Skip to main content
Pollack AI Lab
Agent Infrastructure · Experiments
Helping you build agents that demonstrably work. A growing ecosystem of projects — usable together or individually — for composing, evaluating, and understanding agent behavior. Experiments that prove the approach works, with published data and reproducible results. By Mark Pollack, creator of Spring AI.

​
What’s New

Agent Client 0.19.0
JSONL trace file support — durable per-invocation traces with tool calls, thinking blocks, and result metrics.
May 2026 · Details →
Agent Workflow 0.8.0
Trace file capture through the workflow journal. Every AI step’s tool calls are now traceable via JSONL files.
May 2026 · Details →
ACP Java SDK 0.12.0
7 new protocol methods: session/list, close, resume (stable) + elicitation, fork, config options (unstable).
May 2026 · Details →
AgentWorks BOM 1.2.0
Coordinated versions: agent-client 0.19.0, agent-workflow 0.8.0, agent-journal 1.2.0, ACP 0.12.0.
May 2026 · Details →

​
Core Projects

ACP Java SDK

Agent Client Protocol — build and consume agents for Zed, JetBrains, VS Code

Agent Workflow

Multi-step agentic pipelines with typed context, quality gates, and portable runtimes

Agent Judge

Evaluation with deterministic, command, and LLM judges

Loopy

Interactive coding agent CLI for Java developers

Agent Client

Claude, Gemini, Codex, Amazon Q as unified Spring services

Agent Journal

Behavioral trace capture for analysis and observability

Agent Bench

Benchmarking on real enterprise Java tasks

​
Latest Experiments

Code Coverage v2 — Skills vs Knowledge Bases
Do structured skills outperform flat file injection? 7 variants, N=3
Mar 2026 · In Progress · Details →
Code Coverage v1 — Knowledge Injection Baseline
9 variants testing progressive knowledge injection on Spring Boot test generation
Feb 2026 · Complete · Details →