Google has officially expanded its AI-powered developer toolkit with the launch of Gemini CLI Extensions, a new system that allows third-party integrations directly into its command-line interface (CLI) tool.

Unveiled on October 8, 2025, this new feature gives developers the freedom to connect tools and services like Figma and Stripe straight into Gemini CLI, transforming it into a powerful and flexible development platform.

A Step Toward an Open Ecosystem

Unlike OpenAI’s recently launched ChatGPT apps—which require approval before being listed—Google’s approach embraces openness. With Gemini CLI Extensions, any developer or company can publish their own integrations without needing Google’s endorsement or participation. These extensions are hosted on public repositories such as GitHub, where developers can manually install them.

Taylor Mullen, a senior staff engineer at Google, emphasized this open approach in an interview with TechCrunch, noting that the project’s foundation is “a fair ecosystem that anyone can participate in.”

How It Works

The first available extension connects Gemini CLI to Google’s Nanobanana image generator, letting users create AI-generated images directly from the terminal once the extension is installed. This marks a big step in merging creative and coding workflows inside the command-line environment.

Ryan J. Salva, Google’s senior director of product management for developer tools, explained that the vision behind Gemini CLI Extensions is to turn the CLI into “an extensibility platform—a bridge to other tools and commands in your development workflow.”

Rapid Growth Among Developers

Since its launch in June 2025, Gemini CLI has already surpassed one million users, primarily developers who rely on the AI system for coding, debugging, and project maintenance. Interestingly, Google itself uses Gemini CLI extensively for internal development, demonstrating its practical reliability and scalability.

Why It Matters

The introduction of Gemini CLI Extensions positions Google’s AI developer tools as a serious contender in the race to make coding more collaborative, open, and AI-driven. By enabling anyone to build and share extensions freely, Google is fostering a vibrant ecosystem that could evolve as rapidly as the open-source community itself.

Conclusion

With Gemini CLI Extensions, Google isn’t just expanding its AI coding assistant—it’s empowering developers to shape their own AI-driven workflows. The open model could redefine how developers collaborate, create, and automate within their coding environments, setting a new benchmark for flexibility in AI development tools.

OpenAI has taken another major step toward empowering developers by launching AgentKit, a comprehensive toolkit for building and deploying AI agents. The announcement was made by OpenAI CEO Sam Altman during the company’s Dev Day 2025 event.

Altman described AgentKit as “a complete set of building blocks” that helps developers take AI agents from concept to full-scale production with minimal friction. The toolkit aims to simplify the creation, deployment, and optimization of AI-driven workflows—marking a clear move to make AI agent development faster, more accessible, and enterprise-ready.

What’s Inside AgentKit

AgentKit introduces four key components that streamline the process of building intelligent, autonomous agents:

• Agent Builder – Dubbed as the “Canva for AI agents,” this feature allows developers to visually design the logic, workflows, and behaviors of agents. It’s built on top of OpenAI’s existing Responses API, which already supports hundreds of thousands of developers worldwide.
• ChatKit – A plug-and-play chat interface that developers can embed directly into their applications. It offers customization options to match brand identity and integrate unique workflows, allowing businesses to deliver AI-powered conversations tailored to their users.
• Evals for Agents – A performance evaluation framework that enables step-by-step trace grading, dataset benchmarking, and automated prompt optimization. It also supports evaluations on external models, ensuring high-quality agent behavior and decision-making.
• Connector Registry – This feature lets developers securely connect their AI agents to internal tools and third-party systems through an admin control panel. It ensures both flexibility and data security—critical for enterprise adoption.

A Demonstration of Speed and Simplicity

During the event, OpenAI engineer Christina Huang demonstrated AgentKit’s capabilities live on stage—building a complete AI workflow and two functional agents in under eight minutes. The demonstration showcased just how streamlined the process has become for developers.

Altman emphasized that this toolkit represents everything OpenAI’s team wished they had when building their first AI agents. He also revealed that several launch partners are already leveraging AgentKit to scale their AI solutions rapidly.

