The Answer Engine AI News · Sydney

Cognition, the company behind AI coding assistant Devin, has reached a $48 billion valuation—higher than Cursor's valuation before it was acquired by SpaceX. This massive number tells us something important about how investors are viewing the AI coding tools market: they don't think one player will dominate everything. In typical software markets, when a clear winner emerges, competitors struggle to raise money at high valuations. But investors are still betting big on multiple AI coding companies simultaneously, suggesting they believe there's room for several winners. The market is likely to segment—different tools for different developer needs, team sizes, or coding approaches. Cognition's sky-high valuation, even with strong competitors already in the field, shows venture capitalists think AI coding assistance is becoming a fundamental part of software development, not a niche feature. For context, these are tools that help programmers write, debug, and understand code faster using AI.

Meta has launched Muse, a personal AI agent that requests deep access to users' private data including email, calendars, payment information, and health services. This represents Meta's most ambitious consumer AI product to date, but it also puts the company's troubled privacy reputation front and centre. Meta has faced years of criticism over data handling practices, including the Cambridge Analytica scandal and numerous regulatory penalties. The success of Muse hinges on whether everyday users are willing to hand over intimate personal information to a company that many still view with suspicion. Unlike standalone AI assistants that operate with limited permissions, Muse's value proposition depends on comprehensive access to personal data to provide truly useful, personalised assistance. This creates a fundamental tension: the more access Muse gets, the more helpful it becomes — but also the more risk users must accept. For Meta, Muse is a calculated gamble that enough time has passed and AI utility is compelling enough to overcome lingering trust issues.

Danijar Hafner, an AI researcher, is building a stealth-mode startup in San Francisco focused on creating AI agents that can plan for uncertain situations. While the provided summary offers limited detail about the technology itself, the focus appears to be on developing AI systems that go beyond simple task completion to handle unexpected scenarios and adapt their plans accordingly. This represents a shift from current AI assistants that typically follow predetermined patterns or respond to immediate prompts, toward systems that can anticipate problems and adjust their approach proactively. Hafner's work suggests the next generation of AI tools may be capable of more sophisticated decision-making in dynamic, unpredictable environments. The startup is in early stages, with minimal staff and infrastructure, but the research direction points to AI agents that could eventually handle complex, multi-step business processes where conditions change frequently and human-like foresight is valuable.

Researchers at MIT are now using OpenAI's GPT-5.6 Sol model, combined with Codex, to autonomously run quantum computing experiments. The AI system handles the full experimental loop: it runs tests on quantum hardware, analyzes the results, and even calibrates qubits—the basic units of quantum computers—without human intervention. This represents a significant step in automating highly complex scientific work that previously required constant hands-on expertise. Quantum computing experiments are notoriously difficult because qubits are fragile and require precise calibration, making them time-consuming to work with. By letting AI manage these tasks, researchers can potentially run far more experiments in less time, accelerating the pace of quantum research. While this is still in the research phase and far from everyday business applications, it demonstrates how AI models are moving beyond generating text and images into controlling real-world scientific equipment and making technical decisions autonomously.

OpenAI is highlighting how increasingly capable and affordable AI tools are expanding what individuals and businesses can realistically accomplish. As AI models become more powerful yet cheaper to use, tasks that were previously too expensive, time-consuming, or technically complex are now within reach for smaller operations. This shift means businesses can take on work they would have outsourced, automate processes they previously handled manually, or tackle projects they simply couldn't afford before. The key message is about economic accessibility: AI isn't just getting better, it's getting more cost-effective, which changes the calculus for what's commercially viable. Small businesses can now access capabilities that were recently only available to larger competitors with bigger budgets. This democratisation of AI tools means the barrier to entry for sophisticated automation, analysis, and content creation continues to drop, potentially levelling the playing field across different business sizes.

Common Crawl has analysed over 584,000 llms.txt files and found widespread confusion about what this new file format actually does. The llms.txt format was designed to help sites share information *with* AI systems—things like what your site does, what content you want highlighted, and useful links. It's meant to be helpful documentation, not a blocking mechanism.

However, Common Crawl found many sites are treating it like robots.txt, attempting to use it to block AI crawlers or set access rules. That simply doesn't work—llms.txt has no enforcement capability. AI companies' crawlers aren't required to respect it, and most don't even check for blocking instructions there. Common Crawl also found that many llms.txt files were copied from templates with no actual links or useful information, and some contained nothing at all.

The confusion appears to stem from the similar name and file structure to robots.txt, but the two serve completely different purposes. If you want to block AI crawlers, you still need to use robots.txt or other technical controls. The llms.txt file is purely for providing helpful context to AI systems that are already accessing your content.

ChatGPT launched its advertising platform six months ago, but advertisers are still flying blind when it comes to measuring success. Cost-per-click data is all over the map – ranging from single digits to $22 – but that's where the clarity ends. The platform doesn't offer auction insights, so advertisers can't see how competitive their bids are or what others are paying. There are no industry benchmarks to compare performance against, making it impossible to know whether a campaign is doing well or poorly relative to others. Reliable audience data is also missing, leaving advertisers guessing about who's actually seeing and clicking their ads. This lack of transparency is a stark contrast to mature platforms like Google Ads, where detailed performance metrics and competitive intelligence have been standard for years. For businesses considering ChatGPT advertising, this means you're essentially experimenting without a roadmap – unable to determine if your spend is efficient or if your results are competitive.

The United States has accused Chinese artificial intelligence companies of engaging in "malicious" copying of AI technology through what they describe as "distillation activities at an industrial scale". Distillation is a technical process where a smaller AI model is trained to mimic the behaviour of a larger, more sophisticated model—essentially allowing companies to replicate advanced AI capabilities without doing the original development work. US authorities allege Chinese firms are systematically using this technique to copy American AI models, potentially undermining the competitive advantage of US tech companies that have invested billions in developing cutting-edge AI systems. The accusations suggest this isn't isolated copying but rather an organised, large-scale effort to replicate proprietary AI technology. This escalates existing tensions between the US and China over technology leadership and intellectual property. The allegations come as both countries compete for dominance in artificial intelligence, which is increasingly seen as critical to economic and national security. For AI companies globally, the claims highlight growing concerns about protecting sophisticated models from being reverse-engineered through distillation techniques.