Claude is back on top, and your sessions just got longer. Anthropic dropped Claude Opus 5.5 this week, and the short version is: it's stronger than Fable 5.1, cheaper than Opus 5, and developers who had drifted toward OpenAI's Codex are coming back. If you use Claude Code for development work, the five-hour session limit just got bumped up 20%. That means fewer "sorry, I've lost the thread" interruptions on big tasks. GPT-6 Luna and Sol also shipped at 50% lower prices than their predecessors, but early head-to-head testing has Opus 5.5 winning on code quality. More on the pricing comparison in the headlines below. Full breakdown here.

YouTube is now picking your thumbnails for you. At its Made on YouTube event, the platform announced a set of AI tools baked directly into YouTube Studio. The one worth stopping on: dynamic thumbnails. YouTube will generate three variants, test them on different audience segments, and swap out the underperformer automatically, without you touching anything. Also coming: A/B testing for your video's opening hook, with the system crowning a winner after seven days. Forty million title and thumbnail experiments have already run on the platform since 2024, so the data behind this isn't thin. If you use YouTube for marketing, this is the ad-network optimization loop you used to pay for, now sitting free inside your Studio dashboard. One flag from Shelly Palmer worth keeping in mind: if you have brand approval clauses in any creator agreements, an auto-swapping agent could technically trigger a breach. Worth a quick review before you turn it on. Full announcement from YouTube here.

Your Claude outputs are about to carry an invisible watermark. Starting September 30, Anthropic is rolling out statistical watermarks to outputs from Claude Fable 5, Sonnet 5, and Opus 4.8. Think of it like a fingerprint hidden in word choice, imperceptible to a reader but detectable by the right tool. This is Anthropic's response to the EU AI Act, which requires AI-generated content to be identifiable. No change to quality, pricing, latency, or your API format. If you're on AWS Bedrock, Google Cloud, or Microsoft Foundry, it may take a few extra days to reach you. No action required on your end. But if you build products where clients ask "was this written by AI?", you now have a cleaner answer coming. Details straight from Anthropic's support page.

Tools Worth a Look

  • Vercel AI Gateway: See per-request AI spend broken out by model inside your Vercel dashboard, swap providers by changing one line of code, and cache duplicate requests so you're not billed twice. For technical founders using Vercel's AI SDK. Included in the SDK, no separate account needed.

  • Adobe for Claude (expanded plugin): 80-plus pro-grade tools across imaging, design, video, and documents now inside Claude, with interactive editors for PDFs and Adobe Express designs. For owners already using both Claude and Adobe who don't want to keep switching windows. Requires an Adobe subscription.

  • ChatGPT Voice (upgraded): Now runs on GPT-6 and can access your email, calendar, and Slack, so you can verbally tell it to send a message or build a page. For owners who want a voice assistant that actually does things. Included in ChatGPT Work subscription.

  • Amazon Seller Assistant (upgraded): Automated workflows for inventory, pricing, and account health monitoring running 24/7 with no extra cost. For small business owners selling on Amazon who don't want to pay for a third-party tool to do what Amazon now does natively.

  • Equativ ChatGPT Ads partnership: Brands and agencies can now run contextual and native ads inside ChatGPT conversations. For marketing-minded owners who want early access to ChatGPT's ad inventory before it becomes crowded.

  • rohitg00/ai-engineering-from-scratch: A structured curriculum for learning AI engineering from the ground up, then building and shipping real products. (347 stars today)

  • vectorize-io/hindsight: Gives AI agents a memory that improves over time, so the agent gets better at your specific tasks the more it runs. (1,668 stars today)

  • NVIDIA/Model-Optimizer: A toolkit for making AI models smaller and faster to run, so you spend less on compute at deployment. (44 stars today)

  • FxEmbed/FxEmbed: Fixes broken X/Twitter and Bluesky link previews on Discord and Telegram, so your shared posts actually show images and videos. (182 stars today)

  • obra/superpowers: A practical framework for giving AI coding agents specific skills and a reliable development workflow. (611 stars today)

Also Worth Knowing

-Scott

SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email.