- Everyday AI
- Posts
- Ep 676: 2025 AI Roadmap Rewind Part 2 - Human vs Machine, AI Models Shrink, and AGI No One Noticed
Ep 676: 2025 AI Roadmap Rewind Part 2 - Human vs Machine, AI Models Shrink, and AGI No One Noticed
Our AI Roadmap Rewind, ChatGPTās App Store launches, Lovable raises $330 million, Amazon makes big AI switch and more
Everyday AI
December 18, 2025
š Subscribe Here | š£ Hire Us To Speak | š¤ Partner with Us | š¤ Grow with GenAI
Sup yāall! š
Real talk, this is an important one to listen to, as it kinda serves as a āState of AIā and is great for beginners and experts alike.
Did we crush it or miss completely?
(Hit us with a reply to lettuce know.)
āļø
Jordan
Outsmart The Future
Today in Everyday AI
8 minute read
š Daily Podcast Episode: As 2025 wraps up, we revisit our AI predictions to see what actually came trueāand what everyone missed. Give todayās show a watch/read/listen.
šµļøāāļø Fresh Finds: Leaks show more on Claude agents, Gemini 3 Flashās video uploads, GPT-5.2 Codex updates and more. Read on for Fresh Finds.
š Byte Sized Daily AI News: ChatGPTās App Store launches, Lovable raises $330 million, Amazon makes big AI switch and more. Read on for Byte Sized News.
šŖ Leverage AI: A lot of AI predictions are noise. Weāve spent hundreds of hours each year on ours. Today, we break down the results and audit ourselves. Keep reading for that!
ā©ļø Donāt miss out: Miss our last newsletter? We covered: Google Drops Gemini 3 Flash, Amazon in talks to invest $10 billion to OpenAI, Bernie Sanders calls for stop on AI and more Check it here!
Ep 676: 2025 AI Roadmap Rewind Human vs Machine, AI Models Shrink, and AGI No One Noticed
Did AI end up being a political force this year? š¤
Did AI kill start to kill off influencers?
Did we achieve AGI in 2025?
We take our AI predictions seriously. And in January of this year, we made some kinda hot-take predictions about the above and more.
How did we do?
Tune in for Part 2 of our 2025 AI Roadmap Rewind as we give a State of AI and see if we were your AI BFF or if we led you astray.
Also on the pod today:
ā¢ $1.5B Anthropic copyright settlement šø
ā¢ AI avatars replacing influencers š¤
ā¢ Trump blocks state AI laws šŗšø
Itāll be worth your 47 minutes:
Listen on our site:
Click to listen
Subscribe and listen on your favorite podcast platform
Listen on:
Hereās our favorite AI finds from across the web:
New AI Tool Spotlight ā AI Gateway Uses popular AI models in your code, without needing to manage API keys or external accounts, Shadow Understands What Was Said, and ShownāFrom Every Meeting, MujoAI is an AI agent for eācommerce listing content
AI Job Market ā AI-exposed jobs are growing fasterāat least for now.
AI In Love ā A woman in Japan married her AI boyfriend. It raises big questions about where love and technology are headed.
Gemini 3 Flash ā Gemini 3 Flash lands across Search, apps, and dev tools
Claude Task Mode ā Claude might soon become a hands-on task agent with live planning and artifacts
Gemini 3 Flash Video Analysis ā Quick AI video analysis draws buzz, praise, and backlash in real-world use
OpenAI Codex ā Codex hints at Caribou, a major GPTā5.2ābased coding upgrade.
Gemma 3 ā Googleās new Gemma 3 270M packs strong instruction-following and fast fine-tuning into a tiny 270M model, able to run on phones, browsers, and Raspberry Pi. Hereās how.
Salesforce Agreement ā Salesforce moves to acquire Qualified, bringing agentic AI marketing into Agentforce
AGI Rebranded ā Big Tech drops āAGI,ā invents safer-sounding intelligence instead
1. Micron Rockets on AI Memory Boom š
Micron stock surged more than 12% today after the chipmaker delivered a massive earnings beat and forecast far stronger demand tied to AI and data centers, signaling that this cycle is moving faster than Wall Street expected.
