Adaline launches Adaline 2.0, an automated optimization layer that tests and upgrades AI agents using production execution traces

@arshdilbagi This video just motivated me to become the worlds first bajillionaire

Adaline just launched a self-improvement layer for AI agents that turns messy production traces into fresh evals, synthetic edge cases, and better agent candidates for humans to approve.

I expected it to be a regular trace viewer, but it is reading my production traffic and building evals I would never have considered.

It reads production traffic and user feedback, then clusters the mess into recognizable agent behaviours rather than asking a human to manually inspect every strange conversation.

Introducing Adaline 2.0 - The Agent Self-Improvement Layer

Adaline turns Traces into Behaviors, Behaviors surface Issues, Issues become auto-generated Evals + Data, Adaline then generates new agent candidates and tests them.

You review the winners and ship!

Agents now generate more data than humans can comprehend.

Every message. Every tool call. Every retrieval.

Observability shows you the traces. Adaline reads production traces and clusters them into Behaviors.

Actual recurring patterns in how your agent behaves in the wild.

The loop is simple:

Traces → Behaviors Behaviors → Issues Issues → Evals + Data Evals + Data → Better agent candidates Best candidate → Human approval → Ship

Ship once. Improve forever.

Observability was read-only. Agents need read-write infrastructure.

Adaline 2.0 gives every AI team a way to understand, test, and improve their agents continuously.

Self-improve your agents now: http://adaline.ai

@arshdilbagi Really really cool! Congrats :)

Then those Behaviors become coverage.

Adaline generates evals and data.

So every production failure becomes an offline eval with data and part of the agent’s memory.

Now the agent can improve.

Adaline generates multiple new agent candidates, tests them against the evals + data, and shows you which ones actually perform better.

Not just a new prompt.

A tested candidate.

@arshdilbagi “the bottleneck is you”. felt that one. 🥲

@arshdilbagi Bullish

@arshdilbagi AI agents finally getting the same feedback loop humans use to improve.

@arshdilbagi 🔥🔥🔥

@arshdilbagi Man this was actually needed tbh

@kimmonismus Finally, a feedback loop that doesn’t ignore the human

@arshdilbagi thats so fire

@arshdilbagi Nice! Congrats Arsh

@arshdilbagi Was I the bottleneck the whole time?👀

@arshdilbagi

@arshdilbagi hell yeah