Good morning, AI enthusiasts. Google’s recent AI moves may have had Sam Altman slamming the ‘Code Red’ button, but the AI giant’s latest releases are rising to the challenge.

OpenAI’s counter to Nano Banana Pro is officially here, with a long-overdue GPT-image upgrade that vaults the company back to the frontier of image leaderboards (for now).

In today’s AI rundown:

• OpenAI counters Nano Banana Pro with new Images upgrade

• HubSpot CEO Dharmesh Shah on SEO for the AI era

• Quickly iterate on Sora videos with a simple automation

• Google, MIT study finds pitfalls in multi-agent systems

• 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

OPENAI

🚀 OpenAI counters Nano Banana Pro with new Images upgrade

Image source: OpenAI

The Rundown: OpenAI just released GPT Image 1.5, a major update to ChatGPT’s image generator that creates visuals up to 4x faster, improves text rendering, and maintains consistency across edits — arriving as an answer to Google’s recent creative momentum with Nano Banana Pro.

The details:

• Image 1.5 brings significantly upgraded generation speeds, with the model also now able to preserve faces, lighting, and composition across edits.

• Text rendering also gets a big improvement, with 1.5 handling long content, infographics, and varied text sizes compared to GPT-image-1’s rampant issues.

• The new model moves to first place on both Artificial Analysis and LM Arena’s text-to-image and editing leaderboards.

• OpenAI also released a new dedicated creative panel, joining the typical chat-based workflow to offer users quick-start templates and curated style options.

Why it matters: Despite GPT-image-1 being a viral success at the time, it doesn’t take long to fall far behind the frontier curve — making this new upgrade very overdue. 1.5 delivers on the benchmarks and, like the recent GPT-5.2 release, brings OpenAI at least on par with Google’s buzzy releases heading into the new year.

TOGETHER WITH UIPATH

👋 Goodbye Chaos. Hello Orchestration.

The Rundown AI: UiPath 2025.10 upgrades how work moves across the enterprise with stronger coordination and clearer oversight. It sharpens the way teams manage processes that span systems, people, and intelligent agents — pushing automation into a faster, more intelligent era.

What’s in the newest UiPath release:

• Expanded Maestro upgrades for agentic orchestration

• Smarter AI across documents, testing, and workflows

• Faster build cycles with streamlined Studio and Autopilot tools

Explore UiPath 2025.10 and the features people are talking about.

ROWAN X DHARMESH

🎙️ HubSpot CEO Dharmesh Shah on SEO for the AI era

Image source: The Rundown

The Rundown: We sat down with HubSpot CEO Dharmesh Shah for an exclusive interview on how the traditional SEO playbook is changing with LLMs and how to prepare for a world run by agents.

On SEO’s shift to AEO (Answer Engine Optimization):

Dharmesh: We’re moving from an SEO mindset (solving for Google) to an LLM mindset (solving for AI). Ask, “How can I take the ideas and content that I have and ‘translate’ it to make it more easily consumable by LLM vs. search engines and humans directly?”

On the hidden risk of low-quality AI content:

Dharmesh: The worst that can happen is actually not zero return. The worst that can happen is negative return. Because if you build a reputation online, in the algorithm’s mind that you are crappy content, not trustworthy... it’s hard to dig yourself back out of that hole.

On the bottlenecks for AI agents:

Dharmesh: If you think of agents as teammates... You wouldn’t just hand them a computer and access to all the internals of the organization and say, “Here’s what we want done, go do it.” Agents need to be trained, tested, and need to have the equivalent of performance reviews.

Catch Rowan’s full interview with Dharmesh on YouTube, Spotify, or Apple Podcasts.

AI TRAINING

📹 Quickly iterate on Sora videos with a simple automation

The Rundown: Learn how to generate bulk Sora videos automatically using a simple Make.com automation that watches your Google Sheet prompts and uploads completed videos to Google Drive.

Step-by-step:

1. Duplicate this Google Sheet Template, then sign into Make.com and create a new scenario with the Google Sheets ‘Watch Rows’ trigger

2. Add a Sora module with your API key (create one if needed + verify org), then create a Google Drive folder and add ‘Upload a File’ module

3. Add a Google Sheets ‘Update a Row’ module that updates the drive link column with the ‘Web View link,’ then write prompts in your sheet with ID numbers and click ‘run once’ in Make

4. Within a few minutes, your Google Sheet will update with links to completed videos in your Google Drive

Pro tip: Instead of waiting every 15 minutes for the automation to fire, you can set up Google Sheets to fire a webhook straight to Make every time a row is updated.

PRESENTED BY AMAZON

⚙️ Automate UI workflows with Amazon Nova Act

The Rundown: Nova Act just launched — designed for developers to build, deploy, and manage #NormcoreAgents to automate tasks so you don’t have to.

How does it work?

• Prototype agents in the playground

• Write, test, and debug agents directly in your IDE

• Deploy agents to production

Try it for free in the Nova Act Playground.

AI RESEARCH

😵 Google, MIT study finds pitfalls in multi-agent systems

Image source: Reve / The Rundown

The Rundown: Researchers from Google and MIT published a new study testing whether throwing more AI agents at problems improves results, finding that performance swung wildly depending on the structure of the task.

The details:

• The team ran 180 experiments across models from OpenAI, Google, and Anthropic, using the same prompts and token budgets.

• Financial analysis tasks split across agents saw an 81% improvement, while Minecraft tasks requiring step-by-step work degraded by up to 70%.

• When a single agent already hit 45% accuracy on a task, adding more typically led to worse performance — with multiple agents eating through tokens quickly.

Why it matters: The agentic hype is pushing companies and users towards complex multi-agent workflows, but this research may show that more isn’t always better. For many enterprise tasks that require step-by-step reasoning, a well-designed single agent may outperform an elaborate system at a fraction of the cost.

QUICK HITS

🛠️ Trending AI Tools

• 🌌 ChatGPT Images - OpenAI’s upgraded image generation system

• 🎥 Wan 2.6 - Alibaba’s new native multimodal model for image and video

• 📪 CC - Google Labs’ experimental AI productivity agent in Gmail

• 🔊 SAM Audio - Meta’s new model for segmenting sounds using text prompts

📰 Everything else in AI today

Sonatype just launched Guide — ensuring your AI assistants select the right open-source components so you can spend less time debugging. Watch the unboxing webinar to learn more.*

OpenAI released FrontierScience, a new benchmark to evaluate research-level reasoning in scientific research tasks, with GPT 5.2 taking the top spot in testing.

Google Labs rolled out CC, an experimental AI assistant powered by Gemini that connects to user’s Gmail, calendar, and files to send personalized morning summaries.

Black Forest Labs released FLUX.2 [max], the startup’s new top image model featuring upgraded editing and the ability to create visuals from real-time web data.

AI2 introduced Molmo 2, an SOTA open model family for analyzing video by following objects across frames, counting events, and locating precise moments within footage.

*Sponsored Listing

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Bo in Cincinnati, OH:

“I used Claude to help my law firm pivot to an AI-native firm, using it for annual planning, project management, workload integration, and more. Recently I created AI agents to serve as my CFO, CTO, and content creator. With my digital CTO, we started building our own small language model trained on our proprietary data to scale our law practice.”

How do you use AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: Nvidia’s powerful open AI model play

• Read our last Tech newsletter: The AI boom’s phone problem

• Read our last Robotics newsletter: 1X’s home humanoid gets a factory job

• Today’s AI tool guide: Iterate on Sora videos with a simple automation

• RSVP to next workshop @ 4PM EST Friday: Google NotebookLM For Work

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown