VIDEO
HIGHLIGHT
Log InSign up
FORGET Loop Engineering. Agentic Engineering is about THIS
SummaryTranscriptChat

FORGET Loop Engineering. Agentic Engineering is about THIS

The Misconception of Loop Engineering

Understanding the Flaws in Loop Engineering

  • The speaker criticizes "loop engineering" as a misleading term that oversimplifies software development, equating it to a rebranding of the traditional software development life cycle.
  • Emphasizes the importance of clarity and simplicity in building valuable software with agents, suggesting that this approach accelerates progress in the AI industry.
  • Proposes focusing on developer workflows within a "software factory" rather than on loop engineering.

Introduction to Agentic Engineering

  • The speaker introduces himself as Dan Eisler, an experienced software engineer with over 15 years in various programming languages and tools.
  • Highlights his contributions to agentic engineering through information products and consistent content creation aimed at helping engineers advance their skills.

Key Components of Value Creation

Actors in Software Development

  • Identifies three main actors: engineers, agents, and code. Mastering their interactions is crucial for effective agentic engineering.
  • Stresses that while everyone discusses agents and engineers' costs, code remains the most reliable component due to its speed and lack of associated token costs.

Basic Developer Workflow

  • Describes a fundamental workflow where an engineer prompts a language model (LLM), reviews results, and iteratively improves outputs using agents.
  • Introduces conditions within workflows that create loops but argues that loop engineering is too simplistic for complex systems.

Enhancing Developer Workflows

Scaling Up Workflows

  • Discusses adding more deterministic code into workflows to improve validation processes through multiple passes back into build agents.
  • Explains how testing becomes integral by feeding results back into build agents until all tests pass before final review.

Importance of Planning and Review

  • Highlights two critical constraints: prompting (planning phase) and reviewing (validation phase), which are essential for successful agentic engineering at scale.

Advanced Techniques in Agentic Engineering

Creating Isolated Environments

  • Suggestion to provide each agent with its own sandbox enhances isolation and allows parallel processing without interference among agents.

Building Complex Systems

  • Advocates for designing AI developer workflows that leverage engineers, agents, and code effectively to maximize impact across organizations.

Responding to Production Crises

Handling Support Issues Efficiently

  • Outlines a scenario where production issues arise; emphasizes having predefined AI developer workflows ready for rapid response during crises.

Utilizing Specialized Agents

  • Introduces the concept of specialized hotfix agents designed specifically for urgent fixes without unnecessary optimizations or delays.

Towards a Software Factory Model

Structuring Workflows Effectively

  • Describes evolving towards a structured system resembling a software factory capable of handling various tasks like chores, bugs, or features efficiently through specialized agent sandboxes.

Continuous Improvement

  • Concludes by emphasizing the need for continuous refinement of these systems to ensure they operate better than individual efforts alone.

What is Agentic Engineering?

Understanding Products and Companies

  • A product within a company, especially outside of big tech, consists of specialized teams that address specific problems for targeted users. This specialization highlights the value of expertise.

The Shift to Agentic Coding

  • Emphasizing the importance of structured AI developer workflows, agentic coding moves away from "vibe coding," which lacks understanding of system mechanics.
  • Agentic engineering involves mastering your system to the extent that you can operate without constant oversight, elevating engineers to a meta-engineering role.

Building Effective AI Developer Workflows (ADWs)

  • Engineers should design ADWs with customers in mind, treating them as integral nodes in the workflow process.
  • Start with simple workflows; after establishing an agent, engage in iterative prompting while monitoring its performance closely.

Separation of Concerns

  • It's crucial to separate code execution from agent operations. Use an SDK for agents and ensure linting occurs independently to maintain clarity and organization.
  • As complexity increases, begin adding nodes to solve real problems while funneling errors back into the build agent for resolution.

Scaling Up Your Workflows

  • Gradually specialize agents by separating contexts such as front-end and back-end tasks. Maintain simplicity initially but prepare for more complex structures as production demands grow.
  • Remember KISS (Keep It Simple Stupid); start with straightforward skill-based workflows before transitioning into more intricate systems.

Key Strategies for Developing ADWs

Hands-On Experience

  • Engineers should personally execute their workflows end-to-end. This includes running tests and observing function executions to understand each component's role thoroughly.

Utilizing Tools Effectively

  • Consider using tools like Mermaid for visualizing workflows. Creating diagrams can help clarify processes and enhance communication among team members.

Balancing Agents and Code

  • Transition from relying solely on agents to incorporating code as production scales up. This balance enhances performance, reliability, and speed while minimizing hallucinations associated with pure agent use.

Final Thoughts on Agentic Engineering

Importance of Classic Engineering Patterns

  • Adhere to established engineering principles such as decoupling components and maintaining single interfaces; these practices become even more critical when scaling ADWs effectively.

Community Engagement

  • Acknowledgment is given to long-time followers who have supported this journey into agentic engineering concepts.

Further Learning Resources

Tactical Agent Coding Program

  • For those interested in deepening their understanding, consider exploring tactical agent coding lessons that break down key concepts step-by-step over eight lessons plus additional upgrades available.

Additional Reading

  • Explore free resources like "Thinking in Threads" which covers similar ideas discussed throughout this sessionā€”available through links provided in descriptions or channels related to this content.
Video description

Loop engineering is a terrible rebrand that's going to hold you back. šŸ”„ Forget it. The engineers pulling ahead of the entire AI industry aren't building loops, they're building AI developer workflows inside their own software factory. Your prompts go in, a specific workflow runs (a combination of code plus agents), and your results come out. That's the whole game. Loops are just one tiny piece of the picture. āœ… MASTER AGENTIC ENGINEERING Tactical Agentic Coding (TAC): https://agenticengineer.com/tactical-agentic-coding?y=VQy50fuxI34 Thinking In Threads: https://agenticengineer.com/thinking-in-threads?y=VQy50fuxI34 šŸŽ„ VIDEO REFERENCES ā€¢ Peter's Loop Tweet: https://x.com/steipete/status/2063697162748260627 ā€¢ Boris's Loop Tweet: https://x.com/bcherny/status/2064426115255730578 ā€¢ Anthropics loop blog post: https://claude.com/blog/getting-started-with-loops ā€¢ Mermaid.js: https://mermaid.live šŸš€ IndyDevDan here. In this video, we break down why loop engineering is the wrong mental model and what agentic engineering is really about: building AI developer workflows inside your software factory. If you understand this concept properly, you'll accelerate far ahead of the industry, because clarity and simplicity of information give you speed and performance in your work. šŸ”„ There are now three actors of value creation: engineers, agents, and code. Knowing when and where to place each of these is the name of the game of agentic coding. Everyone is talking about AI coding agents, but code is the unsung hero here. It's fast, it's reliable, it runs the same way every time, and it costs zero tokens. Loops are just one small slice of this. If we're going to call it loop engineering, we'd also need condition engineering, function engineering, and exception engineering. It's the software development life cycle with AI bolted on, nothing more. šŸ› ļø We scale a single simple workflow all the way up to a full software factory. Start with an engineer prompting an agent (your Claude Code, your Codex, your Pi coding agent) and reviewing the result. Add deterministic code (linters, formatters, type checkers, tests) with pass/fail conditions that route back into your build agent. Then collapse all that validation into a dedicated test agent. This is how you scale your compute to scale your impact: you add compute to add confidence. You and I always show up at the two constraints, planning at the beginning and reviewing at the end, while the system handles everything in between. šŸ’” Git worktrees give your builder and tester agents isolation and parallelism so they don't trip over each other, but git worktrees are a great place to start, not a great place to end. The upgrade is agent sandboxes: give every single agent its own computer to operate in. You can jump into the sandbox to look at the work, run your review, then merge and ship. Agent sandboxes are going to be the majority of computers in the world. āš” We walk through the Kanban queue, where tickets from support, product, and engineering flow into scout agents, plan agents, build agents, and test agents running inside their own sandboxes. This is multi-agent orchestration and agent orchestration in practice. Advanced teams skip translating every ticket into a low-level prompt and kick off the software factory the moment a ticket lands. A factory router agent reads the codebase and picks the right AI developer workflow for the job at the best price, performance, and speed. šŸŒŸ It all adds up to a software factory that can operate your application better than you, your code, or your agents could alone. This is why your effort moves to the agentic layer, not the app layer. The best teams do the meta work, building the system that builds the system. That's the central thesis inside Tactical Agentic Coding, and it's the opposite of vibe coding. Vibe coding is not knowing how your system works. Agentic engineering is knowing your system works so well you don't have to look. At the highest levels of agentic engineering, you're building software factories that execute the right work with the right combination of engineers, agents, and code across your entire organization. This is the greatest leverage point in agentic coding. Template your expertise into your AI developer workflows and you get a repeatable system that delivers consistent results tens, hundreds, and thousands of times. Stay focused and keep building. Dan šŸ“– Chapters 00:00 Forget Loop Engineering 01:20 Who Is IndyDevDan? 03:38 Your 3 Actors of Value Creation 04:43 Your First Ever AI Developer Workflow 06:02 Adding Code to Your ADW 07:35 Scale Your Compute to Scale Your Impact 11:56 The Kanban Queue 15:27 Production Goes Down 17:48 The Software Factory 26:39 How to Build Great AI Developer Workflows 29:02 Do It by Hand First 30:09 Make Sure You're Not Just Using Agents 32:10 Tactical Agentic Coding Pitch #agenticengineering #softwarefactory #aideveloperworkflows