2025 AI Roadmap Rewind Human vs Machine, AI Models Shrink, and AGI No One Noticed

Summary

Everyday AI Podcast – 2025 AI Roadmap Rewind: Human vs Machine, AI Models Shrink, and AGI No One Noticed
Host: Jordan Wilson
Date: December 18, 2025
Episode Focus:
A fact-checking review of bold AI predictions made for 2025, examining what came true, what missed the mark, and what it means for professionals navigating AI’s rapid evolution. This episode covers predictions 12 through 1, looking at copyright lawsuits, AI influencers, “vibe coding,” the politics of AI, small models, AGI, and more.

Episode Overview

Jordan Wilson closes out the year by revisiting the second half of his "25 for 2025" AI predictions, grading their accuracy with real-world examples and industry news. Listeners get an honest appraisal of what 2025’s most significant AI shifts were, how AI quietly changed society, and whether we might have crossed the line into AGI (artificial general intelligence) without fanfare.

Key Discussion Points & Detailed Insights

The First Big Copyright Case Gets Settled

[03:35]

Prediction: The first major AI copyright lawsuit would be settled in 2025.
Outcome: Came true with the $1.5 billion class-action settlement against Anthropic for allegedly using copyrighted books in model training.
Impact: Signals a shift toward a "pay-to-train" model, where agreements like the Disney-OpenAI partnership ($1B for IP access) become the norm.
Jordan’s Assessment:

“This is, I think, one of the first big pieces that eventually we’re going to move to a pay-to-train model.” [04:55]

AI Influencers Begin to Replace Human UGC Creators

[07:43]

Prediction: AI influencers–virtual avatars–would start to dominate social media user-generated content (UGC).
Reality: Major brands are rolling out UGC campaigns with AI-generated influencers. Example: Lil Maela, with 3M+ followers, earning up to $10M/year, and partnerships with Samsung, Calvin Klein, and Prada.
Jordan’s Commentary:

“Probably by the end of next year, I’d say maybe 80% of what you might see online is going to be AI." [10:46]
Platforms' Role: Tools like TikTok’s Symphony Gen AD and Meta’s AI ad tools are accelerating this trend.

“Vibe Coding”—Non-Techies Build Apps on the Fly

[13:00]

Prediction: Non-technical people would be able to "build on the fly" software.
Result: True, and the term “vibe coding” (coined by Andrej Karpathy in February) now describes this phenomenon. Tools like Google’s experimental Disco, AI Studio, Gemini Canvas, and others allow anyone to create web apps or features with minimal effort.
Notable Quote:

“The people that are winning hackathons are non-technical people.” (Citing a conversation with Paige Bailey at Google) [15:49]
Stat: 70% of new enterprise apps in 2025 were built using low-code AI tools.

Reasoner Wrappers for Enterprise Knowledge

[17:51]

Prediction: Tools would emerge to wrap reasoning models with enterprise data, enhancing decision-making.
What Happened: Some movement (e.g., tools like Leta, growth in agent frameworks and observability), but the prediction was ahead of its time.
Jordan’s Take:

“Maybe just a little early on this prediction…but there is probably this new agency layer.” [18:30]

Virtual Machines (VMs) for Agents Go Mainstream

[20:30]

Prediction: VMs/virtual desktops would regain popularity as agents need “computers” to operate.
Accuracy: Correct. Major cloud platforms now supply dedicated AI agent environments, e.g., Microsoft’s Windows 365 for agents.
Memorable Moment:

“It kind of seemed like a niche prediction…But no one was talking about virtual machines. And now they’re definitely a thing.” [22:53]

AI’s Political Transformation in the U.S.

[24:00]

Prediction: AI would become entangled in politics and policy.
Validation: Clear. President Trump signed an executive order blocking states from regulating AI, making AI regulation a federal-only matter—a break from usual Republican policy.
Quote:

“There is literally no bigger way for AI to become political than at the federal level, saying: hey states, you can't make laws on this.” [25:15]
Other Notes: AI leaders, including Zuckerberg, Altman, and Apple, donated significant sums to political campaigns; U.S. signed the “AI Action Plan,” explicitly banning “woke” or ideologically biased models for federal contracts.

Global Regulation Tightens, But the U.S. Stagnates

[29:29]

Prediction: AI rules globally would grow stricter, except in the U.S.
Evidence: EU’s AI Act, Singapore’s assurance pilot, and Italy’s national law restricted or delayed access/features for major LLMs; U.S. largely remained hands-off.
Effect: Users in regulated regions face delays or limitations; Americans have almost no restrictions.

