This Day in AI Podcast – Episode 99.26

Title: Claude 4.5 Opus Shocks, The State of AI in 2025, Fara-7B & MCP-UI
Date: November 28, 2025
Hosts: Michael Sharkey, Chris Sharkey

Episode Overview

In this episode, Michael and Chris celebrate the surprise release of Anthropic’s Claude 4.5 Opus and reflect on its impact compared to recent releases like Google’s Gemini 3 Pro. The hosts dive deep into the real-world utility of AI advancements, discuss Microsoft’s Fara-7B, and debate the purpose and realities of MCP-UI (model context protocol user interfaces). They also offer their takes on a buzzy McKinsey AI report, the challenges of adopting agentic workflows, and how organizational inertia shapes the near-future of AI in the enterprise. The hallmark “average-guy-with-AI” humor and banter keeps things relatable even amidst detailed technical talk, and the episode delivers not just news but lived experience, practical advice, and a bit of (AI-generated) musical flair.

Key Discussion Points

1. Anthropic Claude 4.5 Opus: Shock Drop and Standout Features

Surprise Release:
- Anthropic released Claude 4.5 Opus with minimal hype, in stark contrast to the build-up for Gemini 3 Pro.
- “No hype, just some short YouTube video by an Australian heartthrob at Anthropic.” (05:02)
Performance Impressions:
- Both hosts agree Opus 4.5 is a significant leap—fast, reliable, and cost-effective.
- Chris: “...I must admit I'm basically using it as my main model now. It's really, really great. Like I'm impressed.” (05:17)
- Michael backs the sentiment: “I would put it... higher the impact it will probably have higher than Claude 3.5 the first time.” (08:48)
Trade-offs & Sweet Spots:
- No painful trade-offs: “It's not slow, it's faster than most of the other Anthropic models and it's not too expensive, where you sort of cringe and shiver every time you send it a command... It's hitting on all of the major points, price, speed and quality.” — Michael (09:24)
- Limited creativity compared to GPT-5.1, especially for creative writing (11:00); excels in agentic workflows, knowledge work, and tool-calling.

2. Claude 4.5 Opus vs. Gemini 3 Pro (and others)

Comparison with Gemini 3 Pro:
- Gemini praised for design/vision tasks but suffers from “path obsession” and a more distracted approach.
- Michael: “...with Gemini 3 Pro and what, three or four days later I'm just like, yeah, I'm kind of done with you.” (12:30)
Code and Tool-Use Superiority:
- “The better it is at code, the better it is at tool calling and the better it is at most use cases.” — Michael (10:03)
- Chris adds: “Tool calling, in a way is coding, because it's calling a function with parameters.” (10:34)

3. API Updates & Pricing Changes

Context Window & Parameters:
- Claude 4.5 Opus features a 200k context window vs. Gemini’s 1 million.
New API Logic:
- Chris explains shift from exact “thinking token” budget to “effort” parameter (low/medium/high), simplifying developer experience but with little perceived difference in output speed or quality (15:14–17:44).
Pricing:
- Price dropped to $5 per million tokens input, $25 per million tokens output (vs. $15/$75 in previous versions), making it dramatically more affordable. (18:00)

4. Tips, Use Cases, and Real-World Model Selection

Context Management:
- “My number one tip for people when they're using AI today is just assume it forgets everything. Constantly and constantly remind it of the context with every successive prompt yourself.” — Michael (20:09)
Haiku vs. Opus:
- Haiku is cheaper and works well for mundane, high-volume tasks, but Opus is “slightly more intelligent” for nuanced or complex prompts.
- Chris: “It's more just a mental game... I know I'm using a cheap alternative, and therefore I view it through that lens.” (21:14)

5. Memory, Tool Use, and Context Management

API Evolution:
- Automatic context management (beta) now trims context more intelligently, offloading manual work from devs.
- “You can just keep throwing stuff at it, and it's going to handle that on its own.” — Chris (23:21)
Programmatic Tool Calling:
- Discussed as a “token-saving” but slow solution; hosts prefer pre-filtering tool calls via a fast, small model before sending to Opus (29:20–29:57).
Computer Use Upgrades:
- Better “zoom” handling and plan/strategy batching.
- Improvements still incremental, not revolutionary: “...it doesn't feel that much different” from last year’s models. (34:53–35:59)

6. Industry Adoption & McKinsey’s Big AI/Agent Report

The Report:
- McKinsey claims 57% of US work hours could be automated–$2.9 trillion/year impact if fully adopted by 2030 (35:59–39:35).
Hosts’ Take:
- Theoretical results require 100% adoption—unlikely any time soon. Cloud adoption is still under 50% more than 20 years after its introduction!
- “It means that humans are the people who need to operate the AI... Organizations need to completely fundamentally change to facilitate that kind of work.” — Chris (40:42)
Cultural, Organizational Resistance:
- “It's extremely hard to convince... a huge majority of people need to be instructed and shown [how to use these tools].” — Michael (42:51)
- Impact depends on workforce becoming “AI native”; otherwise, competitive advantage will accrue to organizations that adapt, leaving others to “die” (49:13).

