This Day in AI Podcast Claude 4.5 Opus Shocks, The State of AI in 2025, Fara-7B & MCP-UI | EP99.26
245 snips
Nov 28, 2025 Claude 4.5 Opus impresses with speed and pricing, possibly outshining its competitors. The hosts analyze a controversial McKinsey report suggesting AI could automate 57% of jobs. They test Microsoft's Farah 7B, which yields some humorous refusals during local execution. Conversations shift to the potential pitfalls of OpenAI's MCP-UI apps, critiquing their user experience. Finally, they preview innovative demos from Claude, including a cheeky AI-generated diss track that adds a fun twist to the tech talk.
AI Snips
Chapters
Transcript
Episode notes
Opus Is The New Agentic Workhorse
- Claude 4.5 Opus is now the authors' preferred daily model because it balances speed, cost, and coding/tool-calling ability.
- Strong tool-calling and code skills make a model more useful for agentic workflows than raw creativity alone.
Opus API Introduces Effort And Thinking Tokens
- Anthropic switched thinking controls to low/medium/high effort and added token references for ongoing thoughts in the API.
- These changes simplify prompting but broke some existing integrations during rollout.
Constantly Refresh Model Context
- Assume models forget context and proactively remind them of the latest relevant content with each prompt.
- This extra token cost often yields faster, more accurate results and reduces context drift.