The company said it is effectively sold out of key memory products, raised capital spending plans, and now expects the market for high-bandwidth memory used in AI systems to reach $100 billion by 2028 as supply struggles to keep up. Analysts rushed to upgrade the stock, with some calling the results among the strongest in U.S. semiconductor history outside of Nvidia.
2. Amazon reshuffles its AI leadership š¤
Just weeks after its major AWS conference, Amazon is reorganizing its AI leadership as it tries to close the gap with rivals like Google, Microsoft, and OpenAI, according to Reuters.
Cloud veteran Peter DeSantis will take charge of a newly expanded group that brings AI models, custom chips, robotics, and quantum computing under one roof, while Alexa AI chief Rohit Prasad exits at the end of the year. CEO Andy Jassy framed the shift as a turning point for technologies that will shape future customer experiences, as Amazon pushes harder to show it can compete at the top tier of AI.
3. Lovableās Funding Frenzy Heats Up āØļø
According to Reuters, Stockholm-based AI startup Lovable just raised $330 million in a funding round that values the company at $6.6 billion, a rapid jump that underscores how fast money is chasing agentic AI right now.
The round was led by Alphabetās CapitalG and Menlo Ventures, with backing from Nvidia and other enterprise tech players, nearly quadrupling Lovableās valuation in under six months. The company builds tools that let people create full software products using plain language, tapping into a surge of interest in AI-driven coding across tech and business. Lovable plans to use the funds to scale infrastructure and enterprise features as it cements a spot among Europeās most valuable private tech firms.
4. Shopify product chief jumps to OpenAI š¤
OpenAI is bolstering its push to make ChatGPT a central computing platform by hiring longtime Shopify product leader Glen Coates to run its app ecosystem.
Coates, who helped shape Shopifyās core software and app marketplace, says his mandate is to turn ChatGPT into an operating system built around AI at its center, a sign of how seriously OpenAI is moving beyond chat into everyday digital infrastructure. The move comes amid heavy executive turnover at Shopify and as OpenAI eyes massive new funding to scale its models and developer tools.
5. ChatGPT Opens Its App Store š„
OpenAI has just started accepting app submissions for ChatGPT, marking an early but significant step in turning the chatbot into a broader platform in 2025.
Developers can now build and publish chatānative apps that appear inside ChatGPT, discoverable through a new ināapp directory and callable directly during conversations. OpenAI says the goal is to let apps feel like a natural part of chat while giving users control over privacy and data sharing. For developers, monetization is limited for now, but OpenAI is clearly signaling bigger business opportunities ahead as this app ecosystem takes shape.
6. AI Jitters Rock Wall Street š
Stocks slid Wednesday as renewed anxiety about the fast-growing AI trade hit markets, sparked by a Financial Times report questioning funding for a massive Oracle data center project, a claim the company strongly denied.
Even so, uncertainty around Oracleās AI infrastructure plans sent its shares down 5.4 percent and helped pull major chip and power stocks lower, dragging the Nasdaq and S&P 500 with them. The sell-off shows how tightly investor confidence in the broader market is now tied to continued AI expansion, from cloud computing to data center energy demand.
We all expected AGI to look like a sci-fi movie. š±
Sam Altman defined AGI seven years ago as systems that outperform humans at "most economically valuable work."
Well, the new GDPVal benchmark on GPT-5.2 Thinking kinda just confirmed AI now tie or beat human experts 74.1% of the time on actual payroll tasks like financial forecasting.
Seems like we kinda crossed the imaginary AGI finish line?
But nobody clapped. Or noticed. Just like we told you back in January.
While you were waiting for a robot to wake up, your competitors started automating elite knowledge work using "boring" files like Excel.
So on today's show, we are grading our own homework from Part 2 of our 2025 AI Roadmap Rewind to see what we got right, what we got wrong, and exactly what it means for your strategy next year. You can watch/listen to Part 1 here.
(And if you wanna go watch/listen to the original 5-part series, you can do that here. Theyāre kinda quick. Promise.)
Letā get auditing. š
Prediction #12: The first big copyright settlement āļø
The Prediction: We said a major AI company would settle a class-action copyright lawsuit to avoid a court ruling.