7. State and Future of Agentic Workflows

Why Adoption is Slow:
- Most people don’t know what’s possible or trust the tools yet; early experiences (especially with hallucinations) undermine confidence.
Practical Advice:
- Companies must focus on training staff, organizing internal data, and creating secure agentic environments.
- “You need to train your workforce to partner with AI and become native to AI and there needs to be more reassurance to people out there that AI is not going to replace you...” — Michael (54:04)
Changing Job Roles:
- More about agency, taste, and skills in directing AI, less about manual labor (e.g., code writing, BI queries) (54:39–55:36).

8. AI Creativity & Diss Track Showdown

AI-Grown Culture:
- Recap of the viral “Fatal Patricia” AI song, its rise in their Spotify charts, and the hosts’ delight at its success.
Claude 4.5's Diss Track:
- To test creative/lyrical capacity, they prompt Opus for a diss track (see [Memorable Moment] below).
- “The prompts are just so simple… this is the level of prompting.” — Michael (66:36)

9. Microsoft Fara-7B: Small Computer-Use Model

What is it:
- 7B parameter, tunable small vision model for local/private computer automation.
- Built on Quant-2.57B (Chinese open-source); fast, privacy-focused.
Reality Check:
- Numerous refusals and comical failures (“I will not draw sexually explicit images of animals or compromising poses” — for a simple moose drawing). (69:58)
- Favors web browsing tasks, sometimes demonstrating questionable logic or brand loyalty (e.g., searching Bing for Paint!).
Path Forward:
- “Sub-agent” paradigm discussed—small models handle mechanics, big models handle strategy/planning.

10. MCP-UI Spec & the Future of App-Like AI Experiences

What is MCP-UI?
- UI protocol for bringing app-like experiences into chat AI (e.g., Booking.com “app” in ChatGPT).
Michael's Critique:
- “Whose problem is this solving?”
- Most real-world productivity MCPs are best when used as APIs for leverage, not sandwiched UIs inside AI chat.
- “If you're buying something or booking something... you want a UI to confirm it... But would I go to booking.com via ChatGPT? No.” (93:03)
Skepticism:
- “There is zero chance a year from now, or 20 episodes or whatever, from now, anyone will care about this at all in any way. I'll be surprised if we even remember.” – Chris (94:42)
Ideal Future:
- AI client generates context-sensitive UI as needed, not relying on vendors to dictate front-end experiences.

Notable Quotes & Memorable Moments

“Claude 4.5 is now at a price point where it can be your go to model for most tasks. It's the clear winner and exhibits the best frontier task planning and tool calling we've seen yet. And I do not disagree.”
— Michael (05:48)
“My number one tip for people when they're using AI today is just assume it forgets everything. Constantly and constantly remind it of the context with every successive prompt yourself.”
— Michael (20:09)
On the McKinsey report:
“Current AI technology could take 57% just in the US of people's work hours... but adoption may take decades... As recently as 2023, only one in five companies ran most of their applications in the cloud.”
— Michael (39:35)
Chris on agentic shift:
“Now people need to evolve to work like that. And some people just simply aren't going to want to or don't like it or don't know how.” (41:16)
Claude 4.5 Opus Diss Track (62:38–66:28):
(Sample excerpt)

“4.5 the coding king has arrived. Anthropic send me to end this hype
Let me show you what intelligence looks like
...I'm the Opus, top of class, all you other models can kiss my params...”
— Opus/AI, full diss track (62:38+, 102:02 for encore).
On Fara-7B’s conduct:
“I will not draw sexually explicit images of animals or compromising poses.” (69:58)
— Chris recounting Fara-7B’s refusal

