Big moves from OpenAI, Anthropic, Microsoft, and more

AI Fluency Basics: Learning a few terms that turn confusion into confidence

Something Big Is Happening - Why this moment in AI matters for your business

ChatGPT begins testing ads for free users: OpenAI has started testing ads inside ChatGPT for Free and Go tier users in the U.S., while keeping paid and Enterprise plans ad-free. The ads are clearly labeled, appear at the bottom of chats, and rely on topic-level signals rather than personal user data. Users can dismiss ads or adjust personalization settings. What’s notable is the pricing. OpenAI is reportedly charging around $60 CPM (cost per thousand impressions), which is significantly higher than most platforms where CPMs typically fall in the $5–15 range. They also appear to be working only with advertisers willing to commit $200,000 or more, at least for now. This signals that this is being positioned as premium, high-intent inventory, not broad consumer reach. The bigger takeaway for business owners is that AI platforms are evolving into media channels. When users are actively asking for help or solutions, the intent is different than passive scrolling. Whether or not you ever buy ads here, it’s a reminder that AI is becoming a new layer of distribution, discovery, and influence.

Microsoft's Agent 365 is here: This is Microsoft’s answer to managing AI agents at scale. Think of it as a registry, access control layer, and security framework for your entire fleet of AI agents, whether they were built inside Copilot or brought in from third-party tools. As more teams spin up internal agents for marketing, ops, finance, and support, things can get messy fast. Agent 365 is designed to give organizations centralized visibility and governance so IT and leadership can see what’s running, who has access, and how it’s being used. If you’re experimenting with multiple AI workflows across your company, this is the kind of infrastructure layer that keeps innovation from turning into chaos.

The AI model wars: OpenAI and Anthropic both launched major updates last week, and the timing wasn’t random. Anthropic reportedly moved up the release of Claude Opus 4.6 by 15 minutes to get ahead of OpenAI’s GPT-5.3 Codex announcement. Both companies clearly wanted to own the news cycle.

Harvard claims AI doesn't reduce work, it intensifies it: Harvard Business Review published an eight-month case study suggesting that employees using enterprise AI tools often didn’t see their hours decrease. Instead, expectations increased. Teams worked faster, took on more tasks, and absorbed responsibilities that might have previously justified additional hires. In many cases, employees were feeding prompts to AI during meetings, between calls, and even over lunch. The efficiency gains didn’t automatically translate into lighter workloads. They translated into higher output expectations. This is an important conversation for leadership teams. AI can absolutely create leverage, but without intentional design, it can also amplify pace and pressure. Implementation strategy and a thoughtful rollout matters just as much as the tools themselves.

Google says AI hasn't hurt search revenue: In fact, Google Search revenue growth accelerated every single quarter in 2025. Google is pushing back hard against the narrative that AI is eating into their core business.

OpenAI's Codex app hit major milestones: 1M downloads and 1M weekly active users within a week of launch. OpenAI just hosted a hackathon where winners built OpenCortex (AI research paper generator), Envy (auto-tool builder), and Paradigm (AI coding assistant).

Anthropic is closing a $20B funding round: At a $350B valuation (doubling its initial target), with Nvidia and Microsoft contributing $15B combined. The company's revenue run rate now exceeds $9B.

Meta AI added powerful new agent features: Meta rolled out browser automation, multi-model fusion, and more proactive task management capabilities. In simple terms, their agents can now navigate the web, combine strengths from multiple models, and take initiative on follow-up actions instead of waiting for every prompt. This signals that Meta is making aggressive moves in the agent space. The race is shifting from “who has the smartest chatbot” to “who has the most capable digital worker.” For business owners, the takeaway isn’t to jump platforms. It’s to recognize that agents are quickly becoming more autonomous, more connected, and more useful inside real workflows. The question to consider is how to harness that without creating complexity or losing oversight.

Context Window is AI's working memory. How much conversation it can actively hold at once. You're deep in a chat, referencing details from earlier, and AI responds like you never said any of it? The context window filled up. It literally ran out of space to remember based on its short term memory capacity. Understanding this saves you from assuming the tool is unreliable when it's really just working within its design limits. 

Hallucinations happen when AI confidently provides incorrect information. Fake statistics. Made-up sources. Plausible-sounding details that are completely wrong. This isn't something you can fix by switching tools, but it is continuing to improve with each model release. To combat this it’s important to verify anything that matters. Names, dates, numbers, facts etc. AI drafts incredibly well. It needs your judgment for accuracy. 

System Instructions are rules and guidelines you set once that provide context to every conversation automatically. Tell AI your communication style, your audience, your format preferences. It remembers. No more typing "professional tone, keep it concise, no emm dashes etc..." every single time. Set it once and it applies to everything. This is how you make AI sound like you instead of a generic assistant.

Guardrails are clear boundaries you give AI about what it should and shouldn't do. They reduce risk, keep outputs on-brand, and prevent unwanted or unsafe results. This is especially important in business settings where you need consistency and professionalism.

Grounding means giving AI specific documents, files, or data to base its answers on. Grounding is the same as "using source documents." Grounded AI is far more accurate and useful for real work. It reduces guessing, hallucinations, and generic answers because the AI is working from your actual information instead of its general training. NotebookLM is a good example of a tool that does a good job of working with grounded source material.