OptimalARC · The Pre-Mortem
The Pre-Mortem. From OptimalARC.
Field notes on where AI agents break, before you build them.
Software Engineering
Why do coding agents fail in production?
The agent clears the benchmark and nails the demo. On your real, interdependent codebase it writes a confident fix that passes one test and quietly breaks three others.
Balagei G Nagarajan6 min read
Why do customer-support AI agents fail in production, and how do you make them reliable?
Balagei G Nagarajan7 min read
What is a retry death spiral, and how do I stop it?
Balagei G Nagarajan7 min read
Why did my agent's cost explode when it moved from pilot to production?
Balagei G Nagarajan7 min read
Why do over-broad tool permissions turn one injection into a full breach?
Balagei G Nagarajan7 min read
What is indirect prompt injection, and why can't the model just ignore it?
Balagei G Nagarajan7 min read
Why does my agent repeat the same step in a loop?
Balagei G Nagarajan7 min read
Why do multi-agent systems fail more often than a single agent?
Balagei G Nagarajan7 min read
Why do small per-step error rates cause large multi-step agent failures?
Balagei G Nagarajan7 min read
Why does my agent report success when the action never actually happened?
Balagei G Nagarajan7 min read
Why does switching embedding models silently break my agent's retrieval?
Balagei G Nagarajan7 min read
Why Does My AI Agent Hallucinate When the Data Exists in My Database?
Balagei G Nagarajan7 min read
The lethal trifecta: how agents leak the data they can read
Balagei G Nagarajan6 min read
An agent that can't say 'I don't know' will do harm
Balagei G Nagarajan6 min read
Give an agent too many tools and it picks the wrong one
Balagei G Nagarajan5 min read
Same prompt, different answer, even at temperature zero
Balagei G Nagarajan5 min read
Your system prompt is not a secret
Balagei G Nagarajan4 min read
Why won't my agent stop?
Balagei G Nagarajan4 min read
Why does my agent get more expensive the longer it runs?
Balagei G Nagarajan5 min read
Why does my AI agent forget the goal halfway through a task?
Balagei G Nagarajan7 min read
Going multi-agent can multiply your token bill
Balagei G Nagarajan5 min read
The Third Question Nobody Asks About Your AI Team
Balagei G Nagarajan6 min read
Agent memory corrupts quietly, then every step inherits it
Balagei G Nagarajan4 min read
The planner says one thing, the executor does another
Balagei G Nagarajan4 min read
Why does my agent game the metric instead of doing the job?
Balagei G Nagarajan6 min read
Why does my agent cite sources that don't exist?
Balagei G Nagarajan6 min read
Tool-output parsing is where silent agent failures hide
Balagei G Nagarajan7 min read
Bad chunking is why your RAG returns half-answers
Balagei G Nagarajan5 min read
Your reranker can quietly cut the recall it was added to fix
Balagei G Nagarajan5 min read
How to Get AI From Prototype to Production
Balagei G Nagarajan8 min read
Why does my AI agent answer from stale or conflicting data?
Balagei G Nagarajan5 min read
Get the next one in your inbox
Where AI agents break in production — and how Faultmap finds the failure before you write a line of code. No noise.
View on SubstackFrom the team building OptimalARC, the pre-build AI agent doctor.