2025年8月16日土曜日

Gemini CLI + VS Code: Native diffing and context-aware workflows

 Get ready to experience the future of command-line apps from within VS Code! Our latest Gemini CLI update brings a new level of intelligence directly to your VS Code integrated terminal. This isn't just any update; it evolves how you interact with your projects, offering a deep IDE integration that understands your context and suggests changes right where you need them.


What's new: A deeper integration

This update fundamentally changes how Gemini CLI interacts with your editor, making your developer workflows smoother and more efficient.

  • Workspace & selection context: When you connect your VS Code IDE to Gemini CLI, Gemini CLI has direct access to your workspace. It becomes aware of the files you have open and can access selected text. This allows the CLI to provide targeted and contextually relevant suggestions, as the tool understands precisely what you're working on at the moment.

  • Native in-editor diffing: Suggestions from Gemini CLI now trigger a full-screen diff view directly inside VS Code. This gives you a comprehensive, side-by-side review of changes. Crucially, you can modify the code right within this diff view before you accept it, giving you full control and flexibility.

Getting started

Ready to try it out? The setup is straightforward.

1: Prerequisites:

  • Gemini CLI version 0.1.20 or higher
  • Run the CLI from the integrated terminal within VS Code

2: One-time setup: From your integrated terminal, run the /ide install command to install the necessary companion extension

3: Toggle integration: After the one-time setup, you can easily manage the integration:

  • To activate it: /ide enable
  • To deactivate it: /ide disable

Enable IDE integration today and experience a new way to build with Gemini CLI and VS Code.

2025年8月13日水曜日

Microsoft Rolls Out GPT-5 Across Its Copilot Suite — Here's Where to Find It

 GPT-5 is here, and Microsoft is integrating it into nearly every corner of its AI ecosystem — from chatbots to enterprise tools, developer platforms, and beyond.

If you're using anything with the "Copilot" brand, chances are you're about to see GPT-5 under the hood. Here's what’s new and where GPT-5 is now live:


🔹 1. Microsoft Copilot (Chatbot)

The main Copilot chatbot — available for free to all users — now includes GPT-5 under a new “Smart mode.”

Expect better reasoning, context retention, and more accurate answers — even in multi-turn conversations.

No subscription needed.


🔹 2. Microsoft 365 Copilot (Word, Excel, Outlook, etc.)

Business and enterprise users get a more powerful assistant across apps like Word, Excel, and Outlook.

  • GPT-5 can reason over emails, documents, and files

  • Improves handling of longer, more complex tasks

  • Designed to “stay on track” during extended conversations


🔹 3. Copilot Studio (Build Your Own AI Agents)

Copilot Studio now lets users build custom GPT-5-powered agents for specific business workflows.

  • Select GPT-5 in your agent’s configuration

  • Great for automating complex processes

  • Supports multi-turn, context-aware interactions


🔹 4. GitHub Copilot (in VS Code & Other IDEs)

GPT-5 is now rolling out in preview to all GitHub Copilot Pro subscribers.

  • Seamlessly integrates into VS Code

  • Offers smarter code suggestions and bug fixes

  • Use with or without your own OpenAI API key (to bypass usage limits)

If you were underwhelmed by GPT-4.1 in Copilot, GPT-5 may be a game-changer.

🔹 5. Azure AI Foundry (For Developers & Enterprises)

Microsoft is also opening the full GPT-5 family to developers in Azure AI Foundry, including:

  • GPT-5 (full model): 272k token context window

  • GPT-5 mini: Optimized for real-time applications

  • GPT-5 nano: Designed for ultra-low-latency responses

  • GPT-5 chat: Multimodal, multi-turn assistant with 128k–272k token context

Azure’s model router automatically chooses the best GPT-5 variant for your task.


⚠️ Security & Safety

Microsoft highlighted GPT-5’s performance in AI Red Team tests — noting major improvements in safety across categories like:

  • Malware generation

  • Scam/fraud automation

  • Other misuse vulnerabilities


💡 Final Thoughts

GPT-5 is no longer just a lab experiment — it's going mainstream across Microsoft’s ecosystem.

Whether you’re writing code, summarizing documents, or building custom agents, expect better performance, deeper reasoning, and much larger context handling — often without needing a paid plan.


How to Use GPT-5 in VS Code with GitHub Copilot Pro Unlock GPT-5, test it for free, and bypass usage limits using your own API key.