Narrow Agents Dominate, Not General Purpose

[32:05]

Prediction: Narrow, task-specific AI agents would thrive, while general-purpose agents would lag.
Reality: 100% accurate. Major agents (e.g., Salesforce Force2O, Oracle’s role-based agents, GitHub Copilot, etc.) are specialized. General agents (OpenAI, Copilot) remain clunky.
Data: Menlo Ventures reports $7.3B enterprise investment in narrow/departmental AI.

LLM Memory Becomes a Focus

[34:56]

Prediction: LLM “memory” (context, persistence) would become standard and a major upgrade.
Validation: All major models now have robust memory features, including OpenAI, Gemini, and Claude; context caching and personalization are common.
Jordan’s Analysis:

“Context plus connectors plus memory was definitely the biggest leap forward…in 2025.” [37:15]

Large Language Models (LLMs) Become Small Language Models (SLMs)

[39:12]

Prediction: Small, efficient models would surpass some older large models and see greater use.
Result: Confirmed with models like GPT-OSS 20B—just 1% the size of old GPT-4, yet equal or superior in many tasks. Also emphasized are Microsoft’s Phi, Google’s Gemma, and domain-specific mini-models.
Jordan’s Summary:

“The future of large language models is using many small domain-specific models.” [42:07]

Mixture of Models Surfaces as a New Standard

[43:11]

Prediction: Systems would begin orchestrating multiple models ("mixture of models" – distinct from "mixture of experts") for enhanced output.
Outcome: This approach emerged—example: Zoom’s ensemble system beat the “Humanity’s Last Exam.” Google’s Interactions API and startups like Area raised big funding on this premise. IDC projects 70% of top AI-driven enterprises will use multi-model architectures by 2028.
Jordan’s Forecast:

“The smartest companies are having a modular approach when it comes to their AI strategy.” [46:01]

Did AGI Happen in 2025… and No One Noticed?

[47:08]

Prediction: AGI (as defined by outperforming humans at most economically valuable work) would arrive quietly, with no public “AGI moment.”
Jordan’s Evidence:
- LLMs outperform top humans on elite tests (e.g., IQ tests, International Mathematical Olympiad).
- OpenAI’s definition: AGI means "highly autonomous systems that outperform humans at most economically valuable work.” (OpenAI charter, 2018)
- GDP VAL Test: GPT-5.2 “Thinking” model outperformed or tied expert humans on 70.9% of tasks; “Pro” model at 74.1%.
Reflection:

“On economically valuable work judged by experts… the AI did better three-quarters of the time… and no one really noticed.” [53:11] “As AI gets smarter, the goalposts keep moving.” [54:00]

Notable Quotes & Memorable Moments

On the speed of AI progress:

“This is how fast time flies in AI world. This was before vibe coding... And now, vibe coding is baked into browsers.” [13:06]
On political influence:

“It doesn’t get more political in AI than the cozying up of big tech leaders with the federal government or President Trump.” [27:43]
On narrow agents:

“There’s no one great agent out there... The agents that have been making the biggest splashes are just narrow agents, agents built around a certain vertical, a certain task.” [32:59]
On AGI’s fuzzy line:

“If you were just to look at those old definitions [of AGI], we've achieved every single old definition. But as AI gets smarter, the goalposts keep moving.” [52:35]

Timestamps for Key Segments

[03:35] – Copyright lawsuit: $1.5B settlement & Disney-OpenAI partnership
[07:43] – Rise of AI influencers over human UGC creators
[13:00] – Emergence of “vibe coding,” non-techies building apps
[17:51] – Reasoner wrappers status
[20:30] – Virtual machines for AI agents become standard
[24:00] – AI becomes a political issue in the U.S.
[29:29] – Global regulations vs. U.S. regulatory approach
[32:05] – Success of narrow agents over general agents
[34:56] – LLM memory and persistence as a 2025 focus
[39:12] – Transition to small language models (SLMs)
[43:11] – Mixture of models approach gains traction
[47:08] – AGI might have quietly arrived: reviews of IQ tests, GDP VAL, and industry definitions

Concluding Reflection

Jordan candidly appraises the accuracy of his predictions and offers both industry insight (“this is what the smartest companies are doing”) and practical encouragement for AI professionals and businesses moving into 2026. The episode is an accessible, honest, and sometimes humorous guide for anyone seeking to keep up with AI’s dizzying pace—reminding us that with each step, the boundaries (and benchmarks) continue to shift.

Essential Call to Action:
Subscribe to the podcast, sign up for the newsletter (youreverydayai.com), and stay tuned for the upcoming 2026 Roadmap and prediction series.

For full segment breakdowns and more AI news, subscribe at youreverydayai.com.

