Large Language Models

Today's Stories

1. 1

   Amex uses AI to speed product development, full workforce impact still ahead

   American Express reported second-quarter net income of $3.11 billion（約5000億円） (up from $2.88 billion（約4600億円） a year ago) and raised its full-year growth projection to 10%, up from a 9–10% range. CEO Steve Squeri said AI is accelerating the company's ability to ship technology products from a large backlog, with early gains in speed to market and efficiency. Amex is investing in AI for both internal operations and customer-facing tools — including using AI to generate hyper-personalized rewards based on individual spending patterns — as it competes for younger customers against fintechs like Block's Cash App and Revolut, and premium card rivals JPMorganChase's Saffire and Citi's Strata Elite. Squeri emphasized that deeper impacts on product development, workflows, and revenue are still to come, and any workforce reduction would occur through attrition rather than layoffs.

   Squeri described the moment as "the preseason," signaling that AI's full impact on the business will emerge over coming years. The company recently completed its "largest investment ever" in upgrading both U.S. consumer and business Platinum cards, including redesigns, new websites, and added incentives, as part of its push to acquire millennial and Gen Z customers.

2. 2

   Meta AI Now Plans and Acts on Tasks via Muse Spark 1.1

   Meta announced Muse Spark 1.1, an updated AI model powering Meta AI and meta.ai that can plan, execute, and follow through on multi-step tasks without needing repeated prompts. Features include kitchen renovation planning, half-marathon training schedules, birthday dinner coordination, daily briefings, and research synthesis—all rolling out today in select markets, with expansion to WhatsApp and other surfaces in the coming weeks. The model shifts Meta AI from answering questions to acting on behalf of users—handling repetitive tasks, tracking calendar conflicts, and delivering ongoing updates automatically. This represents Meta's step toward what it calls "personal superintelligence," an AI that understands context and handles things users would otherwise manage manually, potentially reshaping how people delegate everyday planning and research.

   Meta is deploying these features starting today in select markets via the Meta AI app and meta.ai, with rollout to additional countries and WhatsApp "in the coming weeks." Users can steer tasks in real time (adjusting focus, tone, or content), and all generated content—schedules, slide decks, mood boards—lives in one place for later access and sharing.

3. 3

   Agentic AI token costs soar 24× by 2030; running tasks locally cuts bills 87%

   A simple AI agent consumes up to 15,000 tokens per task; complex multi-agent systems use 200,000 to over a million. Goldman Sachs projects total token consumption will multiply roughly 24 times by 2030, to 120 quadrillion a month. Dell launched Deskside Agentic AI in May, a system that runs production-ready agents on company workstations using open-source models, with governance built in from the start. Between mid-2023 and early 2026, token prices fell 80%, but enterprise AI spending jumped 320% because companies deployed far more agents consuming vastly more tokens. The total bill climbed despite lower per-token prices. For IT teams, the shift means infrastructure, security, budgeting, and governance all have to change — and token strategy is now a board-level question, not just an IT decision.

   Analysis by Signal65 and Futurum shows running agents on-premises saves up to 87% on token spend over two years compared to public-cloud APIs, with break-even in as little as three months. The system handles workflows for coding, research, and private assistants on models from 30 billion to trillion parameters and can migrate to data center servers without redesign.

4. 4

   Solopreneurs warned: which tasks to keep off AI's plate

   Fast Company has published guidance for independent business owners on which work should remain human-controlled rather than delegated to AI systems. As solopreneurs face pressure to automate to stay competitive, the piece argues that certain business functions—likely those requiring personal judgment, client relationships, or strategic direction—carry enough weight that outsourcing them to AI risks damaging the core of the business or client trust.

   The article's specific list of "refusal tasks" will help solopreneurs make deliberate choices about where AI adds value versus where human judgment remains irreplaceable.

5. 5

   Big Tech firms defend open-weight AI models amid regulatory scrutiny

   Major technology companies are publicly defending open-weight AI models—systems where the underlying code and weights are made publicly available—against potential regulatory restrictions or criticism. Open-weight models have become a competitive alternative to closed, proprietary AI systems, allowing researchers and smaller companies to build applications without relying on Big Tech gatekeepers. The defense signals that large companies view this transparency as strategically important and worth protecting.

   The regulatory environment around open-weight models remains uncertain; how policymakers respond to industry arguments will determine whether these models remain widely accessible or face new restrictions.

6. 6

   Apple Store app gains AI shopping assistant

   Apple is adding a "Virtual Shopping Assistant" to the Apple Store app, as revealed by a privacy policy update. The bot will collect account information, device identifiers, carrier information, chat data, and optional location data to personalize responses and help with shopping decisions. The feature lets Apple gather detailed shopping behavior and preferences through the AI assistant's conversations. Users can opt in or out of letting Apple use their chats to improve the assistant, a choice made in Account > Settings > Chat Improvements within the app—though opting out of data collection entirely does not appear to be an option.

   The virtual assistant is not yet live in the Apple Store app. Apple has previously added an AI bot to the Apple Support app, though that feature also remained unavailable to all users.