Strengthening OpenAI’s Developer Ecosystem

The launch of AgentKit follows other major announcements at Dev Day, including the ability to build applications directly inside ChatGPT, which now boasts 800 million weekly active users. Together, these innovations underscore OpenAI’s strategy to become the go-to platform for AI agent development, as competition intensifies with rivals racing to deliver similar enterprise solutions.

Conclusion

With AgentKit, OpenAI is not just providing another developer tool—it’s laying the groundwork for the next wave of AI automation. By giving developers intuitive visual tools, powerful evaluation systems, and secure integrations, OpenAI is making it easier than ever to bring intelligent agents to life across industries.

Google is pushing its AI coding agent Jules deeper into developer toolchains, making it more accessible and powerful as competition in AI-assisted coding intensifies. The company has rolled out Jules Tools, a new command-line interface (CLI), along with a public API, allowing developers to integrate the agent directly into their terminals, CI/CD systems, Slack, and even IDEs.

Jules Tools: AI Coding Inside the Terminal

Previously available only through its website and GitHub, Jules can now live inside the terminal. With the new CLI, developers no longer need to switch between web and GitHub — instead, they can run commands, delegate tasks, and validate results right where they work.

Kathy Korevec, Google Labs’ director of product, explained: “We want to reduce context switching for developers as much as possible.”

Unlike Google’s Gemini CLI, which requires ongoing collaboration, Jules Tools is designed for “scoped tasks” that can run independently once approved by the user.

How Jules Stands Out

Google’s senior developer advocate, Denise Kwan, noted that Jules is intentionally less interactive than Gemini CLI. It executes tasks independently, minimizing back-and-forth. The public API further extends its reach, enabling developers to embed Jules in IDEs like VS Code or even create custom integrations across their workflows.

Google is also working on dedicated IDE plug-ins to strengthen Jules’ presence in developer environments.

New Features and Expanding Beyond GitHub

In recent weeks, Jules has gained several upgrades, including:

• Memory for recalling user interactions and preferences
• Stacked diff viewer layout
• Image uploads
• Pull request comment handling

Currently, Jules is tied to GitHub repositories. However, Google is exploring integrations with other code hosting platforms and even environments without version control.

Oversight and Mobile Challenges

Jules is built with task oversight in mind. If it encounters an issue, it pauses and prompts the user for help rather than stalling silently.

On mobile, though, challenges remain. While Jules works through the web interface, native notifications aren’t supported yet. Korevec confirmed Google is actively improving the mobile experience.

Who’s Using Jules?

So far, Jules has mainly attracted software engineers and professionals, but some casual coders are also using it to extend projects started on lighter “vibe coding” platforms.

Launched in public preview in May and out of beta by August, Jules is now available in structured pricing tiers:

• Free plan: Up to 15 daily tasks, 3 concurrent tasks
• Pro plan ($19.99/month): 5× higher limits
• Ultra plan ($124.99/month): 20× higher limits

Conclusion

By adding a CLI, API, and memory features, Google is transforming Jules from a GitHub-dependent assistant into a flexible, workflow-ready AI coding agent. With competition heating up in the AI coding space, these moves position Jules as a serious contender for developers who want efficiency without heavy context switching.

Opera has officially launched Neon, a next-generation browser built around artificial intelligence, aimed at redefining how people interact with the web. Unlike traditional browsers, Neon is designed to act as a digital assistant, helping users complete tasks, automate workflows, and even build mini-apps — all through AI-driven prompts.

Opera first teased Neon back in May, when it was available only in closed preview. Now, the company is rolling it out more widely, though access is limited to select invitees. Those who get in will need to pay a subscription fee of $19.99 per month.

What Makes Neon Different?

