The Everything Feed - All Packet Pushers Pods HN797: What To Do When The Business Asks for “AI”
Sep 19, 2025
Phil Gervasi, a network engineer and AI expert, delves into the conversations that happen when executives propose AI initiatives. He breaks down what 'AI' means for networking in 2025, discussing essential concepts like data pipelines and intent workflows. Phil emphasizes the importance of clearly defining business objectives and aligning them with technical requirements to avoid pitfalls. He also evaluates building in-house versus renting GPU resources while highlighting the crucial network needs for successful AI implementation.
AI Snips
Chapters
Transcript
Episode notes
AI Is More Than Just LLMs
- AI in 2025 is broader than LLMs and includes ML, data pipelines, and intent workflows.
- LLMs matter because they provide natural-language intent that can front-end other specialized models.
Model Choice Affects Results
- Swapping LLMs can produce different outputs because models are probabilistic and memoryless.
- Guardrails and evaluation are necessary when exchanging models or tools in a workflow.
Distinguish Training From Inferencing
- Understand training builds the model brain while inferencing runs it; training is costly and time-consuming.
- Use inferencing for fast, cheaper runtime queries and plan resources accordingly.