Val Bercovici, Chief AI Officer at WEKA, breaks down agentic AI, AI infrastructure, and why DeepSeek changed the economics of artificial intelligence. If you're building in AI, the bottleneck isn't compute anymore — it's memory.

🤝 Sponsored by ⁠⁠https://TAP4SRV.com⁠⁠ - Quickly find local Service Providers on WhatsApp! 🤝

In this episode, Val explains why AI is not a bubble despite the capital flood, how agentic systems differ from chatbots, and why KV cache has become the silent bottleneck killing AI performance at scale. The conversation spans AI memory architecture, cybersecurity AI agents, AI factories, token economics, model routing, and the infrastructure race between hyperscalers and challengers like DeepSeek.

Val also shares contrarian takes on government regulation, why software development is about to be unrecognizable, and the one shift every founder needs to make: execution is getting cheaper, but imagination is becoming the scarcest resource in AI.

🔔 Listen and Subscribe

• YouTube: https://www.youtube.com/@TheGeneralistsPodcast

• Spotify: https://open.spotify.com/show/2apfGY403NGr1EyTNkJfG4

• Apple Podcasts: https://podcasts.apple.com/sg/podcast/the-generalists

📖 What You’ll Learn

• Why Val believes AI is not a bubble despite unprecedented capital spending

• How modern AI agents differ from traditional chatbot experiences

• What KV cache is and why memory has become AI’s biggest bottleneck

• Why DeepSeek has earned so much respect from AI researchers and engineers

• How model routing is changing the economics of AI applications

• Why token costs are falling while overall AI demand continues to explode

• How cybersecurity teams should prepare for AI-powered attacks

• What AI factories and token warehousing mean in practice

• Why memory infrastructure may become more important than raw compute power

• How Claude Code, Codex, Hermes, and other coding agents compare

• Why guardrails are becoming critical for autonomous AI systems

• How governments should think about AI regulation and governance

• Why AI is removing tasks rather than entire jobs

• Why founders should focus on ideas as execution becomes cheaper

• Why the next few years of AI progress may be impossible to predict

👤👤 Hosts

• Michael Smith Jr.: https://www.linkedin.com/in/smittysgp/

• Raz Kotler: https://www.linkedin.com/in/razkotler/

👤 Guest

• Val Bercovici: https://www.linkedin.com/in/valentinbercovici/

🔗 Links and Resources

• WEKA: https://www.weka.io/

• OpenRouter: https://openrouter.ai/

• SemiAnalysis: https://semianalysis.com/

• TBPN: https://www.tbpn.com/

• DeepSeek: https://www.deepseek.com/

• Anthropic: https://www.anthropic.com/

• OpenAI: https://openai.com/

• Claude Code: https://www.anthropic.com/claude-code

• Codex: https://openai.com/codex/

• NVIDIA: https://www.nvidia.com/

• Oracle Cloud Infrastructure: https://www.oracle.com/cloud/