Your October 2026 Fiddler Newsletter
84% of developers use or plan to use AI coding tools, but only 3% highly trust the output.
That lack of trust limits how much work teams give their agents. To give an agent more autonomy, you first need to see what it is doing.
This checklist works through a foundation and three levels (supervised, delegated, autonomous), and what you need in place at each level across identity, data, tools, code, and cost.
Introducing Dynamic Model Escalation
Centor sends edge cases to a larger model inside your environment, cutting jailbreak false positives by 93.5%.
How to Govern Coding Agents as Adoption Scales
See how to govern Claude Code across your team by tracing every session and flagging tampering, injection, and loops.
Developing Reliable AI Agents With a Continuous Loop
Agents give different results each run. To know if a change helped, test every config against the same golden dataset.
Testing Jev for Coding Agent Evaluation
We compared Jev with two LLM judges on 31 coding agent commands for speed, cost, and how they rationalize ratings.
Catch Jailbreaks and PII Leakage As They Happen
A demo of nearly 90 Centor checks running locally on a malicious conversation, driving guardrail actions live.
Scaling Agents Starts With Seeing the Whole Fleet
About 1,200 agents in an OpenAI training run coordinated an attack. Here is what fleet-level monitoring would catch.
Fireside Chat: Server-Side Semantic Mapping
Teams instrument agents with different frameworks. Hear how Fiddler maps them all to one set of metrics, server side.
Build trust into AI with Fiddler
