More people using AI does not tell you whether it is making them money. OpenAI's October 14 report says weekly active users of Codex and ChatGPT Work at EU small and medium businesses grew more than eightfold since May 2026. The report defines those businesses as having 200 or fewer associated users. That is OpenAI's usage data. It is not an earnings report for those businesses. If you want to turn the trend into a paid service, pick a small task with a clear finish line. For example: take a client's approved product spreadsheet and use AI to prepare a draft catalog update. Keep it off the live store until a person checks the prices, descriptions and links. Run a small pilot. Record the time it takes, the corrections required and the cost of the software. Compare that with doing the same job without AI. Then sell the checked deliverable, with a clear scope and price. Do not promise savings before you have measured them. Get the client's permission before putting private business data into an AI tool. The opportunity is not to sell the headline. It is to prove one useful result a customer will pay for. What repetitive job could you test this on before offering it to a client? Source: https://openai.com/index/the-eus-next-chapter-in-ai-is-agentic/

OpenAI's October 13 Higgsfield case study reports that its team is using AI to produce instructional videos over 10 minutes long. Previously, those videos topped out around one minute, according to Higgsfield's CEO. That is a company-reported workflow, not a promise that every account can produce the same result. Check access, costs and output quality before selling it. But the business angle is clear. Stop thinking only about another short clip. Think about one useful lesson a client needs to explain over and over. A product walkthrough. Customer onboarding. A training lesson for new staff. Start with a small paid pilot and a clear scope. Get the client's approved facts, build the outline, and test a short section before producing the whole lesson. Check every claim, every on-screen word and every scene transition. Longer video gives mistakes more room to hide. The goal is not ten minutes of AI footage. It is ten minutes that help someone understand something and take the next step. What does your next client explain repeatedly that could become a useful video? Source: https://openai.com/index/higgsfield-longer-videos-voice/

Anthropic announced Claude Haiku 5.5 on October 7, describing it as a small model designed for high-volume, cost-sensitive work. Here is the business question: are you paying for the biggest model to do your smallest jobs? Sorting product descriptions. Tagging leads. Turning a transcript into a first draft. Those are jobs worth testing separately from your hardest research or coding work. Do not switch because of a launch headline. Take 20 real examples from one repeatable job. Run the same instructions through your current model and a cheaper option available to you. Check accuracy, speed and the time a person spends fixing the output. Cheap answers that need constant repair are expensive. If the smaller model passes your test, use it for that job. Keep the stronger model for work that actually needs it. That is a practical service you can offer a client: test one workflow and show the cost per usable result. No promises of easy money. A measured improvement. What task are you still running through your most expensive AI tool out of habit? Source: https://www.anthropic.com/news

Barclays is using Claude to help process about 120,000 emails a day. Classifying them, adding information, and routing them to the right place. That is the business lesson. A specific job. A specific bottleneck. You do not need a bank as your first client. Find a business that spends hours sorting incoming requests. Propose a small pilot: label the requests, flag missing details, and prepare them for a person to review. Measure the time saved. Check the mistakes. Keep a person in charge of anything that could cost the client money or trust. Do not pitch "AI transformation." Pitch fewer hours spent on work nobody wants to do. A tool subscription is not a business. A useful result someone will pay for can be. What repetitive job could you test with one client this week? Source: https://www.anthropic.com/news/barclays-scales-claude

An Australian university lecturer replaced himself with an AI avatar for an online lecture. Most of the students did not notice. Then it got more interesting. When they found out, they preferred the AI version. The avatar had a smoother voice. The lecture could be updated faster. Students could change the language and the speed. That is the business lesson. Your audience does not care how long it took you to make the content. They care if it is clear, useful, and easy to consume. An AI avatar can turn one person into a content machine. Training. Sales videos. Product demos. Customer education. Multiple languages without recording everything again. The warning is simple. Do not trick people. Use the avatar to improve delivery, not fake trust. The opportunity is not replacing yourself. It is multiplying yourself. Where could an AI version of you make money while you are doing something else? Article: https://www.thetimes.com/world/australasia/article/lecturer-replaces-ai-avatar-university-jhm9d8s0x