AI Dev 26 x SF | Paul Everitt: The Shift to Agentic Engineering

2026-05-22 · Source: DeepLearningAI · Field: Technology & Digital — Software Development & Engineering, Artificial Intelligence & Machine Learning · Depth: Intermediate, extended

Summary

Paul Everitt of JetBrains presented on "The Shift to Agentic Engineering," addressing critical challenges in current AI adoption within software development. He highlighted issues such as limited organizational value from AI, with productivity gains around 10% rather than 10x, and significant quality concerns including a 50% defect rate. Everitt also noted rising AI token costs, low developer trust (only 3% confidence in AI accuracy), negative impacts on engineer mental health, and governance complexities, particularly in Europe where 84% express sovereignty concerns. He argued that the industry's response, often involving layoffs and "AI washing," misses the opportunity to innovate. The proposed solution, Agentic Engineering, emphasizes building systems that build other systems, augmenting human capabilities, and returning to core engineering principles like robust systems design. This approach involves practical elements such as spec-driven development, rigorous evaluations, harness engineering, specialized tooling, and red-green testing.

Key takeaway

For Directors of AI/ML and engineering leads navigating AI integration, you must shift your focus from merely automating existing tasks to embracing agentic engineering. This means investing in systems that augment your engineers, not replace them, by building "the thing that builds the thing." Prioritize robust systems design, implement rigorous evaluation frameworks for agent-generated code, and foster a culture that addresses fear of obsolescence, steering your organization towards true innovation rather than just profit margin squeezing.

Key insights

Agentic Engineering shifts focus to building systems that augment humans and master systems design, addressing AI's current organizational value gap.

Principles

• Coding is a small part of software engineering.
• Augment humans, don't eliminate them.
• Master systems design and first principles.

Method

Agentic Engineering involves spec-driven development, rigorous evals, harness engineering, specialized tooling, red-green testing, and modular codebases to build systems that construct other systems, augmenting human engineers.

In practice

• Implement spec-driven development for agent steering.
• Prioritize evals to assess agent-generated code quality.
• Reorganize code for modularity with sub-agents.

Topics

• Agentic Engineering
• AI Productivity
• Systems Design
• Software Quality
• AI Governance
• Developer Trust

Best for: AI Architect, AI Engineer, Software Engineer, Director of AI/ML

Related on AIssential

• I Stopped Coding for 30 Days — AI Agents Built Everything Instead — AI agents can autonomously handle a broad spectrum of software engineering tasks, fundamentally altering development paradigms.
• The Pragmatic Shift: How to Actually Become an AI Engineer in 2026 — The AI Engineer role in 2026 prioritizes robust integration and orchestration of existing models for production over foundational model training.
• Fragments: April 2 — AI's integration reshapes software development, shifting focus from code generation to system health, human cognition, and verification.
• An AI state of the union: We’ve passed the inflection point, dark factories are coming, and automation timelines | Simon Willison — AI has fundamentally reshaped software engineering, making code generation cheap and amplifying experienced engineers' capabilities.
• Fragments: April 29 — Agentic engineering shifts focus from building fast to verifying fast, emphasizing automated checks and clear definitions.
• Is developer experience dead? — Agentic AI shifts developer experience from code generation to verification, introducing new cognitive loads and requiring a DevEx paradigm shift.

Open in AIssential →

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