Levels of Agentic Engineering

4 days ago

AI's coding ability is outpacing our ability to wield it effectively, creating a gap between capability and practice.
The gap closes in 8 levels, each representing a significant leap in output and productivity.
The multiplayer effect means your output depends on your teammates' levels, making it beneficial to elevate your team's skills.
Levels 1 & 2 focus on tab completion and AI-focused IDEs, with context being a limiting factor.
Level 3 introduces context engineering, optimizing the information density of each token in prompts.
Level 4 is compounding engineering, where lessons from each session are codified to improve future interactions.
Level 5 enhances capability with MCPs and custom skills, allowing LLMs to interact with databases, APIs, and more.
Level 6 involves harness engineering, building environments and feedback loops for agents to work autonomously.
Level 7 shifts to background agents that can plan and execute tasks without constant human oversight.
Level 8 explores autonomous agent teams, where agents coordinate directly without a central orchestrator, though this is still experimental.
The future may include voice-to-voice interaction with coding agents, making software development more iterative and faster.

Hasty Briefsbeta