From tools to agents
The conversation around AI is no longer only about clever prompts and clever code. It’s expanding into the era of agentic AI — systems that can interpret goals, plan a course of action, and act with a degree of autonomy. Garrett Rose isn’t just tinkering with code; he’s re-learning the craft of guidance, the subtle art of shaping intent without crowding the AI with step-by-step instructions. Industry chatter mirrors this shift, with technology outlets weighing what it means when machines can take on goals rather than merely execute them. Intel Puts Agentic AI to Work with Xeon 6+, Networking, and AI Systems as a concrete example of agentic AI entering production environments.
What counts as guidance in practice
Guiding an agentic system isn’t about micromanaging every action; it’s about shaping objectives, constraints, and feedback loops so that the AI can pursue goals safely and effectively. That requires a blend of prompt design, safety constraints, and real-time monitoring. The discussion also touches on reward modeling, alignment, and how to intervene when an agent begins to stray. The video diary here frames guidance as a dynamic discipline — a middle ground between hands-off autonomy and heavy-handed control. Public discourse around the topic exists as well, with technology outlets debating whether agentic AI could become a foundational infrastructure or simply a new layer of complexity in an already risky landscape. Can Agentic AI become the internet’s next gateway?
Stake, risk, and the human role
Agentic AI changes the calculus for safety and accountability. If a system can set goals, it also inherits the responsibility to avoid harm and to operate within the boundaries humans set. That tension sits at the heart of Rose’s re-learning: how to enable agency without surrendering judgment. The broader industry is watching: analysts and practitioners are weighing how far such autonomy should go and what guardrails are appropriate. Industry coverage suggests this is not a niche debate but a strategic inflection point for software delivery and AI governance. Rewiring software delivery for the agentic era
Who holds the reins? The players and the policy questions
Garrett Rose frames the discussion through a developer’s lens — a reminder that agentic AI isn’t only a technical challenge but a human one. If engineers aim to design systems that can plan and act, they must also confront questions about oversight, transparency, and scope. The conversation sits at the intersection of engineering, ethics, and policy, with real-world deployments (like Intel’s ventures into agentic AI) cited in technical coverage as proof that the era is already underway. Intel Puts Agentic AI to Work with Xeon 6+, Networking, and AI Systems.
Sources & further reading
- TechPowerUp — Shows industry deployment examples of agentic AI, grounding the topic in real-world use.
- Investing.com — Represents ongoing discussions about agentic AI’s potential reach and gateway status.
- McKinsey & Company — Frames the agentic era in software delivery and strategic implications.
- Garrett Rose (YouTube video description) — Provides the subject’s own framing and context for the discussion.
Definitions
- Agentic AI
- AI systems capable of interpreting goals, planning, and taking actions to pursue those goals with a degree of autonomy.
- Guidance of AI systems
- The methods and practices used to shape how an AI behaves, including objectives, constraints, feedback, and oversight.
- AI safety
- Principles and practices to prevent harm, ensure reliability, and keep AI aligned with human values.
- Agentic era
- A period in which AI systems increasingly perform autonomous tasks, moving beyond tool-like behavior.