My Agentic Workbench

2026-07-07 · Source: LLM on Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Advanced, long

Summary

An "Agentic Workbench" developed over six months enables building software primarily with AI agents like Claude 4.5, minimizing manual code writing. This workbench comprises small, specialized open-source tools such as dossier for project memory, ship for task execution, rooms for isolated command execution in Firecracker microVMs, huddle for agent communication, and sense for output grading. The author positions the large language model (LLM) as the orchestrator, with tools designed to support its operations. The workflow involves planning tasks in dossier, preparing batches, and driving execution through ship, with agents handling code generation and review. The system integrates agent-based testing and a "review-coordinator" agent to consolidate feedback from multiple AI reviewers (Copilot, Codex, Claude, Cursor's Bugbot), aiming to reduce human intervention. A core principle is "embracing the slop," tolerating minor stylistic differences in agent-generated code if functionality and reliability are met.

Key takeaway

For AI Engineers building agentic systems, you should shift your focus from writing code to designing robust orchestration and validation layers. Embrace small, specialized tools that empower the LLM as the primary orchestrator, rather than trying to out-build it. Implement agent-on-agent testing and review coordination to ensure confidence in automated outputs, allowing you to tolerate minor stylistic "slop" and accelerate development velocity. Your role evolves to confidence-checking and system design.

Key insights

Building with AI agents requires small, sharp tools and a system-centric approach to orchestration, testing, and review.

Principles

• LLMs orchestrate; build small, sharp tools for them.
• Automate repetitive agent actions into code, not prose.
• Robust agentic development demands agents checking agents.

Method

Plan tasks in dossier, prep batches, drive execution via ship, review with multiple agents and a /review-coordinator, then recap. Automate repetitive agent actions into code.

In practice

• Write house style into CLAUDE.md and enforce with lint.
• Use dossier for project memory and task specifications.

Topics

• Agentic Development
• LLM Orchestration
• Automated Code Review
• MicroVMs
• Software Engineering Workflow
• AI Testing

Code references

• itsHabib/skills
• itsHabib/cortex
• itsHabib/orchestra
• itsHabib/dossier
• itsHabib/ship

Best for: AI Engineer, Machine Learning Engineer, Software Engineer

Related on AIssential

• The Beginner’s Guide to Agentic AI: Building Your First AI Agent as an SDET — AI agents are an engineering pattern integrating LLMs with tools and feedback loops for autonomous goal achievement.
• Agentic System Design Interview Series #1: Design a Coding Agent — Agentic system design requires building deterministic frameworks around non-deterministic LLM cores, emphasizing feedback loops and robust verification.
• A Year Of Vibes — Agentic coding tools are fundamentally reshaping software engineering workflows, challenging traditional practices and human-machine relationships.
• AI Agent Tool Design: What Works and What Doesn’t — Effective AI agent performance hinges on robust tool design, not just advanced model capabilities.
• You Probably Don’t Need an Agent Framework — Explicit LLM workflows offer more reliability and control than autonomous agents for many applications.
• The Accidental Orchestrator — Effective AI-driven development requires a structured approach and human orchestration of AI agents, not just tool use.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by LLM on Medium.