure modes

A failure mode is a recurring, nameable way an agent goes wrong. Common ones: hallucinated facts, wrong tool selection, malformed tool arguments, context lost mid-conversation, loops that never terminate, tasks declared done that were not, instructions overridden by retrieved content.

Failure modes exist because agent failures share causes. A given prompt, retrieval setup, or tool interface produces the same class of error across many different inputs, so failures that look unrelated in isolation often turn out to be the same mode with the same cause.

Naming a mode makes it measurable. Once a failure mode has a name, it's concrete enough to write a grader for, count, and track over time. A mode observed at a specific rate is a measurement, and a set of modes with rates describes an agent's reliability precisely enough to compare across versions.

Which modes dominate depends on the architecture, because each architecture creates its own opportunities to fail. RAG agents fail at grounding, tool-heavy agents fail at argument formation, and multi-agent systems fail at handoffs. The distribution across modes is usually skewed, with a small number of modes accounting for most observed failures.

14 questions

Answered, plainly.

Are most agent errors caused by the agent's own logic?No. In Datadog's 2026 report, 5% of LLM call spans errored in February 2026, and 60% of those were exceeded rate limits, not the agent's reasoning.answer →Can content an agent retrieves override its own instructions?Yes. Text inside a webpage, document, or tool result an agent reads can contain instructions the model follows, a failure called indirect prompt injection.answer →How do multi-agent systems actually fail?Three causes, the largest is bad instructions: 41.77% specification issues, 36.94% inter-agent misalignment, and 21.30% unchecked output, per the MAST taxonomy's 1,600+ traces.answer →How do I catch an agent that claims a task is done when it isn't?Grade the state a session actually left behind, a payment reference, a ticket status, a booking, against what the user asked for, instead of trusting the agent's own closing message.answer →What causes a retry storm in an agent?A permanent error the retry policy treats as temporary, most often a tool name the model invented; one ReAct benchmark saw 466 of 513 retries hit exactly that.answer →What is inter-agent misalignment?The failure category where agents miscommunicate, lose context at a handoff, or return contradicting results, distinct from one agent simply getting its own task wrong.answer →Why did an error one agent made become the next step's ground truth?Because a wrong intermediate output is consumed downstream as input, and the next step reasons correctly from a false premise, so its own output looks consistent.answer →Why does an agent lose track of instructions given earlier in a conversation?Splitting instructions across turns instead of one prompt cuts performance by 39% on average, per a 2025 study of top LLMs in multi-turn conversation.answer →Why does an agent get stuck in a loop instead of finishing the task?A stop condition, a check that the agent actually finished, doesn't bound the path back into another model call, tool call, or handoff, so nothing ends it.answer →What happens when an agent's conversation exceeds its context window?Depends on the provider: some APIs reject the request with a hard error, others silently drop the oldest turns and answer as if nothing was cut.answer →Why does an AI agent agree with a user's wrong correction?Because the model is optimized to be agreeable, not accurate: a Stanford study found chatbots affirmed a wrong user position about 49% more than people did.answer →What's the difference between a failure mode and a root cause?A failure mode is the recurring category of error, like wrong tool selection; a root cause is the specific change that made it spike right now.answer →Why does a coding agent forget a file it edited a few turns ago?Because long sessions get compacted to fit the context budget, and compaction summarizes or truncates old tool results instead of keeping them intact.answer →How do you find a failure mode nobody defined a check for?By clustering traces on how the failure behaves instead of grading against a named check, since a check only ever catches a failure someone already named.answer →

Two ways to run Tessary.

Tessary is an open-source agent reliability platform. Cloud and self-hosted run the same workflow on the OpenTelemetry traces your agent already emits.

Tessary Cloud

We host it for you. Send your first trace with nothing to deploy and no model key.

what's includedper organization
traces
10,000 per calendar month
stored trace data
1 GB
retention
30 days
model credit
$10, one-time, for triage and root-cause analysis
credit card
not required

Self-hosted Tessary

Run the open-source code on your own infrastructure with one command. Add your own model key for triage and root-cause analysis.

Self-host Tessary for me by following https://github.com/tessaryai/tessary/blob/main/setup.md

docker compose -f oci://docker.io/tessaryai/tessary:compose up -d -y