The Receipt: Anthropic just agreed to a massive $1.5 billion settlement in the Authors vs. Anthropic class action suit regarding training data.
The Verdict: We nailed it. The "pay-to-train" era is officially here.
Prediction #11: AI influencers kill human UGC š¤³
The Prediction: We claimed AI avatars would start replacing human creators for user-generated content campaigns.
The Receipt: Virtual influencers like Lil Miquela are now earning $10 million annually working with some of the biggest brands in the world, and TikTok's Symphony tools let brands generate ads without booking human talent.
The Verdict: Unfortunately true. Yahoo Finance reports this trend is set to wipe out 80% of the human influencer industry.
Prediction #10: "Vibe Coding" democratizes software šļø
The Prediction: We said non-techies would build software on the fly.
The Receipt: We kinda called āvibe codingā a whole month before anyone even coined the term. Andrej Karpathy coined "vibe coding" in February, and now reports say up to 70% of new enterprise apps are built with low-code AI tools.
The Verdict: Spot on. Your marketing team is now your engineering team.
Prediction #9: Reasoner wrappers emerge š§
The Prediction: We thought tools would wrap reasoning models in enterprise data to drive decisions.
The Receipt: Platforms like Letta have become essential for parsing hidden thought processes, though this space is still early.
The Verdict: Right direction, but slightly early on this one. This will be a multi-billion dollar industry by 2026. Plan accordingly.
Prediction #8: Virtual Machines make a comeback š»
The Prediction: We said AI agents would need their own computers, bringing VMs back in style.
The Receipt: Microsoft announced Windows 365 for Agents at Ignite, giving AI its own cloud PC to do actual work.
The Verdict: Confirmed. Agents need a desk and a computer just like humans do.
Prediction #7: AI becomes deeply political šļø
The Prediction: We warned AI would get entangled in policy conflicts.
The Receipt: The White House just signed an executive order blocking states from regulating AI to compete with China. Doesnāt get much more political than that, does it?
The Verdict: Sadly accurate. AI is now a central wedge issue in federal versus state power struggles.
Prediction #6: Global regulations tighten (except in the US) š
The Prediction: We said the world would crack down while the US stayed loose.
The Receipt: The EU AI Act is fully enforced with huge fines, while the U.S. has done almost nothing to restrict innovation.
The Verdict: Correct. Your AI strategy now needs a different playbook for every continent.
Prediction #5: Narrow agents dominate general ones š
The Prediction: We said general-purpose agents would flop, but specific ones would win.
The Receipt: OpenAIās "Operator" came and went and general use agents are kinda meh. But narrow agents like Claude Code and GitHub Copilot are printing money in specific verticals.
The Verdict: Validated. Don't buy a generalist; hire a specialist.
Prediction #4: Memory becomes the main character š§
The Prediction: We said LLM memory would be the year's biggest technical focus.
The Receipt: All three major models (GPT, Claude, Gemini) launched long-term memory in 2025 after our predictions, and context caching is now a standard API feature.
The Verdict: 100% correct. Context is the new gold.
Prediction #3: Small models beat big ones š„
The Prediction: We claimed efficiency would trump size.
The Receipt: OpenAIās open-source 20 billion parameter model (GPT-OSS) is outperforming the 2 trillion parameter models from last year.
The Verdict: A landslide win. You are likely overpaying for compute you don't need.
Prediction #2: Mixture of Models becomes standard š
The Prediction: We said systems would bundle multiple models together.
The Receipt: Zoom achieved the highest score on "Humanity's Last Exam" by duct-taping off-the-shelf models together instead of training one big one. Also, big platforms like Airia are cashing in on the mixture of models approach.
The Verdict: Confirmed. The best model is actually three models in a trench coat.
Prediction #1: AGI is achieved (and nobody cares) š¤·
The Prediction: We said AGI would arrive without a cinematic moment.
The Receipt: GPT-5.2 Thinking now ties or beats human experts on 74% of economically valuable work in the GDPVal benchmark.
The Verdict: It happened. We just didn't notice because it showed up as a spreadsheet and the definition of AGI is changing more than the models themselves.
Reply