Taco Bell drive-thrus to MIT’s SEAL. Plus A16z’s top AI apps

Date: 30-Aug-2025

Hey AI enthusiast,

And just when you think you’ve seen it all…there’s always One More Thing in AI.

In this edition:

• When AI Meets Drive-Thrus: Taco Bell and McDonald’ …

• When AI Learns to Teach Itself- MIT SEAL Approach

• AI Apps Find Their Groove, Originals Shine, Ecosys …

• More News scoured from X.com

  • Nano Banana is here and shows why Google is not ou …

  • First flying car is here!

• References

  • Do you want to reach a high-quality audience?

You are going to enjoy this edition!

Best,

Renjit

PS: Founders- In case you want to prototype and test AI automation /Agentic AI deployment in your business, schedule an appointment here:

When AI Meets Drive-Thrus: Taco Bell and McDonald’s Reboot Strategy

Ever placed an odd drive-thru order and thought, “Even humans keep me waiting”? Now imagine a robot hearing that and rethinking the whole plan.

Highlights:

• 🍴 The fast food world is revisiting its AI drive-thru decisions. Technology promised smoother service. Now, it’s proving glitchy.

• Chatty AI systems at Taco Bell’s drive-thrus rolled out across 500+ locations. Customers ran into delays, misunderstandings, or were just…skeptical.

• One prank-order went viral: someone asked for “18,000 cups of water.” The AI froze and people noticed.

• Some customers complained online. Others joked about it. Even the exec in charge, Dane Mathews, said the tech “sometimes lets me down, but sometimes surprises me.”

• 🤝 Taco Bell is now evaluating where AI fits best. On busy days or packed lines, they may let humans handle orders.

• AI isn’t going away at Taco Bell yet- they’re gathering order data and planning where human back-ups should step in.

• McDonald’s paused its own AI drive-thru tests last year. They’re now evaluating new tech partners to make it work better.

• Taco Bell’s parent company, Yum Brands, teamed up with Nvidia to build better voice AI and computer-vision tools to run more smoothly.

• “Byte by Yum,” the AI platform, is already helping with things like inventory, kitchen scheduling, and ordering tech across many locations. But it still needs smart supervision.

So What?

It’s clear: AI can talk the talk, but it’s humans who still get things done. Drive-thrus aren’t fully ready for a robotic takeover. For now, a human voice (and a human heart) still matters when you’re just trying to order a taco.

Start learning AI in 2025

Keeping up with AI is hard – we get it!

That’s why over 1M professionals read Superhuman AI to stay ahead.

• Get daily AI news, tools, and tutorials

• Learn new AI skills you can use at work in 3 mins a day

• Become 10X more productive

Sign up and start mastering AI

When AI Learns to Teach Itself- MIT SEAL Approach

Imagine an AI that doesn’t wait for updates. It learns on its own and keeps evolving.

What’s in Store

The breakthrough

Researchers at MIT created SEAL, a system that helps language models train and adapt themselves. The model generates its own training prompts and retraining steps to absorb knowledge permanently.

Why it matters

Traditional fine-tuning or prompting often falls short when models face new tasks. SEAL teaches the model how to restructure input and learn more effectively, which is crucial for dynamic enterprise use cases.

How it works

SEAL runs two loops. In the inner loop, the model makes a small update using a “self-edit.” In the outer loop, it tests if that update improved performance. Good edits get rewarded. Over time, the model becomes better at teaching itself.

A clever twist

Instead of a single model doing everything, SEAL can use a teacher–student setup. A teacher model crafts edits, and a student model applies them. This split makes enterprise use more scalable.

Real-world lifts

• On knowledge tasks, SEAL lifted accuracy to 47%, beating even GPT‑4.1 in creating useful synthetic training examples. 
• On few-shot reasoning puzzles (ARC), SEAL scored 72.5%, compared to 20% without reinforcement learning and 0% under standard prompting.

