Manis AI is a new autonomous agent that has been making waves in the AI community. Here are some of the key capabilities and demos that have been showcased:

Resume Screening

Manis was able to autonomously screen a zip file of resumes, read through each one, and provide an evaluation of the candidates. It demonstrated the ability to quickly and efficiently process large amounts of information to identify the most relevant candidates.

Property Research

When given a prompt to find a property in New York with specific criteria like a safe neighborhood and low crime rate, Manis was able to autonomously research online resources, use a virtual browser, and provide a detailed report on the top recommendations.

Stock Analysis

Manis also showcased its ability to perform stock analysis, creating a task list and then autonomously completing each task to analyze the stock and provide insights.

Drone Location Scouting

A user provided the prompt to find the best locations to fly a drone for 3D scanning near downtown Austin. Manis created a task list, used online resources and Google Maps, and then provided the top three recommended locations that were both suitable for the drone and allowed drone usage.

Game and Application Development

The community has demonstrated Manis' ability to create 3D games, animations, and even full web applications with just a single prompt. Examples include an endless runner game, a colorful animation, and an SEO audit tool.

While some have argued that Manis is simply combining existing tools, the key value it provides is in seamlessly integrating these capabilities into a single, easy-to-use agent. Manis represents an important step forward in making autonomous agents accessible and useful for a wide range of tasks.

Open AI has been making strides in the agent space as well. This week, they released new tools for developers to help others create AI agents. They released the Responses API, which allows developers to use Open AI's web search, file search, and computer use features. This means we're likely to see more AI agent tools roll out, as developers can now leverage Open AI's tools to build their own agents.

Microsoft also announced that the Responses API is now available in Azure AI Foundry, making it easier for enterprises to create their own AI agents.

Convergence AI also rolled out their "Deep Work" agent, which seems similar to the deep research tools from Open AI, Google, and Anthropic. However, to use Deep Work, you need to upgrade to their $20/month plan, and there doesn't seem to be a way to demo the capabilities before subscribing.

Another new AI agent tool is called Harvey. To use Harvey, you need to request a demo, but they do have a video showcasing its abilities, such as summarizing financial reports and comparing trends to competitors.

These new agent tools from various companies demonstrate the growing interest and capabilities in this space. While they may have some similarities, each platform offers unique features and approaches to building AI agents. Developers and businesses now have more options to leverage AI agents to automate tasks and enhance their workflows.

Google has been very active in the AI space this week, with several major announcements:

Gemini 3:

• Google released Gemini 3, their open-source language model, in Google AI Studio.
• According to ChatbotArena, Gemini 3 almost performs as well as the larger and more powerful model, DeepSeeR1.
• Gemini 3 is a 27 billion parameter model, much smaller than DeepSeeR1's 671 billion parameters, yet still highly capable.
• Gemini is also multimodal, meaning it can understand inputs from images, text, and videos.

Gemini 2.0 Flash:

• Google announced native image generation capabilities with Gemini 2.0 Flash.
• This allows users to generate, edit, and manipulate images directly within the Google AI Studio interface.
• Users can provide text prompts to create images, add elements to existing images, and more.
• The image generation is fast, with results produced in just a few seconds.

Google Deep Research:

• Google is now offering their version of deep research tools for free within the Gemini chatbot interface.
• Users can ask Gemini to research topics in-depth, and it will provide comprehensive, organized reports from various sources.
• This free deep research capability is similar to what is offered by tools like Perplexity and OpenAI's deep research.

Other Announcements:

• Google introduced Gemini Robotics, a Gemini 2.0-based model designed for robotics applications.
• They are also integrating Gemini AI into Google Calendar and Gmail to enhance scheduling and email capabilities.

Overall, Google has made significant strides in expanding the capabilities of its Gemini language model and making advanced AI tools more accessible to users. These announcements showcase Google's continued investment and innovation in the field of artificial intelligence.

Gemma 3 and Gemini 2.0 Flash from Google

• Google made their open-source language model Gemma 3 available in Google AI Studio. It performs almost as well as the larger Deep Seek R1 model.
• Gemma 3 is a 27 billion parameter model that can be run on consumer GPUs, unlike larger models.
• Google also announced native image generation with Gemini 2.0 Flash, which allows users to generate, edit, and manipulate images directly within the AI Studio interface.
• Gemini 2.0 Flash demonstrates impressive capabilities, such as adding accessories to images, creating animations, and maintaining character consistency across frames.

Google's Deep Research and Calendar Integration

• Google is now offering their version of deep research for free within the Gemini chatbot interface.
• The deep research tool can be used to research topics, generate comprehensive reports, and export the results to Google Docs.
• Google is also integrating Gemini AI into Google Calendar, allowing users to quickly check their schedule and add events conversationally.

Other AI Developments

• Perplexity released a Windows app with hotkey support for quick access.
• Grock and Perplexity can now be interacted with by simply tagging them in posts on X.
• Hunan released a new model called Hunan Turbo S, which they claim outperforms GPT-4 and Deep Seek V3 on certain benchmarks.
• Rea AI Labs open-sourced their Rea Flash 3 model, which is said to be on par with 01 Mini.
• Sakana AI generated a scientific paper that passed peer review, believed to be the first fully AI-generated paper to do so.

AI Coding Advancements

• Cursor and Winsurf continue to improve their AI-powered code generation tools with new features.
• Bolt released a Figma app that allows users to create designs in Figma and have Bolt generate the code.
• Anthropic's CEO Dario Amodei predicted that AI will be writing 90% of code within the next 3-6 months, and potentially all code within 12 months.

AI-Generated Video and Advertising

• Moon Valley claims to have created the first world-class, clean AI video model called Marry.
• Captions launched Mirage, which generates energetic, high-converting ads with AI-generated people and animations.
• Snap introduced AI video lenses powered by its own in-house generative model, available to Snapchat Platinum subscribers.

Other Noteworthy Developments

• Xbox announced a new "co-pilot" feature that uses AI to help gamers overcome roadblocks and provide strategy guidance.
• Rivian announced new self-driving features, allowing drivers to take their hands off the wheel on the freeway.
• Meta is reportedly developing its own in-house AI training chips, reducing its reliance on Nvidia.
• Apple is rumored to be working on AI-powered AirPods that can provide real-time language translation.

This week saw some exciting developments in the world of AI coding and development tools:

• Cursor: Cursor rolled out new features including themes, checkpoints, auto-fix errors, and a new navigation bar. These quality-of-life updates make it easier to use Cursor for AI-assisted coding.

• Bolt: The company Bolt released a Figma app that allows you to create designs in Figma and then have Bolt generate the code to bring those designs to life.

• Anthropic CEO: Dario Amodei, the CEO of Anthropic, stated that in the next 3-6 months, AI will be writing 90% of code, and in 12 months, AI may be writing essentially all code.

• Cursor and Winsurf: The author has been using Cursor and Winsurf extensively for AI-assisted coding, including for overhauling the Future Tools website.

• Bolt Figma App: This new app from Bolt makes it easy to generate code directly from Figma designs, streamlining the design-to-development workflow.

• Dario Amodei's Prediction: The Anthropic CEO's bold prediction about AI writing the majority of code in the near future highlights the rapid advancements in this space.

In summary, the AI coding and development landscape continues to evolve rapidly, with tools like Cursor, Bolt, and the potential for AI to write most code in the coming years. Developers should stay informed about these emerging capabilities to leverage them effectively.

This week, we saw some exciting developments in the world of AI-generated video and animation:

• Moon Valley claims to have created the first "world-class, clean AI video model" called Marry. This model is trained exclusively on licensed data and can generate landscape videos, as well as scenes with people, horses, and other elements.

• The company Captions launched "Mirage", which is designed to generate energetic, high-converting ads with AI-generated people, complete with animated body language and micro-expressions. While the visuals look impressive, the audio still has a robotic quality that is a giveaway of the AI generation.

• Snap introduced AI video lenses powered by its own in-house generative model. This allows users to add AI-generated objects and animals to their Snapchat videos, but it requires a Snapchat Platinum subscription costing $16 per month.

• Xbox showcased a new "co-pilot" feature that uses AI to help gamers overcome roadblocks in games like Minecraft and Age of Empires. The AI can provide strategies, recap past gameplay, and even control certain in-game actions.

Overall, we're seeing continued advancements in AI's ability to generate realistic-looking video and animation, though the audio quality still lags behind in some cases. These tools are becoming more accessible to both consumers and creators, opening up new possibilities for content creation and gaming experiences.

In the world of AI, the advancements are not limited to just software and models. Hardware and devices are also seeing significant progress in incorporating AI capabilities.

Meta Developing In-House AI Training Chips

Meta, the parent company of Facebook, is reportedly beginning to test its own in-house AI training chips. Currently, Meta relies on Nvidia GPUs for its AI workloads, but the company wants to reduce its dependence on Nvidia and develop its own custom chips for AI training.

Apple Rumored to be Developing AI-Powered AirPods

Rumors suggest that Apple is developing a new version of its AirPods that will feature real-time language translation capabilities. This would allow the AirPods to automatically translate conversations between different languages, similar to the functionality already available in Google's Pixel Buds.

Nvidia's GTC Conference

Next week, Nvidia is hosting its annual GTC conference in San Jose. This event is a significant showcase for the company's latest advancements in AI hardware and software. If you register to watch the conference virtually for free, you'll be entered to win an Nvidia RTX 5090 GPU signed by Nvidia's CEO, Jensen Huang.

Rivian's New Self-Driving Features

The electric vehicle manufacturer Rivian has announced new self-driving features for its vehicles. Drivers can now take their hands off the wheel, and the car will maintain a safe distance from other vehicles, automatically change lanes when the turn signal is activated, and handle other autonomous driving tasks.

These developments in AI-powered hardware and devices demonstrate the continued integration of AI technology into our everyday lives, from transportation to personal electronics. As the capabilities of these AI-enabled systems continue to evolve, we can expect to see even more innovative applications in the near future.

The rapid advancements in AI technology, particularly in the realm of autonomous agents and AI-powered tools, have been truly remarkable. From the launch of Manis AI, which showcases the ability to automate tasks like resume screening and property research, to the impressive capabilities of OpenAI's Gemini 2.0 model in generating images and animations, the potential of these AI agents is undeniable.

Google has also made significant strides, with the release of Gemma 3 and the integration of AI-powered features into their suite of products, including Google Calendar and Gmail. The ability to leverage these AI tools for tasks like deep research, code generation, and even gaming assistance is a testament to the transformative power of this technology.

As the AI landscape continues to evolve, it's clear that the future will be heavily influenced by these intelligent agents. While some challenges and limitations remain, the pace of innovation suggests that we are on the cusp of a new era where AI will play an increasingly integral role in our daily lives and business operations.

The key takeaway is that the time for AI agents to take on a significant portion of our work has arrived. By embracing these tools and leveraging their capabilities, we can unlock new levels of efficiency, productivity, and creativity, ultimately positioning ourselves for success in the rapidly changing digital landscape.