AI Weekly: Claude Gets Stronger and Runs on a Schedule — April 13-19, 2026

Anthropic had a big week. It shipped a noticeably stronger Claude, gave it the ability to run on a schedule without anyone pressing go, and Google made professional-grade text-to-speech almost free at the same time. If you build anything on Claude, or you have ever wanted a task to just run itself every morning, this was a week worth paying attention to.

The Deep Dive: Claude Gets Stronger and Learns to Run on a Schedule

Claude Opus 4.7 raises the bar on coding and vision

On April 16, Anthropic released Claude Opus 4.7, which scored 87.6% on SWE-bench Verified with high reasoning effort and added three times the vision resolution of the previous version. Better vision matters more than it sounds. It means Claude reads screenshots, scanned documents and charts far more reliably, which is most of what real office work looks like.

For any business already automating with Claude, this is a straight upgrade. The same automations get more accurate without you changing anything, and better document reading means fewer of the small transcription errors that quietly creep into reports.

Claude Code Routines let agents run on a timer

The same day, Anthropic launched Claude Code Routines, cloud-based agents you can trigger on a cron schedule, from a GitHub event, or via an API call. In plain terms: a task can now run itself every morning at 7am, or whenever something happens, with no server for you to manage.

That is the piece most small businesses were missing. Automation is only useful if it runs without someone remembering to start it, and until now most SME setups relied on a person clicking go each time.

Why this matters:

• A stronger Claude with better vision improves every automation that reads documents or screenshots.
• Scheduled agents turn one-off AI tasks into reliable recurring processes.
• Google's new Gemini 3.1 Flash TTS topped the voice quality leaderboard at $0.03 per 60 seconds, making natural voice content genuinely cheap.

What This Means for Australian SMEs

The scheduled-agent update is the one to act on. Think of a task your team does at the same time every week, like pulling numbers into a Monday report or checking overnight orders. That is now something an agent can do on its own before anyone gets in.

The cheap text-to-speech is a quieter win. If you make training material, phone menus or content in more than one language, professional narration used to cost real money. At three cents a minute across 70-plus languages, that maths changes completely.

The practical takeaway for this week:

• List the tasks your team does on a fixed schedule. Those are the best first candidates for a Claude routine.
• If you already automate with Claude, retest your key workflows on Opus 4.7. Accuracy likely improved for free.
• Trial the new text-to-speech on one piece of content before paying for voice-over again.

Turning a manual weekly chore into an agent that runs itself is exactly the kind of work we do at IntelliAgent. If you want help setting up your first scheduled automation, get in touch.