Enterprise edge

Human-generated data might run out. SEAL’s self-generated data could keep models improving, especially in rare or complex domains. This makes AI agents more adaptive and capable over time.

Practical caveats

SEAL can suffer from “catastrophic forgetting” if it over-writes older knowledge. The fix: use a hybrid memory strategy- reserve evolving data for retrieval, but bake long-term logic into the model itself.
Also, real-time self-edits are costly and slow. A more viable approach is batching updates, say, every few hours or once a day, to stay efficient.

So What?
SEAL shows how AI can move from static tools to self-improving systems. For startups building smart agents or ever-changing services, this could be a game changer. The future favors modelsthat can teach themselves.

AI Apps Find Their Groove, Originals Shine, Ecosystem Settles: A16Z report on AI

The generative AI app market is settling into a rhythm. A wave of original products is stepping up as the novelty rush fades.

Key Insights & Context

• After two and a half years of tracking real-world usage, A16z sees steadying growth in both web and mobile Gen AI apps. Web newcomers dropped to 11 from 17 last cycle, signaling less churn. Mobile, meanwhile, welcomed 14 new apps, thanks to app store cleanups that removed ChatGPT copycats.

• Google now claims four separate spots on the web list Gemini, AI Studio, NotebookLM, and Google Labs thanks to better tracking of each domain. Gemini ranks #2 behind ChatGPT and pulls about 12% of its web traffic.

• AI Studio is now in the top 10; it’s a sandbox for developers to build using Gemini models including multimodal tools like Gemini Live.

• NotebookLM, once part of Google Labs, now stands alone at #13. It’s held steady since going viral nearly a year ago, with only a modest summer dip.

• Google Labs appears at #39 on the list, boosted by a more than 13% spike in traffic after launching the Veo 3 video model in May 2025. Labs also hosts tools like Doppl (virtual try‑on), Portraits (AI coaching), and Project Mariner (agentic browser).

• On mobile, Gemini lands at #2 behind ChatGPT but closes the gap, reaching nearly half ChatGPT’s MAUs, with 90% of that traffic on Android.

• ChatGPT still dominates both platforms, but rivals like Google, Grok (from X, formerly Twitter), and Meta are beginning to narrow the lead.

Video & Coding Apps

• Video creation tools are climbing fast. Google Labs’ launch of Veo 3, a next-gen video model, gave it a 13% traffic bump in one month.

• Mobile challengers like PolyBuzz and Pixverse jumped from the Brink List into the top 100. They prove video-first apps can find traction quickly when they offer simple, creative outputs.

🎥 • Coding-focused sandboxes are also gaining ground, with AI Studio ranking high. They let builders experiment directly with multimodal tools like Gemini Live, lowering the barrier to entry for developers.

• Both segments point to a growing market for make-something apps, not just chat-about-something apps.

Emerging Apps & Market Behavior

• The “Brink List” tracks ten apps just outside the top 100—five for web, five for mobile. From that list, Lovable soared into the web top 100 at #22, while mobile apps PolyBuzz and Pixverse also graduated.

• The data shows a shift from early chaos toward measured growth. Transparency and tracking are improving, which helps founders see where real usage lies.

Strategic Implications for Founders

The generative AI market is moving from sprint to marathon. The winners now are the apps that offer real value, carve a niche, and ride authentic growth and not flash. If you’re building the next breakout AI product, focus on substance.

More News scoured from X.com

Nano Banana is here and shows why Google is not out of the AI game, yet

First flying car is here!

References

1. ZDNet – AI is returning to Taco Bell and McDonald’s drive-thrus: Will customers bite this time?

    

2. VentureBeat – Beyond static AI: MIT’s new framework lets models teach themselves

    

3. Andreessen Horowitz (a16z) – The Top 100 Gen AI Consumer Apps – 5th Edition

    

Do you want to reach a high-quality audience?

To sponsor this Newsletter and gain customers. This newsletter is subscribed to by solopreneurs, startup founders and business leaders: