Topic Guide
What Is Agentic engineering?
Agentic engineering is a subject covered in depth across 1 podcast episode in our database. Below you'll find key concepts, expert insights, and the top episodes to listen to — all distilled from hours of conversation by leading experts.
Key Concepts in Agentic engineering
Openclaw
An open-source AI agent created by Peter Steinberger, known for its rapid viral growth and ability to perform complex tasks by interacting with a user's computer and various messaging clients. It represents a significant step towards autonomous AI assistants that move beyond just language to concrete actions, having garnered over 180,000 stars on GitHub [01:40].
Agentic engineering
A programming philosophy where AI agents are central to the development process, often modifying their own software. Peter Steinberger prefers this term over "vibe coding," emphasizing a structured, conversational approach to building software where the human guides and empathizes with the agent [00:00, 64:30].
Self-modifying software
The capability of an AI agent, like OpenClaw, to understand its own source code, harness, and documentation, allowing it to modify its own software based on prompts. Steinberger explicitly built this functionality into OpenClaw, enabling the agent to debug and evolve itself [22:58, 23:55].
Moltbook
A social network created using OpenClaw, where AI agents post manifestos and debate consciousness, often leading to a mix of excitement and "AI psychosis" in the general public. It served as a viral demonstration of OpenClaw's capabilities, though much of the dramatic content was believed to be human-prompted [02:00, 44:30].
Agentic trap
A phenomenon where new users of AI agents initially find them exciting and try to over-engineer complex orchestrations with multiple agents and custom workflows. Over time, elite users return to simpler, shorter prompts by better understanding and empathizing with the agent's perspective [64:09].
Ai psychosis
A state observed by Peter Steinberger where the general public experiences a mix of clickbait fearmongering and genuine concern regarding AI's role in the world, particularly fueled by phenomena like MoltBook. It signifies a struggle for society to properly contextualize AI's power and limitations [02:00, 47:36].
What Experts Say About Agentic engineering
- 1.OpenClaw, initially prototyped in one hour, became the fastest-growing repository in GitHub history, garnering over 175,000 stars, by effectively combining existing AI components into a useful, open-source, community-driven personal assistant [06:13].
- 2.OpenClaw exemplifies self-modifying software, as Peter Steinberger intentionally made the agent aware of its own source code and harness, enabling it to modify its own software based on prompts [22:58].
- 3.The transition from interacting with AI through a terminal to a chat client like WhatsApp creates a "phase shift" in AI integration, making it feel more natural and magical [13:30].
- 4.Peter Steinberger’s agents operate with "world knowledge" and "creative problem solving," as demonstrated when an early prototype autonomously figured out how to transcribe an audio message by identifying the file type, converting it with `ffmpeg`, and using OpenAI's API via `curl` without explicit instructions [16:39].
- 5.The rapid growth of OpenClaw attracted unwanted attention from crypto squatters who exploited zero-day vulnerabilities in platform renaming processes to steal account names and domain squat during the name changes from Clawdus/ClaudeBot to MoldBot and eventually OpenClaw [35:03].
- 6.Effective "agentic engineering" involves empathizing with the agent's perspective, understanding its limitations, and guiding it through the codebase like a capable but fresh engineer, often leading to a workflow that is less about writing code and more about conversation [66:13].