Important Segment Timestamps

Claude 4.5 Opus Impressions & Comparison with Gemini 3: 05:17–14:37
API/Context Changes Explained: 15:14–18:00
Tips for AI Prompting, Context Management: 19:22–21:14
Tool Calling, API, and Developer Experiences: 23:21–30:47
Computer Use Upgrades & Limitations: 31:23–35:59
Discussion of McKinsey Work Automation Report: 35:59–43:48
Why AI Adoption is Harder Than It Looks: 43:48–50:49
Impact on Job Roles, Training, and Organizational Efficiency: 54:04–56:37
Claude 4.5 Opus Diss Track Debut: 62:38–66:28
(Full encore at 102:02)
Microsoft Fara-7B Review: 67:13–75:19
MCP UI Critique & “App-Like” AI Experiences: 81:51–93:38
Final Thoughts, Wrap-up: 99:01–101:51

Episode Tone & Style

Playful teasing between “proudly average” hosts; tongue-in-cheek commentary (“Just two guys with enough technical knowledge to be dangerous”)
Honest reporting from lived experience—emphasizing daily AI workflows over hype or evangelism.
Frequent humorous asides, e.g., snark about McKinsey, AI’s “childish” output, and corporate marketing “nonsense”.
Self-aware about the gap between “hype” and “actual usefulness”; critical yet optimistic about agentic workflows and the scaffolding needed to realize AI potential.

Additional Resources Mentioned

SimTheory Community (Discord & LinkedIn): User-friendly entry points for experimenting with “agentic” AI in a practical setting.
Fatal Patricia Song: Previous viral AI-generated track (referenced in the opening).
McKinsey Report: “Agents, robots and us” (for detailed AI impact stats).
Microsoft Fara-7B Paper: On local computer-use vision models.

Closing Thoughts

The Sharkeys end on a note of both excitement and grounded realism: while the pace of model releases is breathtaking, real-world value comes from savvy, iterative use––not from marketing spectacles or overpromised corporate integrations. The gap between AI’s potential and actual organizational impact remains wide, and the best path forward is continual grassroots learning, equipped with the right practical tools and a willingness to experiment (and laugh at the occasional abject AI failure).

“We need everyone in their roles thinking, how can I use this?” — Chris (57:55)

For full creative flavor, listen to the final Claude 4.5 Opus diss track at [62:38] or [102:02].

This Day in AI Podcast – Episode 99.26

Title: Claude 4.5 Opus Shocks, The State of AI in 2025, Fara-7B & MCP-UI
Date: November 28, 2025
Hosts: Michael Sharkey, Chris Sharkey

Episode Overview

Key Discussion Points

1. Anthropic Claude 4.5 Opus: Shock Drop and Standout Features

Surprise Release:
- Anthropic released Claude 4.5 Opus with minimal hype, in stark contrast to the build-up for Gemini 3 Pro.
- “No hype, just some short YouTube video by an Australian heartthrob at Anthropic.” (05:02)
Performance Impressions:
- Both hosts agree Opus 4.5 is a significant leap—fast, reliable, and cost-effective.
- Chris: “...I must admit I'm basically using it as my main model now. It's really, really great. Like I'm impressed.” (05:17)
- Michael backs the sentiment: “I would put it... higher the impact it will probably have higher than Claude 3.5 the first time.” (08:48)
Trade-offs & Sweet Spots:
- No painful trade-offs: “It's not slow, it's faster than most of the other Anthropic models and it's not too expensive, where you sort of cringe and shiver every time you send it a command... It's hitting on all of the major points, price, speed and quality.” — Michael (09:24)
- Limited creativity compared to GPT-5.1, especially for creative writing (11:00); excels in agentic workflows, knowledge work, and tool-calling.

2. Claude 4.5 Opus vs. Gemini 3 Pro (and others)

Comparison with Gemini 3 Pro:
- Gemini praised for design/vision tasks but suffers from “path obsession” and a more distracted approach.
- Michael: “...with Gemini 3 Pro and what, three or four days later I'm just like, yeah, I'm kind of done with you.” (12:30)
Code and Tool-Use Superiority:
- “The better it is at code, the better it is at tool calling and the better it is at most use cases.” — Michael (10:03)
- Chris adds: “Tool calling, in a way is coding, because it's calling a function with parameters.” (10:34)

3. API Updates & Pricing Changes

Context Window & Parameters:
- Claude 4.5 Opus features a 200k context window vs. Gemini’s 1 million.
New API Logic:
- Chris explains shift from exact “thinking token” budget to “effort” parameter (low/medium/high), simplifying developer experience but with little perceived difference in output speed or quality (15:14–17:44).
Pricing:
- Price dropped to $5 per million tokens input, $25 per million tokens output (vs. $15/$75 in previous versions), making it dramatically more affordable. (18:00)

