The Answer Engine AI News · Sydney

Garry Tan, the president of prestigious startup accelerator Y Combinator, revealed on Walter Isaacson's podcast that he spends as much as $200,000 annually on tokens to power his AI agents. Tan, who leads one of Silicon Valley's most influential organisations, described himself as so enthusiastic about working with AI agents that he's experienced what he jokingly calls "cyber psychosis" — sometimes barely sleeping because he's excited to collaborate with these tools. The admission offers a rare glimpse into how deeply some tech leaders are integrating AI agents into their daily workflow, and what that integration actually costs at scale. While $200K per year is well beyond what most businesses would spend, the figure illustrates both the intensity of agent usage among early adopters and the real costs associated with heavy AI deployment. For context, tokens are the units that measure AI usage — every query, response, and interaction with AI tools consumes tokens that users pay for, usually through subscription or pay-as-you-go models.

Trajectory, a one-year-old startup, has raised $45 million in funding and is now valued at $300 million. The company's pitch is straightforward: it wants to make AI model training accessible to people who aren't technical experts. Currently, training custom AI models typically requires data scientists, engineers, and significant technical know-how. Trajectory is building tools that let non-technical users — think subject matter experts, business analysts, or operations teams — train and customise AI models without needing to write code or understand the underlying mathematics. The funding round signals strong investor confidence in the idea that AI development needs to move beyond technical teams if it's going to scale across industries. While details are limited in the announcement, the core promise is democratisation: businesses that can't afford dedicated AI teams could potentially build custom models tailored to their specific needs, data, and workflows. It's an ambitious goal, and at just one year old with a $300 million valuation, Trajectory is betting that simplifying AI training is a massive market opportunity.

Generative AI has democratized enterprise analytics over the past few years, letting employees across organizations access data insights without specialized technical skills. Now, agentic AI—systems that can automate entire workflows and act on their findings—promises to take this further. However, there's a significant hurdle: the quality of insights these AI agents currently produce isn't meeting enterprise standards. The issue centres on whether these automated agents can deliver the depth, accuracy, and reliability businesses need when making decisions or taking actions based on AI-generated analysis. While agentic systems offer the appeal of hands-free analytics that can both interpret data and execute responses, organizations are discovering that automation without adequate insight quality creates new risks. The challenge is ensuring these AI agents don't just work faster, but actually deliver better or equally reliable analysis compared to human-assisted generative AI tools. For businesses considering agentic analytics, this quality gap represents a real barrier to adoption, particularly when automated actions based on flawed insights could have material business consequences.

AI coding agents work very differently from simple chatbots. When you give an agent a coding task, it doesn't just reply once and finish. Instead, it plans the work, writes code, runs tests, spots errors, then loops back to fix problems – sometimes repeating this cycle many times until the task is complete. This means a single coding task can burn through 10 to 50 times more tokens than a straightforward ChatGPT question. The more capable and autonomous these agents become, the more API calls they make and the more tokens they consume. For developers and businesses using AI agents to build software, this creates a pricing headache: you're no longer paying for a simple question and answer, but for an entire back-and-forth work process. The article's core message is that businesses should shift focus from obsessing over token counts to measuring what actually gets delivered – the shipped features and completed tasks that provide real value.

In 2001, Tim Berners-Lee — who invented the World Wide Web a decade earlier — laid out a vision for an internet navigated by autonomous software agents that could handle tasks on our behalf. That vision is now becoming reality. We're entering the era of the "agentic web," where AI-powered assistants don't just answer questions but actively browse websites, compare options, make purchases, and complete complex tasks without constant human direction. Instead of you searching Google, clicking through results, and reading pages yourself, an AI agent does the legwork: it visits sites, extracts information, evaluates choices, and presents recommendations or completes transactions. This shift represents a fundamental change in how people interact with the internet. Rather than navigating the web directly, users increasingly delegate tasks to AI intermediaries that act on their behalf. For businesses, this means the audience visiting your website may soon be software agents rather than human eyes — and those agents will prioritise different signals than traditional SEO optimised for human readers and search engine crawlers.

The New York Times has filed a lawsuit against Brave, the privacy-focused web browser company, alleging it scraped paywalled content and sold access to AI companies. According to the suit, Brave allegedly bypassed the Times' paywall to collect articles, then made this content available to AI firms for training purposes. The Times claims this deprives publishers of both subscription revenue (by circumventing paywalls) and advertising income (by allowing AI companies to use the content without proper licensing). This is part of a broader pattern of media companies pushing back against how their content is being used to train AI systems. The lawsuit adds to growing tension between publishers who invest heavily in original journalism and tech companies that use that content to build commercial AI products. Brave has built its reputation on blocking trackers and ads, but the Times argues the company has crossed the line from protecting user privacy to facilitating content theft. The outcome could set important precedents for how browser companies and AI firms can legally access and use published content, particularly material behind paywalls.

Chris Sachs and Aris Vrakas have outlined a four-step framework for making products discoverable in AI-powered shopping experiences, including Google's AI Overviews and ChatGPT. The approach focuses on four practical moves: making product pages retrievable by search engines and AI systems, using the actual language shoppers use when searching (not just industry jargon), connecting your product feeds directly with your product pages, and regularly testing how your products appear in AI results. This matters because more shoppers are starting their product research with AI assistants rather than traditional search. If your product pages aren't optimised for these AI systems, you simply won't show up when potential customers ask ChatGPT or Google's AI for recommendations in your category. The framework treats AI shopping as a technical challenge with specific, actionable solutions rather than a vague future trend. For e-commerce businesses, this represents a shift from traditional SEO practices to ensuring your products are structured in ways that AI can understand, retrieve, and confidently recommend to users.

Google has confirmed it has no updates to share about adding AI search data to Google Search Console, despite an approaching UK compliance deadline. The UK's Digital Markets, Competition and Consumers Act requires designated platforms to provide more transparency around how their services distribute traffic. Website owners have been asking Google to report clicks and impressions from AI Overviews (the AI-generated summaries that appear in search results) separately in Search Console, similar to how regular search clicks are tracked. Currently, there's no way for businesses to see whether their content is being cited in AI Overviews or how much traffic those citations generate. This lack of visibility makes it impossible to measure the impact of Google's AI features on website traffic or optimize content for AI search results. While the UK law may eventually force Google's hand on transparency, the company's current silence suggests no immediate changes are coming. For now, businesses remain in the dark about how AI search is affecting their organic visibility and click-through rates.

Westpac has disclosed that using AI helped slash the cost of migrating its data warehouse in half — from an estimated $70 million down to $35 million. The bank achieved this by using AI to create a reusable pattern for the migration work, rather than manually handling each component of the move. This approach suggests AI was used to automate repetitive technical tasks involved in shifting large amounts of data and infrastructure to a new system. The 50% cost saving is significant for a project of this scale and points to genuine productivity gains from applying AI to complex, rule-based IT work. While Westpac hasn't released detailed specifics about which AI tools or techniques were used, the creation of a "reusable pattern" indicates the technology learned from initial migration tasks and then applied that knowledge across the rest of the project. For a major bank dealing with strict regulatory requirements and vast amounts of sensitive data, achieving this kind of efficiency gain while maintaining data integrity is noteworthy.