Neon isn’t just another browser with a chatbot bolted on. It has several standout features that set it apart from competitors like Perplexity’s Comet and The Browser Company’s Dia:

• Neon Do: A productivity-focused tool that can perform tasks like summarizing a Substack post and automatically sharing it to Slack, or fetching details from a YouTube video you watched last week.
• AI Coding Support: The browser can generate code snippets to help create visual reports with tables and charts.
• Cards (Repeatable Prompts): Much like Dia’s “Skills,” Cards let you build reusable AI prompts. Think of them as IFTTT for AI, where you can chain together commands like “pull-details” and “comparison-table” to automate workflows across tabs.
• Tasks (Tab Workspaces): A new organizational feature that works like a hybrid of Tab Groups and Arc Browser’s Spaces, but with added AI context.

Aiming for Power Users

Opera is pitching Neon as a premium tool for heavy AI users rather than a casual browsing option. By charging a subscription fee, the company is signaling that it sees Neon as a productivity investment, not just another free browser.

In its demos, Opera showcased Neon handling tasks as ambitious as ordering groceries on your behalf. However, like many AI-driven products, the real test will be whether these features hold up outside of controlled demos.

The Bigger Picture

Opera enters an increasingly competitive space. Beyond Perplexity and Dia, major players like Google and Microsoft are also rapidly infusing AI into their browsers. But Opera’s bet is that a dedicated, AI-first browser with powerful automation tools will appeal to a niche audience of professionals and enthusiasts who rely heavily on AI in their daily workflows.

Conclusion

Opera Neon represents a bold step into the future of browsing — where the browser isn’t just a window to the internet, but an active collaborator in your work. Whether it can deliver on its promise and justify its monthly fee will determine if Neon becomes a game-changer or just another experiment in AI-powered browsing.

Anthropic has unveiled Claude Sonnet 4.5, its latest frontier AI model that promises state-of-the-art performance on coding tasks. Unlike earlier models that often stopped at prototypes, Anthropic claims this version can build production-ready applications, marking a leap in reliability and practical use.

Pricing and Availability

Claude Sonnet 4.5 is available starting today through the Claude API and the Claude chatbot. Pricing remains the same as its predecessor, Claude Sonnet 4: $3 per million input tokens and $15 per million output tokens. For context, one million input tokens equals roughly 750,000 words—longer than the entire Lord of the Rings trilogy.

What Makes It Stand Out

Anthropic says Claude Sonnet 4.5 sets new benchmarks in coding, excelling on tests such as SWE-Bench Verified. However, researchers emphasize that its real-world impact goes beyond numbers. During enterprise trials, the model coded autonomously for up to 30 hours straight, handling not just application building but also standing up databases, purchasing domain names, and even conducting a SOC 2 audit for security compliance.

Industry leaders are taking note. Cursor CEO Michael Truell called the model “state-of-the-art” for long-term coding tasks, while Windsurf CEO Jeff Wang described it as a “new generation of coding models.”

Competitive Landscape

Anthropic has gained traction with developers and enterprises alike, powering platforms such as Cursor, Windsurf, and Replit. Apple and Meta reportedly use Claude models internally as well. But competition is intensifying: OpenAI’s GPT-5 has recently outperformed Claude in some coding benchmarks, challenging Anthropic’s dominance.

In response, Anthropic highlights that Claude Sonnet 4.5 is also its most aligned frontier model yet, with reduced tendencies toward sycophancy and deception, plus improved defenses against prompt injection attacks.

New Tools for Developers

Alongside the model, Anthropic is rolling out the Claude Agent SDK, the same infrastructure behind Claude Code. This toolkit will let developers build their own AI-powered coding agents.

Additionally, Anthropic is previewing “Imagine with Claude”, a research feature for Max subscribers that showcases the model’s ability to generate software in real time—without prewritten code.

Conclusion

The launch of Claude Sonnet 4.5 underscores how quickly the AI race is accelerating, with new flagship models emerging every few months. Arriving less than two months after Claude Opus 4.1, this release positions Anthropic as a continued leader in AI coding innovation. Whether it can maintain that lead against rivals like OpenAI remains to be seen, but for now, developers have access to one of the most capable coding models available.