 With GPT-5 now integrated into GitHub Copilot Pro, developers can tap into the latest large language model directly within Visual Studio Code (VS Code). Here's how to enable GPT-5, start a free trial, and optionally connect your own OpenAI API key to lift Copilot limits.


✅ Step 1: Enable GitHub Copilot Pro (Free Trial Available)

To use GPT-5 in VS Code, you’ll need Copilot Pro. Here’s how to activate it:

  1. Open VS Code.

  2. Click the Copilot icon in the sidebar.

  3. Click on the current model (e.g., GPT-4.1).

  4. Select “Add Premium Models.”

  5. You’ll be redirected to GitHub. Sign up for Copilot Pro — a 30-day free trial is available.

  6. You’ll need to enter your credit card (not charged unless you continue after trial).

Once activated, restart VS Code for the new model options to appear.


✅ Step 2: Switch to GPT-5 in VS Code

After restarting:

  1. Click the model dropdown again (usually GPT-4.1).

  2. Scroll down and select GPT-5.

  3. Issue any prompt in the Copilot chat (e.g., ask GPT-5 to refactor or search your codebase).

  4. When the “Enable” button appears, click it to activate GPT-5.

That’s it — you’re now using GPT-5 via GitHub Copilot Pro.


✅ Step 3: Optional — Bypass Usage Limits with Your OpenAI API Key

Copilot Pro may have usage caps per model. If you want more flexibility, you can connect your own OpenAI API key.

Here’s how:

  1. Visit https://platform.openai.com/account/api-keys.

  2. Log in and click “Create API Key.”

  3. Copy the key.

  4. In VS Code, open the Copilot model dropdown and choose Manage Models.

  5. Select OpenAI as your provider.

  6. Paste your API key, press Enter to confirm.

You're now using GPT-5 through your personal OpenAI key in VS Code, sidestepping Copilot's usage limits.


Bonus Tips:

  • You can still toggle between GPT-4, GPT-4o, and GPT-5 depending on task complexity and response quality.

  • Using your own API key gives access to OpenAI usage logs, useful for tracking token spend and optimizing prompts.


🧠 Final Thoughts

GitHub Copilot Pro with GPT-5 brings serious coding power into your IDE — from generating full functions to debugging complex systems. Whether you're just exploring with the free trial or going all-in with your own API key, it’s now easier than ever to work with the latest LLMs directly in VS Code.

Have you tried GPT-5 in Copilot? Share your thoughts or prompt ideas in the comments!

Goodbye, $165K Tech Jobs. Hello, Chipotle? As AI reshapes the tech industry, new grads are left searching for jobs

 Just a few years ago, learning to code seemed like a guaranteed path to a six-figure salary. Tech giants, politicians, and even school curriculums pushed students to major in computer science, promising a future of high-paying jobs and world-changing innovation.

But in 2025, that dream is unraveling.

Take Manasi Mishra, a recent computer science graduate from Purdue. After a year of job searching and no offers from tech firms, the only interview she landed was with Chipotle. Her experience isn’t unique — it reflects a growing crisis for newly minted CS grads across the U.S.

“The only company that called me for an interview is Chipotle,” she said in a TikTok video that's since gone viral.

A Brutal Market for New Developers

The numbers tell the story. Computer science was once the golden ticket — in 2014, about 80,000 students majored in it. In 2024, that number more than doubled to over 170,000. But now, job opportunities are shrinking just as fast.

The rise of AI coding tools — like GitHub Copilot and CodeRabbit — has made it easier for companies to automate the very entry-level tasks that junior developers used to do. At the same time, mass layoffs at Amazon, Microsoft, Intel, Meta, and others have flooded the job market with experienced engineers.

The result? Recent CS grads are stuck in limbo.

Among 22- to 27-year-olds, CS and computer engineering grads have some of the highest unemployment rates:

  • 6.1% for CS

  • 7.5% for computer engineering
    (That’s more than double the rate for art history and biology grads.)

"Soul-Crushing" Job Hunts

Zach Taylor, who graduated in 2023 from Oregon State with a CS degree, has applied to over 5,700 jobs. His efforts have yielded just 13 interviews — and zero full-time offers. At one point, even McDonald’s rejected him due to “lack of experience.”

He’s not alone.

Many grads say they feel ghosted by companies, despite spending months on applications, online coding tests, and interviews. Some even describe the experience as “soul-crushing” or “gaslighting”, especially after being told for years that CS was a guaranteed career win.