Summary

Episode Overview

Key Discussion Points & Detailed Insights

The First Big Copyright Case Gets Settled

[03:35]

Prediction: The first major AI copyright lawsuit would be settled in 2025.
Outcome: Came true with the $1.5 billion class-action settlement against Anthropic for allegedly using copyrighted books in model training.
Impact: Signals a shift toward a "pay-to-train" model, where agreements like the Disney-OpenAI partnership ($1B for IP access) become the norm.
Jordan’s Assessment:

“This is, I think, one of the first big pieces that eventually we’re going to move to a pay-to-train model.” [04:55]

AI Influencers Begin to Replace Human UGC Creators

[07:43]

Prediction: AI influencers–virtual avatars–would start to dominate social media user-generated content (UGC).
Reality: Major brands are rolling out UGC campaigns with AI-generated influencers. Example: Lil Maela, with 3M+ followers, earning up to $10M/year, and partnerships with Samsung, Calvin Klein, and Prada.
Jordan’s Commentary:

“Probably by the end of next year, I’d say maybe 80% of what you might see online is going to be AI." [10:46]
Platforms' Role: Tools like TikTok’s Symphony Gen AD and Meta’s AI ad tools are accelerating this trend.

“Vibe Coding”—Non-Techies Build Apps on the Fly

[13:00]

Prediction: Non-technical people would be able to "build on the fly" software.
Result: True, and the term “vibe coding” (coined by Andrej Karpathy in February) now describes this phenomenon. Tools like Google’s experimental Disco, AI Studio, Gemini Canvas, and others allow anyone to create web apps or features with minimal effort.
Notable Quote:

“The people that are winning hackathons are non-technical people.” (Citing a conversation with Paige Bailey at Google) [15:49]
Stat: 70% of new enterprise apps in 2025 were built using low-code AI tools.

Reasoner Wrappers for Enterprise Knowledge

[17:51]

Prediction: Tools would emerge to wrap reasoning models with enterprise data, enhancing decision-making.
What Happened: Some movement (e.g., tools like Leta, growth in agent frameworks and observability), but the prediction was ahead of its time.
Jordan’s Take:

“Maybe just a little early on this prediction…but there is probably this new agency layer.” [18:30]

Virtual Machines (VMs) for Agents Go Mainstream

[20:30]

Prediction: VMs/virtual desktops would regain popularity as agents need “computers” to operate.
Accuracy: Correct. Major cloud platforms now supply dedicated AI agent environments, e.g., Microsoft’s Windows 365 for agents.
Memorable Moment:

“It kind of seemed like a niche prediction…But no one was talking about virtual machines. And now they’re definitely a thing.” [22:53]

AI’s Political Transformation in the U.S.

[24:00]

Prediction: AI would become entangled in politics and policy.
Validation: Clear. President Trump signed an executive order blocking states from regulating AI, making AI regulation a federal-only matter—a break from usual Republican policy.
Quote:

“There is literally no bigger way for AI to become political than at the federal level, saying: hey states, you can't make laws on this.” [25:15]
Other Notes: AI leaders, including Zuckerberg, Altman, and Apple, donated significant sums to political campaigns; U.S. signed the “AI Action Plan,” explicitly banning “woke” or ideologically biased models for federal contracts.

Global Regulation Tightens, But the U.S. Stagnates

[29:29]

Prediction: AI rules globally would grow stricter, except in the U.S.
Evidence: EU’s AI Act, Singapore’s assurance pilot, and Italy’s national law restricted or delayed access/features for major LLMs; U.S. largely remained hands-off.
Effect: Users in regulated regions face delays or limitations; Americans have almost no restrictions.

Narrow Agents Dominate, Not General Purpose

[32:05]

Prediction: Narrow, task-specific AI agents would thrive, while general-purpose agents would lag.
Reality: 100% accurate. Major agents (e.g., Salesforce Force2O, Oracle’s role-based agents, GitHub Copilot, etc.) are specialized. General agents (OpenAI, Copilot) remain clunky.
Data: Menlo Ventures reports $7.3B enterprise investment in narrow/departmental AI.

LLM Memory Becomes a Focus

[34:56]

Prediction: LLM “memory” (context, persistence) would become standard and a major upgrade.
Validation: All major models now have robust memory features, including OpenAI, Gemini, and Claude; context caching and personalization are common.
Jordan’s Analysis:

“Context plus connectors plus memory was definitely the biggest leap forward…in 2025.” [37:15]

Large Language Models (LLMs) Become Small Language Models (SLMs)

[39:12]

Prediction: Small, efficient models would surpass some older large models and see greater use.
Result: Confirmed with models like GPT-OSS 20B—just 1% the size of old GPT-4, yet equal or superior in many tasks. Also emphasized are Microsoft’s Phi, Google’s Gemma, and domain-specific mini-models.
Jordan’s Summary:

“The future of large language models is using many small domain-specific models.” [42:07]

Mixture of Models Surfaces as a New Standard

[43:11]

Prediction: Systems would begin orchestrating multiple models ("mixture of models" – distinct from "mixture of experts") for enhanced output.
Outcome: This approach emerged—example: Zoom’s ensemble system beat the “Humanity’s Last Exam.” Google’s Interactions API and startups like Area raised big funding on this premise. IDC projects 70% of top AI-driven enterprises will use multi-model architectures by 2028.
Jordan’s Forecast:

“The smartest companies are having a modular approach when it comes to their AI strategy.” [46:01]

Did AGI Happen in 2025… and No One Noticed?

