New Google Gemini & Search Upgrades

PLUS: 5 SaaS Ideas to build with Google Veo 3

Welcome back to Daily AI Skills.

Here’s what we are covering today:
1. Google Gemini and Search Upgrades
2. Devstral by Mistral
3. AI System 'Robin' Discovers New Treatment for Vision Loss

+ 5 SaaS Ideas to build with Google Veo 3

Google Unveils Major Gemini & Search AI Upgrades

Google has announced major upgrades to its Gemini and Gemma AI models at Google I/O, along with a suite of new AI-powered search, shopping, and agent tools.

Gemini / Models:

  • Gemini 2.5 Pro and Flash received significant performance boosts. Pro now leads in benchmarks and Arena rankings, while Flash has improved in capability without sacrificing speed.

  • A new “Gemini 2.5 Deep Think” model, focused on advanced reasoning, is now in testing and shows state-of-the-art performance in math, coding, and multimodal tasks.

  • Gemma 3n, a lightweight, mobile-first open model, has launched in preview. Despite its size, it performs on par with much larger models like Claude 3.7 Sonnet and is optimised for on-device use.

  • Gemini Live is now available for free to all users, enabling camera and screen sharing. More personalised features will roll out soon.

Search / Agents:

  • AI Mode in Google Search—powered by Gemini 2.5—is now available to all U.S. users. It introduces features like Deep Search and integrated Gemini Live.

  • New tools include virtual try-on for shopping, an AI shopping assistant, and Search Live for real-time, multimodal voice interactions.

  • Jules, Google’s coding assistant, is now in public beta. It can handle developer tasks in the background and directly integrate with codebases.

  • A new Agent Mode for both Search and Gemini can autonomously complete up to 10 tasks at once on a user’s behalf.

OpenAI Unveils Codex: AI Agent for Developers

French AI startup Mistral has teamed up with All Hands AI to launch Devstral, a cutting-edge open-source coding model that outperforms much larger competitors on real-world software engineering tasks — and it's lightweight enough to run on a laptop or single GPU.

Here’s what stands out:

  • Devstral surpasses all open-source models (and some closed ones) on benchmarks like SWE-Bench Verified, which tests performance on real GitHub issues.

  • It’s designed for agentic development, meaning it can understand entire codebases, make edits, and tackle complex programming challenges.

  • The model is small enough to run locally on consumer hardware like a Mac.

  • It’s released under the Apache 2.0 license, making it freely usable for both personal and commercial projects.

Mistral also hinted that a larger, more powerful agentic coding model is on the way soon.

AI System 'Robin' Discovers New Treatment for Vision Loss

FutureHouse has announced a major milestone for its multi-agent AI system, Robin, which has discovered a promising new treatment for dry age-related macular degeneration (dAMD), a leading cause of blindness.

Key details:

  • Robin autonomously handled the entire research pipeline—hypothesis generation, experimental design, data analysis, and figure creation—while humans carried out the lab work.

  • The AI identified ripasudil, a glaucoma drug already approved in Japan, as a potential treatment for dAMD. Lab tests confirmed its effectiveness.

  • Next week, FutureHouse will open-source Robin’s code and data, along with its supporting AI agents: Crow (literature search), Falcon (in-depth review), and Finch (data analysis).

5 SaaS Ideas to build with Google Veo 3

📩Forward it to people you know who are keeping pace with the changing AI world, and stay tuned for the next edition to stay ahead of the curve!