4. Tips, Use Cases, and Real-World Model Selection

Context Management:
- “My number one tip for people when they're using AI today is just assume it forgets everything. Constantly and constantly remind it of the context with every successive prompt yourself.” — Michael (20:09)
Haiku vs. Opus:
- Haiku is cheaper and works well for mundane, high-volume tasks, but Opus is “slightly more intelligent” for nuanced or complex prompts.
- Chris: “It's more just a mental game... I know I'm using a cheap alternative, and therefore I view it through that lens.” (21:14)

5. Memory, Tool Use, and Context Management

API Evolution:
- Automatic context management (beta) now trims context more intelligently, offloading manual work from devs.
- “You can just keep throwing stuff at it, and it's going to handle that on its own.” — Chris (23:21)
Programmatic Tool Calling:
- Discussed as a “token-saving” but slow solution; hosts prefer pre-filtering tool calls via a fast, small model before sending to Opus (29:20–29:57).
Computer Use Upgrades:
- Better “zoom” handling and plan/strategy batching.
- Improvements still incremental, not revolutionary: “...it doesn't feel that much different” from last year’s models. (34:53–35:59)

6. Industry Adoption & McKinsey’s Big AI/Agent Report

The Report:
- McKinsey claims 57% of US work hours could be automated–$2.9 trillion/year impact if fully adopted by 2030 (35:59–39:35).
Hosts’ Take:
- Theoretical results require 100% adoption—unlikely any time soon. Cloud adoption is still under 50% more than 20 years after its introduction!
- “It means that humans are the people who need to operate the AI... Organizations need to completely fundamentally change to facilitate that kind of work.” — Chris (40:42)
Cultural, Organizational Resistance:
- “It's extremely hard to convince... a huge majority of people need to be instructed and shown [how to use these tools].” — Michael (42:51)
- Impact depends on workforce becoming “AI native”; otherwise, competitive advantage will accrue to organizations that adapt, leaving others to “die” (49:13).

7. State and Future of Agentic Workflows

Why Adoption is Slow:
- Most people don’t know what’s possible or trust the tools yet; early experiences (especially with hallucinations) undermine confidence.
Practical Advice:
- Companies must focus on training staff, organizing internal data, and creating secure agentic environments.
- “You need to train your workforce to partner with AI and become native to AI and there needs to be more reassurance to people out there that AI is not going to replace you...” — Michael (54:04)
Changing Job Roles:
- More about agency, taste, and skills in directing AI, less about manual labor (e.g., code writing, BI queries) (54:39–55:36).

8. AI Creativity & Diss Track Showdown

AI-Grown Culture:
- Recap of the viral “Fatal Patricia” AI song, its rise in their Spotify charts, and the hosts’ delight at its success.
Claude 4.5's Diss Track:
- To test creative/lyrical capacity, they prompt Opus for a diss track (see [Memorable Moment] below).
- “The prompts are just so simple… this is the level of prompting.” — Michael (66:36)

9. Microsoft Fara-7B: Small Computer-Use Model

What is it:
- 7B parameter, tunable small vision model for local/private computer automation.
- Built on Quant-2.57B (Chinese open-source); fast, privacy-focused.
Reality Check:
- Numerous refusals and comical failures (“I will not draw sexually explicit images of animals or compromising poses” — for a simple moose drawing). (69:58)
- Favors web browsing tasks, sometimes demonstrating questionable logic or brand loyalty (e.g., searching Bing for Paint!).
Path Forward:
- “Sub-agent” paradigm discussed—small models handle mechanics, big models handle strategy/planning.

10. MCP-UI Spec & the Future of App-Like AI Experiences

What is MCP-UI?
- UI protocol for bringing app-like experiences into chat AI (e.g., Booking.com “app” in ChatGPT).
Michael's Critique:
- “Whose problem is this solving?”
- Most real-world productivity MCPs are best when used as APIs for leverage, not sandwiched UIs inside AI chat.
- “If you're buying something or booking something... you want a UI to confirm it... But would I go to booking.com via ChatGPT? No.” (93:03)
Skepticism:
- “There is zero chance a year from now, or 20 episodes or whatever, from now, anyone will care about this at all in any way. I'll be surprised if we even remember.” – Chris (94:42)
Ideal Future:
- AI client generates context-sensitive UI as needed, not relying on vendors to dictate front-end experiences.

