Podcast Summary: This Day in AI – "What did OpenAI Announce at DevDay? Apps SDK, MCP UI & Impact to SaaS" (EP99.20-APPS)

Hosts: Michael Sharkey & Chris Sharkey
Date: October 10, 2025
Main Theme:
A deep-dive, candid conversation about OpenAI's major announcements at DevDay 2025: the launch of the Apps SDK, changes to MCP UI, new AI models, and what it means for SaaS products and everyday AI users. Michael and Chris, self-proclaimed "average" tech enthusiasts, dissect the news, critique the developer/user experience, and share their vision (and skepticism) of an "AI-first" software future.

Main Topics Discussed

1. OpenAI DevDay 2025 Announcements

Timestamps: 00:53–03:20

• Major releases:
  • Apps in ChatGPT (essentially MCP with UI components)
  • Agent Kit (for deploying 'production grade' chat experiences in SaaS apps)
  • Sora 2 API (video generation, watermark-free)
  • New Models:
    • GPT-5 Pro (API release)
    • Real Time Mini (cheap, high-quality voice)
    • GPT Image 1 Mini (cheaper, slightly lower quality)
• Initial excitement but skepticism about how game-changing these will be for devs & users.

"This holiday season I will be having the Christmas tree lights when I'm working on my Mac because I do think it's cool. But anyway, we should talk about this..."
—Michael (01:00)

2. Apps SDK & MCP UI: A Developer’s Perspective

Timestamps: 03:20–10:00

• Apps SDK is a formalization of what many were already doing in a “hacky way” for MCPs—OpenAI now provides official, standard UI components.
• Under the hood, it feels reminiscent of React/XML. OpenAI enables devs to define input/output controls for their apps—easing custom app creation inside ChatGPT and similar interfaces.
• Resource referencing (rather than tool-calling for every step) lets ChatGPT render UI for users "straight away".

“What OpenAI has done is basically made a standard way of doing that, where you use it...looks like React... they've taken that and made a standard around it.”
—Chris (05:17)

Key Insights:

• While the UI toolkit is now standardized, it doesn’t give any one company an advantage—the missing link was the UI layer for these agentic tools.
• Apps SDK could enable dynamic UIs, but is currently focused on one app/resource at a time.

3. First Impressions: App Installs, UI Friction, and Limitations

Timestamps: 10:00–19:00

• App Discovery/Installation:
  • Confusing, multi-step process to add apps like Canva in ChatGPT. Not yet consumer-friendly (will likely be improved).
• Single-App Focus:
  • You use/“focus on” one MCP/app at a time (no parallel multi-tool magic).
  • Testing Canva: UI allowed starting a slide deck but not iterating via chat.
  • Testing Booking.com: No rich UI feedback; felt subpar.
  • Testing Spotify: Despite safety checks, providing a "wealthy party playlist" was surprisingly good—but overall, the hands-on experience is often clunky.

“You can only focus on one MCP at a time. That... just cuts out so many research use cases...”
—Michael (14:41)

Memorable Critique:
"It's really just a worse interface for whatever the application is... The human in the loop interactivity on every step is a massive step away from what's already possible."
—Chris (15:21)

4. The Agency/Aggregator Dilemma: Why This Misses the Mark

Timestamps: 19:00–29:00

• Using MCPs for complex workflows and true agency/automation isn’t possible in the current setup.
• The future feels like it should have AI agents wielding tools for you—without constant human intervention—but OpenAI’s new SDK keeps the human in every loop.
• Concern: Is AI just recreating every SaaS app as a window/panel in ChatGPT, or is ChatGPT on a path to Sherlock (absorb and replace) the most-used apps?

“It’s just temporary, that all these apps are integrating. ...this idea of treating it like a sort of generic piece of software with plugins with brand names and logos... seems like a misguided way of getting towards what everyone actually wants.”
—Chris (20:06)

“All these companies go out and... rush to go and embed themselves in ChatGPT... and then the inverse happens where ChatGPT is like ‘okay, we’ll take the top five... and clone them in ChatGPT.’ That whole Sherlocking that Apple’s famous for; you can kind of just see this playing out so clearly.”
—Michael (23:00)

5. Strategic Risks & The SaaS Industry Impact

Timestamps: 24:00–34:00

• SaaS companies face a lose-lose: integrate deeply with ChatGPT or risk irrelevance, yet risk giving up their user relationships and being replaced.
• Most initial MCP-based integrations from major companies are "half-assed," described as more for brand protection and FOMO than actual user benefit.
• Real value is for power users or those leveraging AI for multiplicative productivity, not consumers wanting an easier way to check the weather.

“They're not... embracing it in a way that's actually useful. So many of the MCPs that especially big companies have released are not that useful.”
—Chris (26:24)

6. The Agency ‘Agent Builder’: Automation vs. True Agency

Timestamps: 42:05–49:30

• Visual Agent Builder (“Agents”): More like a visual, step-by-step workflow tool (e.g., Zapier), not true agents.
• Irony: The more a workflow looks like a flowchart, the less agentic/intelligent (in the AI sense) it actually is.
• Hosts are skeptical about who will benefit from these tools; devs might prefer to write code, and non-coders may find the tools overwhelming.