[47:08]

Prediction: AGI (as defined by outperforming humans at most economically valuable work) would arrive quietly, with no public “AGI moment.”
Jordan’s Evidence:
- LLMs outperform top humans on elite tests (e.g., IQ tests, International Mathematical Olympiad).
- OpenAI’s definition: AGI means "highly autonomous systems that outperform humans at most economically valuable work.” (OpenAI charter, 2018)
- GDP VAL Test: GPT-5.2 “Thinking” model outperformed or tied expert humans on 70.9% of tasks; “Pro” model at 74.1%.
Reflection:

“On economically valuable work judged by experts… the AI did better three-quarters of the time… and no one really noticed.” [53:11] “As AI gets smarter, the goalposts keep moving.” [54:00]

Notable Quotes & Memorable Moments

On the speed of AI progress:

“This is how fast time flies in AI world. This was before vibe coding... And now, vibe coding is baked into browsers.” [13:06]
On political influence:

“It doesn’t get more political in AI than the cozying up of big tech leaders with the federal government or President Trump.” [27:43]
On narrow agents:

“There’s no one great agent out there... The agents that have been making the biggest splashes are just narrow agents, agents built around a certain vertical, a certain task.” [32:59]
On AGI’s fuzzy line:

“If you were just to look at those old definitions [of AGI], we've achieved every single old definition. But as AI gets smarter, the goalposts keep moving.” [52:35]

Timestamps for Key Segments

[03:35] – Copyright lawsuit: $1.5B settlement & Disney-OpenAI partnership
[07:43] – Rise of AI influencers over human UGC creators
[13:00] – Emergence of “vibe coding,” non-techies building apps
[17:51] – Reasoner wrappers status
[20:30] – Virtual machines for AI agents become standard
[24:00] – AI becomes a political issue in the U.S.
[29:29] – Global regulations vs. U.S. regulatory approach
[32:05] – Success of narrow agents over general agents
[34:56] – LLM memory and persistence as a 2025 focus
[39:12] – Transition to small language models (SLMs)
[43:11] – Mixture of models approach gains traction
[47:08] – AGI might have quietly arrived: reviews of IQ tests, GDP VAL, and industry definitions

Concluding Reflection

Essential Call to Action:
Subscribe to the podcast, sign up for the newsletter (youreverydayai.com), and stay tuned for the upcoming 2026 Roadmap and prediction series.

For full segment breakdowns and more AI news, subscribe at youreverydayai.com.

wavePod

Powered by Wave AI

Summary

Episode Overview

Key Discussion Points & Detailed Insights

The First Big Copyright Case Gets Settled

AI Influencers Begin to Replace Human UGC Creators

“Vibe Coding”—Non-Techies Build Apps on the Fly

Reasoner Wrappers for Enterprise Knowledge

Virtual Machines (VMs) for Agents Go Mainstream

AI’s Political Transformation in the U.S.

Global Regulation Tightens, But the U.S. Stagnates

Narrow Agents Dominate, Not General Purpose

LLM Memory Becomes a Focus

Large Language Models (LLMs) Become Small Language Models (SLMs)

Mixture of Models Surfaces as a New Standard

Did AGI Happen in 2025… and No One Noticed?

Notable Quotes & Memorable Moments

Timestamps for Key Segments

Concluding Reflection

Summary

Episode Overview

Key Discussion Points & Detailed Insights

The First Big Copyright Case Gets Settled

AI Influencers Begin to Replace Human UGC Creators

“Vibe Coding”—Non-Techies Build Apps on the Fly

Reasoner Wrappers for Enterprise Knowledge

Virtual Machines (VMs) for Agents Go Mainstream

AI’s Political Transformation in the U.S.

Global Regulation Tightens, But the U.S. Stagnates

Narrow Agents Dominate, Not General Purpose

LLM Memory Becomes a Focus

Large Language Models (LLMs) Become Small Language Models (SLMs)

Mixture of Models Surfaces as a New Standard

Did AGI Happen in 2025… and No One Noticed?

Notable Quotes & Memorable Moments

Timestamps for Key Segments

Concluding Reflection