The Captain's Log

AI Weekly – 21.07.25

This week, generative AI gets kinetic: Runway brings full-body motion capture, Midjourney V7 becomes the version with video and voice prompting, Goldman Sachs adds a coding partner, Meta jousts with regulators (again), Anthropic targets finance pros and OpenAI makes AI agents mainstream.

In this week’s briefing:

1. AI in Content and Creativity
2. AI in Marketing and Sales
3. AI in Business and Adoption
4. Tools, Models and Platforms
5. AI Agents and Automation
6. Policy, Power and Ethics
7. AI in Society, Health and Education
8. And Finally…

1. AI in Content and Creativity

Runway’s Act‑Two animates full‑body motion: Runway launched Act‑Two this week, a next generation motion capture AI model that transfers full body, face, hand and head movements from video to animated characters. No special hardware required. Source

Midjourney V7 becomes the default with draft, video & voice: After Midjourney has seen incremental updates over the last few months, the highly respected image generator (and now also a video generator), version 7 has now officially become their standard model. It offers a fast “Draft Mode” (≈10× speed), built‑in voice prompting, and it has just added, fast 5‑second video generation from a single image.  Source

Google's Veo3, is now available in the Gemini mobile app: You can create 8 second video clips - with sound - in your Gemini app, just with a prompt (providing your prompt doesn't fall foul of Google's Guardrails). Source

2. AI in Marketing and Sales

AI outreach triggers US TCPA risk warnings: Reuters reports AI-generated marketing calls are now triggering fines under the Telephone Consumer Protection Act, exposing businesses to legal risk.  Source

71 % of CMOs to invest over $10 m in AI, with caution: Boston Consulting Group (BCG) BCG’s annual Post‑Cannes survey of more than 200 CMOs reveals that 71% plan to invest over $10 Million annually in GenAI in the next three years, but remain wary of ROI and compliance. Source

3. AI in Business and Adoption

Goldman Sachs deploys Devin ai  2.1 coding agent: Goldman Sachs has launched Cognition's “Devin” version 2.1, the AI coding assistant that’s now part of its developer team—handling large codebases under human oversight. Source

Anthropic rolls out Claude for Financial Services: Anthropic unveiled a finance-specific Claude offering for modeling, risk assessment, compliance and integrated data from platforms like Databricks and Snowflake . Source

4. Tools, Models and Platforms

Google Gemini 2.5 Pro, NotebookLM & AI wearables: Google confirmed Google Gemini announcements is now live and widely available, along with a new feature for NotebookLM - a range of curated pre-populated notebooks - from Shakespeare to science - and Gemini AI available on your wrist - Wear OS 4+ watches, to be more precise, all part of the expanded Google AI Pro/Ultra plans. Source

OpenAI's ChatGPT Record Mode arrives on Mac for Plus/Pro subscribers: Record Mode, ChatGPT’s meeting/audio recording and summarisation feature, has rolled out to macOS apps, allowing Plus and Pro users to record, transcribe and analyse audio for meetings, brainstorms etc. Source

5. AI Agents and Automation

ChatGPT Agent Mode launches: OpenAI has released “Agent Mode”, heralding the introduction of 'agentic AI' into the mainstream: ChatGPT can now think and act autonomously using a toolbox of research and operator skills to complete tasks for most users across the world, that is unless you are based in the UK, EU, Norway, Switzerland and Lichtenstein (again). Source

Google’s AI will call businesses directly: Google announced AI-powered “Google AI” will soon begin calling businesses itself to handle bookings, reviews and inquiries for customers. Source

6. Policy, Power and Ethics

European Union AI code: Microsoft signs, Meta refuses: Microsoft has signed onto the EU’s voluntary AI code of conduct, but Meta has declined, citing implementation costs and complexity. Source

Meta says Australian privacy laws are AI blockers: Meta warns that Australia’s strict privacy regulations are hindering its ability to build AI systems, calling for regulatory reform. Source

7. AI in Society, Health and Education

Centaur predicts human decisions: A Nature paper introduced Centaur, a foundation model trained on over 10 million human experiment choices. It predicts behaviour and reaction times and generalises across tasks. Source

Professors monetise AI teaching avatars: India’s Master’s Union now allows professors to license their AI avatars and earn royalties per student - the creator-economy meets education. Source

8. And Finally…

AI Appreciation Day origin story uncovered: Worry no longer, pcgamer has done the hard work for you and traced the quirky origin of AI Appreciation Day - celebrated each year on July 16 - to freelance advertiser Jason Kirton. A passionate advocate for AI ethics and apparently a devoted Elon Musk fan (yes, for some those two subjects do go together), Kirton spent nearly a year camping outside SpaceX’s Starbase, hoping to meet Musk and urge him to take AI regulation seriously. His vigil sparked the now-annual event. Source