AI is moving beyond bigger models and better chatbots. Google is bringing multimodal AI directly onto devices. Georgia Tech is building an AI degree for working professionals. And OpenAI is training agents on complex, real-world business workflows. Together, these moves reveal where AI is heading next: into our devices, careers, and everyday work.

In today’s post:

• Google just shrank multimodal AI

• Georgia Tech’s new AI degree reveals a bigger shift

• OpenAI’s agents are learning actual jobs

SPONSORED BY

What’s Trending Today

LAUNCH

AI search is moving from the cloud to your device

Google DeepMind just released EmbeddingGemma 2. It understands text, code, images, video, and audio. But that isn’t the interesting part. It does this with just 740 million parameters. That changes where useful AI can actually run.

• EmbeddingGemma 2 puts different media into one shared embedding space. That means text can find audio, images can find video, and voice can search media.

• The model is designed for consumer hardware, not datacenters. Quantized text-only weights can use around 191MB of active RAM.

• Developers also don’t need the entire model. Text workloads need roughly 270M parameters, while vision and audio remain optional.

• Storage gets cheaper too. Developers can shrink 768-dimensional embeddings down to 128 dimensions, cutting storage requirements by up to six times.

• The context window has grown to 8K tokens. One request can handle 29 images or 58 video frames.

• Code retrieval received a notable upgrade. Its MTEB Code score jumped from 68.76 to 78.68.

• The bigger opportunity is private, offline RAG. Apps can search personal files and media without sending everything elsewhere.

The model itself isn't the biggest story. The important shift is where AI happens. Cloud models made intelligence accessible. Small models could make intelligence ambient. Your files stay local. Search becomes instant. Apps understand multiple formats. That opens an interesting question: What gets built when AI stops needing the cloud?

TECHNOLOGY

AI education is moving beyond computer science

Georgia Tech just announced a new online AI master’s. But it isn’t designed only for AI engineers. It targets professionals who already know another field. Business. Healthcare. Decision-making. Human-AI interaction. That distinction matters more than the degree itself.

• The Online Master of Science in Artificial Intelligence launches in 2027. It targets working professionals who want applied AI expertise.

• Students won’t just learn how AI works. They’ll study where it belongs, when to use it, and what responsible deployment requires.

• Six concentrations are planned initially. They include business, health, AI theory, predictive AI, decision-support, and human-AI interaction.

• Ethics won’t sit outside the technical curriculum. Privacy, safety, accountability, and governance are built into the program.

• Every student will complete an AI practicum. That forces classroom knowledge into actual organizational problems.

• The interdisciplinary structure signals something bigger. AI knowledge is becoming useful across professions, not just computing roles.

• Georgia Tech also plans to evolve the curriculum. New concentrations can emerge as AI changes industries and workforce needs.

For years, AI education focused on building models. The next wave may focus on applying them well. That requires something different. Domain expertise becomes more valuable when paired with AI fluency. The advantage may not belong to AI specialists alone. It may belong to people who understand two worlds deeply.

RESEARCH

AI agents are moving from clicking buttons to understanding workflows

OpenAI just published new research with Ironclad. The goal sounds simple. Teach AI agents to complete real contracting workflows. But these tasks require more than computer control. Agents must remember rules, handle exceptions, and verify results.

• OpenAI and Ironclad created 11 tasks across legal and procurement work. Each represents work an experienced user might spend 30–40 minutes completing.

• The tasks weren’t simple button-clicking exercises. Agents configured agreements, approval processes, procurement rules, and reusable legal clauses.

• Every task had between 8 and 50 evaluation criteria. Success meant completing the whole workflow correctly, not just individual steps.

• OpenAI then created synthetic training tasks around representative workflows. Reinforcement learning helped models improve through practice and feedback.

• GPT-6 Astra scored 55% across the evaluation. GPT-5.6 Sol reached 41.6% under its strongest tested reasoning setting.

• Astra also completed simulated attempts much faster. Estimated time fell from 37 minutes to 19.2 minutes per attempt.

• There’s still plenty of room to improve. Even Astra missed almost half the evaluation criteria on average, reinforcing why human oversight still matters.

The interesting part isn’t the 55% score. It’s how these models are being trained. Software companies understand their workflows better than model labs. Model labs understand training better than software companies. Put those together, and agents learn actual professional work. That could become an important competitive advantage. The best AI may learn directly from how businesses operate.

Free Guides

My Free Guides to Download:

🚀 Founders & AI Builders, Listen up!

If you’ve built an AI tool, here’s an opportunity to gain serious visibility.

Nextool AI is a leading tools aggregator that offers:

• 500k+ page views and a rapidly growing audience.

• Exposure to developers, entrepreneurs, and tech enthusiasts actively searching for innovative tools.

• A spot in a curated list of cutting-edge AI tools, trusted by the community.

• Increased traffic, users, and brand recognition for your tool.

Take the next step to grow your tool’s reach and impact.

Reach 150,000+ READERS:

Expand your reach and boost your brand’s visibility!

Partner with Nextool AI to showcase your product or service to 140,000+ engaged subscribers, including entrepreneurs, tech enthusiasts, developers, and industry leaders.

Ready to make an impact? Visit our sponsorship website to explore sponsorship opportunities and learn more!