DeepSeek has unveiled its new experimental model, V3.2-exp, designed to make AI more affordable by significantly reducing API costs in long-context operations. The company announced the release on Hugging Face and also shared a linked academic paper on GitHub.

At the core of this advancement is DeepSeek Sparse Attention, an innovative mechanism that improves efficiency without sacrificing performance. The system works in two steps:

• A lightning indexer pinpoints and prioritizes the most relevant excerpts from the context window.
• A fine-grained token selection system then identifies the most important tokens within those excerpts to fit into the model’s limited attention window.

Together, these components allow the model to handle long-context workloads while keeping server demands lower than traditional transformer models.

Cutting API Costs in Half

DeepSeek’s initial testing shows promising results: API calls in long-context scenarios could be reduced by up to 50%. This breakthrough could deliver huge savings for developers and businesses struggling with high inference costs. Since the model is open-weight and freely available, researchers and developers worldwide will soon be able to validate its performance.

Why It Matters

Inference costs—the ongoing expenses of running pre-trained AI models—are one of the biggest challenges in deploying large-scale AI systems. By rethinking how transformer models process long contexts, DeepSeek demonstrates that there are still major opportunities to improve efficiency at the architectural level.

DeepSeek’s Role in Global AI

China-based DeepSeek has already made a name for itself with the R1 model, which relied heavily on reinforcement learning and was trained at a fraction of the cost of its U.S. competitors. Although R1 didn’t spark the sweeping industry changes some expected, it showed DeepSeek’s ability to innovate in cost efficiency.

The new sparse attention approach may not cause the same buzz as R1, but its practical impact could be profound. By cutting inference costs, DeepSeek is offering a path to more sustainable and scalable AI operations — a lesson that could prove invaluable for U.S. and global AI providers alike.

Conclusion

With V3.2-exp, DeepSeek has introduced a practical and powerful solution to one of AI’s toughest hurdles: keeping API costs under control while managing longer context windows. If the early results hold true, this could become a landmark step toward building more cost-effective AI systems.

Meta has rolled out a brand-new feature called “Vibes”—a short-form video feed filled entirely with AI-generated clips. Available in the Meta AI app and on meta.ai, the feature looks and feels like TikTok or Instagram Reels, except every single video is machine-made.

What Is Vibes?

Meta CEO Mark Zuckerberg announced the launch on Instagram, sharing a set of quirky AI-generated videos:

• A group of fuzzy-looking creatures hopping across colorful cubes
• A cat kneading dough
• An ancient Egyptian woman snapping a selfie with pyramids in the background

According to Meta, Vibes will showcase AI-generated videos from both creators and users, and the feed will adapt to your preferences over time.

How It Works

Users can:

• Generate videos from scratch
• Remix existing ones on the feed
• Add visuals, music, and different styles before publishing

Once ready, videos can be shared directly on Vibes, sent via DM, or cross-posted to Instagram and Facebook Stories/Reels.

The Tech Behind It

For now, Meta is relying on partnerships with Midjourney and Black Forest Labs to power Vibes while it continues developing its own AI models. Alexandr Wang, Meta’s chief AI officer, confirmed the collaboration in a recent post.

User Backlash

While Meta sees Vibes as the next big thing, the reception hasn’t been kind. Under Zuckerberg’s announcement, top comments included:

• “gang nobody wants this”
• “Bro’s posting AI slop on his own app”
• “I think I speak for everyone when I say: What…?”

The criticism highlights growing frustration with what many now call “AI slop”—low-value, machine-generated content flooding social media feeds.

Why It’s a Strange Move

This launch feels oddly out of step with the industry. Platforms like YouTube are actively cracking down on AI spam, while Meta itself previously encouraged creators to focus on authentic storytelling. Launching an AI-only feed goes directly against that advice, raising questions about whether Vibes will actually resonate with users—or quickly fade away.

Conclusion

With Vibes, Meta is betting that people want an AI-first TikTok alternative, but early reactions suggest the opposite. As AI-generated content continues to spark debate, it remains to be seen whether Vibes will thrive—or become another experiment that quietly disappears.