Oracle moves to usage-based AI billing, Claude and OpenAI both stumble on October 27, GM settles over driver data, Accenture screens with agents and Runnings au
By Avik Ghosh, Managing Founder, illuminis
Five stories from the past week that matter to anyone buying, running or regulating AI, with our take on each.
At Oracle's investor day on October 28, Chris Leone, executive vice president of application development, said Oracle will add usage-based revenue to its subscription model, with customers consuming "AI units" as AI does work inside its applications. "We don't have to add more seats in order for us to now generate more revenue," he said. Oracle also said it plans to spend $90 billion to $95 billion on datacenters and cloud infrastructure in fiscal 2027, up from $55.7 billion in fiscal 2026. Source.
Our take: When software vendors bill per AI task, every wasted step and every redo shows up on the invoice. Seat counts no longer cap the bill, so the cost to finish each piece of work becomes the number finance needs to watch and control. That is the case for CompletionPrism™.
On October 27, Anthropic reported elevated errors on Claude Sonnet 5.5 twice. The first incident was opened at 09:14 UTC and resolved at 09:56 UTC. The second ran from 14:42 to 15:17 UTC. Both affected claude.ai, the Claude API, Claude Code and Claude Cowork. The same day, OpenAI's status page reported degraded performance for 6.1 Sol across its Chat Completions, Responses and Batch APIs for about 80 minutes before it was resolved. Source.
Our take: None of these incidents was long, but three on one day, from two providers, show that no single model is always available. A product that can only call one model passes every one of those minutes on to its customers. Failover between models from different providers should be part of the product from the start. See the Embedded License.
On October 29, the Washington state attorney general's office said General Motors had reached a $220 million settlement with 36 attorneys general over collecting driving data, including precise location, speed and seatbelt use, and selling it to data brokers that supplied insurers, without drivers' informed consent. Investigators said dealership staff in some cases accepted terms on customers' behalf, and that GM's OnStar app told more than 370,000 drivers they were not enrolled while their data was still being collected and sold. GM agreed to limit the data it collects and set up a comprehensive data privacy program. Source.
Our take: Consent given by someone else on the person's behalf is not consent, and regulators are treating it that way. Any marketing database built on contacts that nobody actually agreed to share carries the same risk. That is why RelationLens shares a contact only when the employee approves it, and forgets what nobody reviews.
The Register reported on October 30 that Accenture uses AI agents to screen resumes, assess candidates' skills and schedule interviews, and offers the same approach to clients. Kate Clifford, Accenture's global chief leadership and HR officer, said "a human always makes a hiring decision," and claimed the process cut time to hire by 28 percent and improved candidate quality by 30 percent. Source.
Our take: A human making the final call does not settle the question. The candidates an agent screened out never reach that human. When a rejected applicant or a regulator asks why, the employer needs a record of what the agent saw and decided, one that nobody could have changed afterwards. See InferTrust™.
Supply Chain Dive reported on November 2 that Runnings, a farm, home and outdoor goods retailer with 90 stores in 12 states and five distribution centers, is rolling out Relex Solutions' AI platform in stages, starting with store forecasting and replenishment and then distribution centers, with both scheduled to go live by mid-2027. Runnings' director of IT, Terry Kriz, said the goal is exception-based processes "where we're not having to look at every single order," with planners reviewing large, unexpected forecast changes before orders are placed. Source.
Our take: Growing a network without adding planners means the system handles routine orders on its own and sends people only the exceptions. The value depends on getting the exceptions right: too many and planners go back to checking everything, too few and stockouts get through unnoticed. That balance is what StockBalancer™ is built for.
In the News comes out every week. Subscribe from the Insights page to get it by email.