Agentic Engineering and the Lost Art of Verification

2026-07-17 · Source: Vanishing Gradients · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Data Science & Analytics · Depth: Advanced, extended

Summary

Wes McKinney, Jeremiah Lowen, and Randy Olson discussed their advanced AI agent workflows and tools in a "Show Us Your Agent Skills" episode. Wes McKinney detailed his "agentic engineering" stack, including RoboRev for automated code review (using GPT 5.5 to review code 4-5 times), Agents View for session analytics, Middleman for local GitHub dashboard management, and Kata for issue tracking, processing over a billion tokens daily. Jeremiah Lowen presented his "second brain" approach, using OpenClaw to store voice memos and meeting transcripts, alongside his "explain" skill for context-aware agent communication, and Prefab, a Python DSL for generative UIs. Randy Olson demonstrated a data visualization skill that encodes Edward Tufty's principles, enabling agents to research data, generate chart variants, and refine visualizations through a "verifier loop" that acts as an LLM judge. The discussion highlighted the shift from writing and reading code to agent-driven verification and custom software development, while acknowledging challenges like agent reliability and the need for precise instruction.

Key takeaway

For AI Engineers and MLOps teams aiming to scale development and ensure code quality, integrating agentic workflows is crucial. You should implement automated code review systems like RoboRev, leveraging advanced LLMs (e.g., GPT 5.5) for rigorous, multi-pass verification of agent-generated code. Additionally, consider encoding specific design principles or human judgments into agent skills and employing "verifier loops" to iteratively refine outputs, significantly reducing manual oversight and accelerating project delivery while maintaining high standards.

Key insights

Advanced AI agents enable scalable software development, automated verification, and personalized knowledge management, shifting focus from manual coding to oversight.

Principles

• Automate code review with agents for thorough, multi-pass verification.
• Treat agent memory as a "second brain" for continuous context and knowledge accumulation.
• Encode human judgment and design principles into agent skills for consistent output.

Method

Implement a "verifier loop" where agents generate outputs (e.g., data visualizations), an LLM acts as a judge based on encoded criteria, and provides feedback for iterative refinement until quality standards are met.

In practice

• Use a daemon like RoboRev for continuous, asynchronous code review on agent-generated commits.
• Develop an "explain" skill to guide agents in providing concise, context-aware summaries of their actions.
• Create Python DSLs (e.g., Prefab) to stream generative, interactive UIs directly from agent processes.

Topics

• Agentic Engineering
• AI Agents
• Automated Code Review
• Data Visualization
• LLM Evals
• Custom Software Generation
• Workflow Automation

Best for: AI Architect, AI Engineer, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• A Year Of Vibes — Agentic coding tools are fundamentally reshaping software engineering workflows, challenging traditional practices and human-machine relationships.
• Only the best are using them... — AI coding loops automate development by defining a verifiable end goal, allowing agents to self-iterate until completion.
• Self-Improving AI Coding Agents Through Accumulated Behavioral Rules: A Closed-Loop Framework — Codifying human review feedback into persistent behavioral rules enables AI coding agents to self-improve across sessions.
• My Agentic Workbench — Building with AI agents requires small, sharp tools and a system-centric approach to orchestration, testing, and review.
• The End of Code Review: Coding Agents Supersede Human Inspection — Coding agents now supersede human code review by fulfilling its goals more efficiently and scalably.
• Exploring LLM-based Verilog Code Generation with Data-Efficient Fine-Tuning and Testbench Automation — Multi-agent LLMs can automate testbench generation, creating high-quality data for efficient Verilog code generation.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Vanishing Gradients.