The CFO AI Digest: October 1

Happy October. The major AI players continue to ship upgrades, leaning into consumer use cases and generative media. And in a significant move, a top financial firm mandates AI literacy across the workforce.

Let’s get into it:

1. OpenAI releases flagship video and audio generation model

OpenAI’s Sora 2 generates realistic video and synchronized sound from text prompts, with major leaps in physical accuracy, continuity, and user control. It can now simulate real-world physics like rebounds and missed shots, follow multi-shot directions, and even insert your likeness into scenes. OpenAI is releasing it through a new invite-only social app, where users can remix each other’s content and appear in videos via “cameos.”

CFO takeaway: Days after YouTube launched AI-generated video highlights, Sora arrives as competition to TikTok and Instagram Reels. This reshapes the economics of video content creation. Production costs fall with AI, and competitive benchmarks could evolve from follower counts and views to how easily content can be adapted, reused, and personalized by others. If OpenAI leans into social distribution, it could also become a new player in the ad ecosystem and change how brands allocate spend across platforms.

2. Anthropic launches new, top-performing model

Anthropic released Claude Sonnet 4.5, calling the new model its most capable yet. It leads on real-world coding (SWE-bench), tops computer-use benchmarks (OSWorld), and shows major gains in reasoning, math, and domain knowledge — including finance. New agentic features allow Claude to run code, create files, and handle multi-step tasks for more than 30 consecutive hours. Alongside the model, Anthropic launched the Claude Agent SDK, giving developers the same infrastructure the company uses internally to build agent workflows.

CFO takeaway: A member of Anthropic’s GTM team noted that this launch is relevant for healthcare and biotech companies building complex products: superior long-running agents, enhanced coding capabilities, and better context management through new Developer Platform features. While other AI giants double down on consumer use cases and general chat, Anthropic may try to differentiate itself by going deep on agent infrastructure for technical users, especially in R&D-heavy industries.

3. Citi requires AI prompt training across global workforce

Citi has mandated that the bulk of its global workforce complete a training program on AI prompt writing. The initiative, described by Citi tech head Tim Ryan as teaching a “key skill for the future,” aims to make effective use of generative AI a baseline capability across the organization.

CFO takeaway: Ramp’s head of operations recently pointed out that the limiting factor with AI is no longer the power of models but the empowerment and knowledge of employees. That insight explains Citi’s move. Prompt writing is becoming core job literacy, as extracting value from AI now depends on how well your workforce can interact with these tools.

4. OpenAI adds native shopping to ChatGPT

ChatGPT users can now discover relevant products, confirm payment, and complete single-item purchases inside the chat from U.S.-based Etsy sellers. Over a million Shopify merchants, including Glossier, SKIMS, and Spanx, will soon make their products available on the platform. The technology underlying this experience is “Instant Checkout,” powered by the Agentic Commerce Protocol that OpenAI co-developed with Stripe. It’s an open standard for AI commerce that enables AI agents to securely pass purchase and payment details between users and merchants.

CFO takeaway: With discovery and checkout happening in one interaction, traditional signals from the buyer journey may be more difficult to track. You’ll see the transaction, but perhaps not the query or the influence behind it. Marketing spend may need to shift from audience targeting to agent optimization, showing up where AI does the selling.

5. Meta acquires chip startup Rivos to boost AI infrastructure

Meta is acquiring Rivos, a GPU-focused chip startup, to accelerate its in-house silicon efforts for AI workloads. The company has pledged $72B in capex this year, including $29B for a massive data center in Louisiana, and has been actively scouting chip companies to speed up internal development.

CFO takeaway: Meta’s pursuit of superintelligence depends on scale. That requires more compute, lower cost, and tighter control over the infrastructure that runs its models. Acquiring Rivos brings Meta closer to owning that full stack in yet another signal of AI giants prioritizing vertical integration.

6. Amazon launches Echo devices built for Alexa+

Amazon unveiled four new Echo devices, the Echo Dot Max, Echo Studio, Echo Show 8, and Echo Show 11, designed specifically for Alexa+, its next-gen AI assistant. These are the first Echo products built around “ambient AI” experiences: AI that’s always on in the background and able to act proactively based on sensor input, like motion, sound, and temperature. Early access to Alexa+ now comes out of the box with any new Echo purchase.

CFO takeaway: Like Google’s push to bring generative AI into Pixel and Nest, Amazon is upgrading Alexa from a command-based assistant to an adaptive one. The strategic advantage is shifting to AI platforms that can anticipate user needs and act in context, without waiting for a prompt.

See you next week.

Don't miss these

How to build your team to capitalize on AI

As Ramp strives to save customers more time and money, we have also set the ambitious internal goal to become the most productive company in the world.

4 principles from Ramp’s journey to $1B in annualized revenue

Ramp's mission is to save customers time and money — and move quickly to deliver it. The results are a natural outcome of staying true to that focus.

The CFO AI Digest: September 24

AI infrastructure is expanding, pricing for AI productivity tools is becoming tiered, and GenAI features are moving deeper into daily workflows.