• AIdeations
  • Posts
  • 25% of Google’s Code is Now Written by AI

25% of Google’s Code is Now Written by AI

AI is reshaping coding—what it means for you.

OpenAI Just Partnered with Broadcom to Build Its First AI Chip

Quick Byte:

OpenAI isn’t just making ChatGPT—it’s diving into the hardware game, teaming up with Broadcom and TSMC to build its own custom AI chip. After ditching plans for a billion-dollar foundry network, OpenAI’s new strategy is all about custom design, cost-cutting, and maybe, just maybe, shaking up the chip world.

Key Takeaways:

  • The First Custom Chip: OpenAI is working with Broadcom to create a custom AI chip that focuses on inference, the core function that powers ChatGPT’s real-time responses. Forget just borrowing from Nvidia—OpenAI is designing its own tech now.

  • AMD Jumps In: Nvidia might hold 80% of the market, but OpenAI is looking beyond that, adding AMD’s MI300X chips via Microsoft Azure. Think of it as a tactical move to reduce its dependence on Nvidia and gain some breathing room on costs.

  • Foundry Plans on Hold: OpenAI initially considered building its own chip factories but dropped the idea due to the insane costs. Instead, they’ve built a chip team and are securing partnerships to hit the ground running in 2026.

Bigger Picture:

OpenAI’s custom chip strategy signals a huge shift in how AI companies approach their infrastructure. It’s not just about “what’s cheapest” or “what’s available”—it’s about creating a future where AI companies don’t need to rely on Nvidia’s hardware to get things done. By working with Broadcom for design and TSMC for manufacturing, OpenAI is skipping the billion-dollar investment in building foundries, opting instead for partnerships that let them stay agile, ramp up production, and possibly drive down the cost of AI.

In a world where Nvidia’s GPUs still dominate, OpenAI’s shift to AMD’s MI300X chips on Microsoft’s Azure marks a clear move toward more diversified, cost-effective tech stacks. This isn’t just some side project—it’s a sign of OpenAI scaling fast and learning to build without relying on a single player. As other companies catch on, we could see a more competitive AI chip market with more affordable AI applications for businesses everywhere.

AI is Writing Over 25% of Google’s New Code—What Does That Mean for the Future of Work?

Quick Byte:

More than a quarter of Google’s new code is being generated by AI, according to CEO Sundar Pichai. Google’s internal AI model, “Goose,” trained on decades of engineering expertise, is now helping coders crank out code faster than ever.

Key Takeaways:

  • AI-Created Code: Google’s new AI model, Goose, is now producing over 25% of the company’s new code. Engineers review and refine the AI’s output, but the productivity gains are undeniable.

  • A Boost in Efficiency: According to Pichai, AI-driven coding is helping engineers “move faster” and get more done. This efficiency boost underscores Google’s commitment to internal AI adoption, helping streamline workflows.

  • Job Security? The fact that AI is generating a significant chunk of code may raise concerns about job security among engineers. While Google’s leaders insist that AI isn’t replacing jobs (yet), this trend certainly reshapes the future of coding at Google.

Practical Tips for Business Owners:

  • Consider AI for Repetitive Tasks: Like Google’s approach, using AI for repetitive work could speed up processes in your business, especially if you have tasks that don’t require human creativity or complex problem-solving.

  • Streamline with AI Assistants: If Google’s model works, consider implementing an AI solution to help your teams handle day-to-day tasks and focus on more strategic work.

  • Balance AI with Human Review: Even with AI, human oversight is key. Google has engineers reviewing AI-generated code to ensure quality. If your business uses AI, make sure employees are involved in reviewing and refining the output.

Bigger Picture:

Google’s internal AI model, Goose, represents a major shift in how companies approach coding. By producing over 25% of new code, AI is turning what used to be a purely human task into a hybrid process. While Google insists AI isn’t here to take jobs, the rapid adoption of AI in coding suggests a new kind of workforce synergy where AI handles repetitive or foundational work, leaving engineers to focus on the bigger, more complex problems.

The takeaway? As AI gets more integrated into high-skill fields like software engineering, it’s likely that other industries will follow. This isn’t about robots taking over—it’s about figuring out how humans and AI can work together to get more done in less time. Google’s experiment with Goose might just be a glimpse into a future where “AI teammates” become the norm across sectors.