AI: Both the Problem — and the Future?

Ironically, the same technology that made coding more accessible is now making jobs harder to land.

  • Companies are using AI to screen resumes, often rejecting candidates in seconds.

  • Job seekers are using AI tools to mass-apply — making competition even more intense.

  • AI coding tools are replacing junior engineers, especially in roles focused on repetitive tasks.

“The unfortunate thing is, entry-level positions are the easiest to automate,” said economist Matthew Martin from Oxford Economics.

There’s a growing sense among grads that they’re stuck in a doom loop: AI tools are taking their jobs and blocking them from even getting interviews.

A New Direction for Some

Some students are pivoting.

  • Mishra, for example, realized she preferred tech marketing and sales over engineering — and just landed a job in tech sales.

  • Audrey Roller, a data science grad from Clark University, is highlighting her human traits — like creativity — in hopes of standing out.

  • Others are focusing on AI literacy, hoping that new skills will align them with emerging job demands.

Still, it’s clear the tech job landscape has changed — fast.

So, What Now?

The boom in computer science education created a generation of hopeful developers. But as AI disrupts everything from hiring pipelines to the code itself, many of those grads are discovering that the job market no longer has room for them — at least not in the roles they were trained for.

That doesn’t mean the end of opportunity — but it does mean the rules have changed. Quickly.

Claude Sonnet 4 Can Now Handle a Million Tokens — But What Does That Mean for Developers (and Their Budgets)?

 Anthropic just took a massive leap in AI usability. On Tuesday, the company announced that its Claude Sonnet 4 model can now handle 1 million tokens of context in a single request — a 5x increase that changes how developers can interact with large-scale data.

So, what does that actually mean?

Think of context tokens like memory: the more your AI model can “remember” at once, the more it can understand, connect, and reason about. With 1 million tokens, developers can now feed Claude entire codebases, full-length research papers, or massive documentation — all in one go. No more splitting projects into chunks and losing the thread between files.

Why This Matters for Developers

With this update (available now in public beta via Anthropic's API and Amazon Bedrock), Claude can now:

  • Analyze full software projects — even those with 75,000+ lines of code

  • Generate suggestions across an entire system, not just isolated parts

  • Handle long-running sessions without losing context or logic

This opens up entirely new possibilities for AI-powered development — from real-time debugging across massive apps to AI agents that can reason through complicated workflows or documentation libraries.

A Big Deal for Real-World AI Tools

Industry voices are already calling this a game-changer.

Sean Ward, CEO of iGent AI, said the upgrade “supercharged autonomous capabilities” in their coding assistant Maestro. Eric Simons of Bolt.new highlighted that with 1 million tokens, developers can finally scale AI tools to match real-world project sizes — and still get reliable output.

But there’s another side to this coin: price.


Let’s Talk Pricing — and What It Means for You

With great context comes great… compute costs.

Anthropic has updated Claude’s pricing to reflect the heavier processing needed for longer prompts:

Context SizeInput Cost per Million TokensOutput Cost per Million Tokens
≤ 200,000 tokens$3$15
> 200,000 tokens$6$22.50

At first glance, those higher prices for large prompts might give teams pause — especially when compared to cheaper options like OpenAI’s latest models. Some estimates suggest Claude Opus 4 can cost up to 7x more than GPT-5 for similar workloads.

But here’s where Anthropic’s strategy gets interesting.

They argue that quality and efficiency matter more than token price alone. Features like prompt caching allow frequently used large datasets to be stored and reused, helping companies reduce the real-world cost of repeated queries. And unlike traditional RAG (retrieval-augmented generation) methods, Claude’s long context lets the model see the entire data landscape at once — which can lead to more accurate, more connected responses.

As one Anthropic spokesperson put it:

“Large context lets Claude see everything and choose what’s relevant... often producing better answers than pre-filtered RAG results.”


The Takeaway

Anthropic’s million-token context update for Claude Sonnet 4 is a huge step forward in how AI can support real-world development. But it also forces teams to think harder about cost vs. capability.

If you're building apps or systems that rely on large datasets, deep context, or multi-step logic, this could be a worthwhile investment — especially if quality and continuity are mission-critical.

But if you're just looking for quick, lightweight prompts or smaller tasks, the old 200k-token ceiling might still be enough — and much cheaper.

Either way, the bar for context-aware AI just got a lot higher.