"My agentic engineering workflow" | George Pickett (MTS @ Parallel)

2026-06-30 · Source: Greg Kamradt · Field: Technology & Digital — Software Development & Engineering, Artificial Intelligence & Machine Learning · Depth: Advanced, extended

Summary

George Pickett, a software engineer at Parallel, details his evolving agentic engineering workflow, primarily using Codeex. Initially, his process involved an "exec plan" based on an OpenAI blog post, where agents create detailed planning documents for long-running tasks, potentially for 25 hours. He enhanced this with "Grill Me," an agent-led interview to clarify intent, generating a "decisions.mmd" file for durable memory. Pickett then integrated "Goalcraft" to create long-running goals with context, boundaries, and verification. His latest iteration, "Grillcraft," combines grilling, planning, and goal activation, ensuring durable artifacts like decisions.mmd and exec plan are linked to the goal for structured progression. He also employs "review recent work" and adversarial code review using different LLMs like Claude and Codto for robust bug detection.

Key takeaway

For AI engineers building complex, long-running agentic systems, you should prioritize upfront intent clarification and the creation of durable planning artifacts. Implement an agent-led "grilling" phase to thoroughly define project scope and decisions, then link these to an "exec plan" and a long-running goal. This structured approach, combined with adversarial code review using diverse LLMs, will significantly reduce mid-session steering and improve the autonomy and reliability of your agents.

Key insights

Clarifying intent and creating durable artifacts are crucial for effective, long-running AI agent workflows.

Principles

• Agents need explicit planning documents.
• Clarify intent via agent-led questioning.
• Use adversarial review for robust checks.

Method

The workflow progresses from initial instruction to agent-led grilling for intent clarification, generating durable decision and execution plans, then activating a long-running goal with linked artifacts for structured, autonomous execution.

In practice

• Implement "Grill Me" for project scoping.
• Chain multiple review skills for thoroughness.
• Integrate external LLMs for adversarial review.

Topics

• Agentic Workflows
• AI-Assisted Coding
• Codeex
• LLM Orchestration
• Software Engineering Principles
• Code Review Automation
• Goal-Oriented Agents

Best for: Software Engineer, AI Engineer

Related on AIssential

• A Year Of Vibes — Agentic coding tools are fundamentally reshaping software engineering workflows, challenging traditional practices and human-machine relationships.
• Long-running Agents — Long-running agents overcome AI limitations by externalizing state, enabling sustained, multi-session progress on complex tasks.
• Bridging intent and execution in agentic systems — Minimizing the intent-execution gap in agent harnesses is crucial for state-of-the-art LLM performance.
• AIE Miami Day 2 ft. Cerebras, OpenCode, Cursor, Arize AI, and more! — Successful AI agent adoption and performance hinge on effective context management, optimized inference, and flexible tooling.
• You Probably Don’t Need an Agent Framework — Explicit LLM workflows offer more reliability and control than autonomous agents for many applications.
• Agentic Engineering — Effective agentic engineering requires structured workflows to plan, steer, and verify AI-generated code.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Greg Kamradt.