Headline

The AI Boom is Built on Billions of Tons of Concrete—And It’s Undermining Big Tech’s Climate Promises

Quick Byte:

The rise of AI isn’t just about software—it’s about physical infrastructure, and that means a lot of concrete. Data centers require massive amounts of it, pushing the tech industry further from its carbon goals. Big tech’s solution? Greener concrete innovation and carbon capture, but progress remains slow.

Key Takeaways:

  • Concrete and Carbon: Concrete is essential for building data centers, but its production is a huge carbon emitter. Cement, the main ingredient in concrete, accounts for about 6% of global CO2 emissions.

  • Big Data Needs Big Materials: With Google, Meta, Microsoft, and Amazon racing to build data centers, demand for concrete is soaring. Analysts project data centers alone will emit 2.5 billion tons of CO2 each year by 2030, nearly half of what the U.S. emits annually.

  • Innovation in Green Concrete: As tech companies scramble to lower their carbon footprint, they’re investing in green concrete. AI-powered solutions, carbon capture, and low-carbon materials like fly ash and calcined clay offer promising alternatives but are still in the early stages of adoption.

Practical Tips for Business Owners:

  • Emphasize Sustainable Choices: If construction is a big part of your business, consider exploring low-carbon concrete or recycled materials to cut down on emissions.

  • Leverage Partnerships for Green Goals: Collaborate with green tech providers to explore sustainable materials. Even smaller shifts toward green building practices can make a significant difference over time.

  • Watch for Policy Changes: As governments push for sustainable construction, subsidies and requirements for greener materials may impact business. Staying ahead of these trends could benefit your business in the long run.

Bigger Picture:

The AI boom is pushing big tech companies like Google and Microsoft further from their climate goals. As the world builds out data centers to support AI, concrete demand is skyrocketing—and with it, CO2 emissions. While tech giants have committed to carbon neutrality by 2030 (or 2040, in Amazon’s case), the reality of their construction needs paints a different picture. Last year alone, Microsoft’s emissions jumped 30%, largely due to new data center materials.

The industry’s carbon footprint isn’t just an abstract figure. Each data center expansion pulls in trucks of concrete, hundreds of electricians, and an army of structural engineers. Demand for concrete is expected to grow as companies rush to keep up with AI infrastructure demands. The global scale is staggering: about 35 billion tons of concrete are used annually, or roughly 4 tons for every person on Earth.

On the flip side, a green concrete revolution is gaining momentum. Tech companies, along with governments, are experimenting with low-carbon alternatives and carbon capture systems. Projects like the Open Compute Project Foundation, joined by Amazon, Google, Meta, and Microsoft, aim to accelerate green concrete in data centers. And with AI helping discover better concrete formulas, there’s hope that technology itself might provide a solution to the problem it’s creating.

AI Agents Could Transform Your Job—And They’re Closer Than You Think, Says Research Firm Gartner

Quick Byte:

AI is about to go beyond answering questions. Gartner predicts that AI agents—smart, autonomous bots—will soon be handling tasks across IT, security, and marketing, potentially surpassing human performance on repetitive tasks and creating a new wave of “AI colleagues.”

Key Takeaways:

  • The Rise of AI Agents: Unlike chatbots, AI agents don’t just respond to questions—they actively handle tasks autonomously. From coding to managing expense reports, these agents could soon be a staple in workplaces, with many companies adopting or testing them by 2025.

  • Expanding AI Roles: Gartner VP Arun Chandrasekaran highlights three primary areas where AI agents will be game-changers: Information Technology, Security, and Marketing. By performing tasks like content generation and process automation, AI agents offer time and cost savings for businesses.

  • Adoption on the Horizon: According to Gartner’s forecasts, by 2026, over 100 million people will interact with AI agents as “colleagues,” with 80% of these interactions semi-automated—meaning agents will handle tasks with minimal human oversight.

Practical Tips for Business Owners:

  • Explore AI in Repetitive Work: Evaluate areas where AI agents could handle repetitive tasks, such as data entry or customer support, allowing your team to focus on higher-value work.

  • Start Small with AI Trials: Test AI agents in a limited capacity to assess their impact on productivity and identify any operational kinks. Consider tools from Microsoft’s Copilot Studio or Salesforce’s AI agent system to get started.