“The second you put a flowchart up, you’re like, that’s not agency... You wouldn’t have an employee who has to follow an exact procedure every time and say, oh, they’re an agent...”
—Chris (43:25)

“If you’re someone who can code already... it’s quicker to vibe code this... if you don't have the skills... this isn’t good for that either. To me you have that problem of who’s this for? And I don't think they know.”
—Michael (48:08)

7. Model Announcements – Pricing, Performance, Reactions

Timestamps: 49:30–56:50

• GPT-5 Pro API: Hugely expensive, making it inaccessible for most real use. Tokens can cost so much that a single heavy output could bankrupt an app if not careful.
• Real Time Mini Model:
  • Voice-based, near real-time conversations at affordable pricing (though still more expensive for audio tokens).
  • Useful for voice UIs, fast but not as smart.
• Sora 2 API:
  • Video generation, 10–30 cents per second (more for high-res).
  • Good for serious video work, but VO3 (by another company) is considered “just better” even if harder to access.

"It's expensive. It's too expensive. You got onto me immediately and said, Chris, you need to add this into Sim Theory... and I'm like, people are going to use up their entire token allocation on their first request."
—Chris on GPT-5 Pro (49:33)

8. Vision: The AI Superapp Future & Power Users

Timestamps: 33:30–40:30

• Consumer Angle: OpenAI's clear direction is to become the UI hub for many daily tasks (search, calendar, flights, shopping, etc.).
• But for power users and companies, value will accrue to those platforms/SaaS tools that integrate via well-designed APIs/MCPs facilitating truly agentic workflows.
• There's skepticism about whether consumers or companies will hand over their data and user relationships so easily.

“The future is agentic CRUD with instantaneous generative UIs when needed to display data.”
—Michael, quoting Kyle Corbett (36:46)

9. Final Thoughts – Disappointment & Hope for the Real Agency

Timestamps: 39:42–end

• Disappointment at “apps” that are effectively chat-based wrappers for existing services and lack the true promise of “AI agents doing work for you.”
• Dynamic, context-aware agentic workflows driven by tool selection, output collation, and only temporary/custom UI are seen as the real prize.
• Short-term: consumer interfaces may remain clunky, and SaaS companies will be in defensive/experimental mode for a while.

“It's not what ChatGPT has done with it. It's almost like they've sort of made the thing we all need to get through this period... but they haven't really used it in a way that's actually beneficial.”
—Chris (39:20)

“The custom UI is dynamic, it's context driven... that's what AI is brilliant at. It's amazing at classification, it's amazing at holistically understanding a problem and presenting you with a way to resolve that problem.”
—Chris (38:20)

Notable Quotes & Moments

• "It's almost the inverse equivalent of having a little Clippy AI helper—it's just a worse interface..."
  —Chris (15:21)
• "Treating it like one universal console just isn't as useful because you don't want it distracted by all the other stuff."
  —Chris (41:27)
• "It's such a letdown. It's painful."
  —Michael (35:48)
• "Maybe the benchmark should take into account affordability. Are you really getting that much more intelligence?"
  —Michael (51:27)
• "Make it so expensive no one can verify your claims. So you're like, 'Oh no, it's AGI,' but no one can actually afford to try it."
  —Chris (51:45)

Conclusion & Takeaways

• OpenAI Apps SDK brings welcome standards to MCP user interfaces, but the single-app focus and lack of parallel tool/agent support is a major limitation for power users.
• Agent Builder is more like a visual scripting tool than true agency. Useful for narrow, repeatable tasks— not for real AI-powered agents.
• New Models (GPT-5 Pro, Sora 2, Real Time Mini): Some exciting potential; pricing and accessibility remain pain points.
• Big Picture: OpenAI is building the “OS layer” for consumer web/app activity, but there's a tension between utility, control, and who ultimately owns the user relationship.
• Hosts’ take: Current implementations underwhelm for “pro”/power users; killer use cases may be yet to come, with most AI's real productivity powers still waiting on more open, flexible agentic frameworks.

If you’re a developer, SaaS founder, or ambitious AI user: wait and watch, experiment with the new UIs, but don’t cede your user relationship to a panel in someone else’s superapp just yet.

Bonus:
Stick around at the end for the Sharkey brothers' AI-generated diss tracks & reminders about their "perfectly average" Spotify albums. Because “most of us are figuring this out as we go along.”

Episode Key Segments

| Time | Segment | |----------|---------------------------------------------------------------| | 00:53 | OpenAI DevDay Announcements Overview | | 03:20 | Apps SDK/MCP UI Structural Deep Dive | | 10:00 | App Installation/First Impressions | | 14:41 | Frustrations: Single-app Limitations | | 19:00 | Agency and Agency Paradoxes | | 24:00 | SaaS Industry Fears and Strategic Risks | | 33:30 | Superapp, Aggregator, Consumer vs Power User Viewpoints | | 42:05 | Agent Builder (Visual/No-Code Tool) Critique | | 49:30 | Model Pricing & Accessibility (GPT-5 Pro, Mini, Sora 2) | | 56:50 | Sora 2 vs VO3—Video Model Smackdown | | 61:01 | Outro, Diss Tracks, and Personal Reflections |

For those who want the real AI productivity boost: think less about “apps,” more about agency. Demand more from your tools, and maybe don’t pay $120 per million tokens just yet.