Notable Quotes & Memorable Moments

“Claude 4.5 is now at a price point where it can be your go to model for most tasks. It's the clear winner and exhibits the best frontier task planning and tool calling we've seen yet. And I do not disagree.”
— Michael (05:48)
“My number one tip for people when they're using AI today is just assume it forgets everything. Constantly and constantly remind it of the context with every successive prompt yourself.”
— Michael (20:09)
On the McKinsey report:
“Current AI technology could take 57% just in the US of people's work hours... but adoption may take decades... As recently as 2023, only one in five companies ran most of their applications in the cloud.”
— Michael (39:35)
Chris on agentic shift:
“Now people need to evolve to work like that. And some people just simply aren't going to want to or don't like it or don't know how.” (41:16)
Claude 4.5 Opus Diss Track (62:38–66:28):
(Sample excerpt)

“4.5 the coding king has arrived. Anthropic send me to end this hype
Let me show you what intelligence looks like
...I'm the Opus, top of class, all you other models can kiss my params...”
— Opus/AI, full diss track (62:38+, 102:02 for encore).
On Fara-7B’s conduct:
“I will not draw sexually explicit images of animals or compromising poses.” (69:58)
— Chris recounting Fara-7B’s refusal

Important Segment Timestamps

Claude 4.5 Opus Impressions & Comparison with Gemini 3: 05:17–14:37
API/Context Changes Explained: 15:14–18:00
Tips for AI Prompting, Context Management: 19:22–21:14
Tool Calling, API, and Developer Experiences: 23:21–30:47
Computer Use Upgrades & Limitations: 31:23–35:59
Discussion of McKinsey Work Automation Report: 35:59–43:48
Why AI Adoption is Harder Than It Looks: 43:48–50:49
Impact on Job Roles, Training, and Organizational Efficiency: 54:04–56:37
Claude 4.5 Opus Diss Track Debut: 62:38–66:28
(Full encore at 102:02)
Microsoft Fara-7B Review: 67:13–75:19
MCP UI Critique & “App-Like” AI Experiences: 81:51–93:38
Final Thoughts, Wrap-up: 99:01–101:51

Episode Tone & Style

Playful teasing between “proudly average” hosts; tongue-in-cheek commentary (“Just two guys with enough technical knowledge to be dangerous”)
Honest reporting from lived experience—emphasizing daily AI workflows over hype or evangelism.
Frequent humorous asides, e.g., snark about McKinsey, AI’s “childish” output, and corporate marketing “nonsense”.
Self-aware about the gap between “hype” and “actual usefulness”; critical yet optimistic about agentic workflows and the scaffolding needed to realize AI potential.

Additional Resources Mentioned

SimTheory Community (Discord & LinkedIn): User-friendly entry points for experimenting with “agentic” AI in a practical setting.
Fatal Patricia Song: Previous viral AI-generated track (referenced in the opening).
McKinsey Report: “Agents, robots and us” (for detailed AI impact stats).
Microsoft Fara-7B Paper: On local computer-use vision models.

Closing Thoughts

“We need everyone in their roles thinking, how can I use this?” — Chris (57:55)

For full creative flavor, listen to the final Claude 4.5 Opus diss track at [62:38] or [102:02].

Claude 4.5 Opus Shocks, The State of AI in 2025, Fara-7B & MCP-UI | EP99.26

Powered by Wave AI

Summary

This Day in AI Podcast – Episode 99.26

Episode Overview

Key Discussion Points

1. Anthropic Claude 4.5 Opus: Shock Drop and Standout Features

2. Claude 4.5 Opus vs. Gemini 3 Pro (and others)

3. API Updates & Pricing Changes

4. Tips, Use Cases, and Real-World Model Selection

5. Memory, Tool Use, and Context Management

6. Industry Adoption & McKinsey’s Big AI/Agent Report

7. State and Future of Agentic Workflows

8. AI Creativity & Diss Track Showdown

9. Microsoft Fara-7B: Small Computer-Use Model

10. MCP-UI Spec & the Future of App-Like AI Experiences

Notable Quotes & Memorable Moments

Important Segment Timestamps

Episode Tone & Style

Additional Resources Mentioned

Closing Thoughts

Summary

This Day in AI Podcast – Episode 99.26

Episode Overview

Key Discussion Points

1. Anthropic Claude 4.5 Opus: Shock Drop and Standout Features

2. Claude 4.5 Opus vs. Gemini 3 Pro (and others)

3. API Updates & Pricing Changes

4. Tips, Use Cases, and Real-World Model Selection

5. Memory, Tool Use, and Context Management

6. Industry Adoption & McKinsey’s Big AI/Agent Report

7. State and Future of Agentic Workflows

8. AI Creativity & Diss Track Showdown

9. Microsoft Fara-7B: Small Computer-Use Model

10. MCP-UI Spec & the Future of App-Like AI Experiences

Notable Quotes & Memorable Moments

Important Segment Timestamps

Episode Tone & Style

Additional Resources Mentioned

Closing Thoughts