  • Address Employee Concerns: Be transparent about AI’s role in the workplace. While AI agents can free up time, communicate openly with your team about how this technology complements their work rather than replacing it.

Bigger Picture:

Gartner’s latest insights reveal how AI agents are likely to reshape the workplace over the next few years. As the tech matures, AI agents will extend beyond simple automation, taking on tasks like managing workflows, coding, and even handling basic sales functions. This shift could make AI an integral part of daily operations in IT, security, and marketing—where companies are already gearing up to deploy “fleets” of agents to maximize efficiency.

While the transition may spark concerns about AI replacing jobs, Gartner’s findings suggest a more nuanced reality: AI agents can handle the mundane tasks that often bog down employees, allowing people to focus on strategic, creative, and complex responsibilities. For now, these digital coworkers are in early stages, but with companies like Microsoft, Salesforce, and Google all in the game, AI agents are closer to becoming an office staple than many might think.

How To Install Claude’s New AI Agent That Can Control Your Computer

Authors:
Alexey Dontsov, Dmitrii Korzh, Alexey Zhavoronkin, Boris Mikheev, Denis Bobkov, Aibek Alanov, Oleg Y. Rogov, Ivan Oseledets, Elena Tutubalina
Affiliated with: AIRI, MIPT, Skoltech, Sber, University of Sharjah, HSE University

Summary:
CLEAR introduces a new benchmark specifically for evaluating machine unlearning (MU) in multimodal models (those that handle both text and images). The benchmark allows researchers to assess how well AI can forget certain information, such as private or sensitive data, without losing critical knowledge from other areas. This benchmark includes 200 fictitious individuals with 3,700 images and question-answer pairs, offering a synthetic dataset that helps measure forgetting performance across text and visual domains. Researchers adapted ten unlearning techniques for these tasks and found that multimodal forgetting comes with unique challenges, particularly around maintaining knowledge in non-forget areas.

Why This Research Matters:
Machine unlearning is crucial for privacy and data safety in AI, especially as models become more complex and integrated into real-world applications. As AI models are often trained on vast datasets, they might inadvertently memorize sensitive information, posing privacy risks. The CLEAR benchmark provides a structured way to test unlearning techniques in multimodal setups, helping researchers develop models that forget specific data while retaining necessary information—a critical need for industries that prioritize privacy and data compliance.

Use Cases:

  1. Privacy Compliance: CLEAR can help in creating AI models that comply with privacy laws by removing sensitive data upon request.

  2. Personalization and Security: AI in personal assistants or customer service could unlearn customer-specific data after interactions to enhance security.

  3. Healthcare AI: Patient data in healthcare applications could be unlearned as needed, preserving privacy while maintaining general knowledge.

  4. Law Enforcement and Surveillance: Facial recognition systems could unlearn data about specific individuals, supporting privacy-centric applications.

Immediate Impact:
CLEAR offers a direct tool for researchers to gauge and improve machine unlearning capabilities. This benchmark could accelerate the adoption of privacy-conscious AI in sensitive fields, ensuring models handle only the data they need while discarding sensitive information accurately.

Future Impact:
In the long run, CLEAR could set the standard for privacy-respecting AI, particularly as multimodal systems grow in use. It might lead to regulatory frameworks that demand unlearning capabilities in AI systems, pushing for tools that not only learn but can also effectively forget, supporting data safety in sectors like healthcare, finance, and social media.

Lasso - Turn metadata into revenue with industry-leading accuracy. As the only platform indexing every business with a website, LinkedIn page, and at least one employee, we deliver 100% additive, verified contact data—helping sales teams find net new opportunities and unlock millions in revenue.

Bolt - Prompt fullstack web applications into existence, see them executed in real-time, debug errors as they occur & deploy a fully functional app—all without ever leaving your browser or personally writing a single line of code!

Wand - AI-powered design tool for iOS to draw and edit anything. Upload reference material and Wand will create a custom private style based on those images. Use that style to draw anything you can imagine.

AI Studios - AI-powered platform that simplifies video creation. It converts text into videos with realistic avatars and supports multiple languages.

Inncivio - Transforms your training material into fun, bite-sized knowledge, completed with personalized learning paths that ensure easy upskilling and retention.

Renamify - AI-powered tool renames your files intelligently. Start organising your memories for free, today.