Promises meet workflows

Today’s signal is sharper than yesterday’s enterprise-deployment story: AI is being tested in the places where promises become operational reality — finance workflows, banking compliance, CFO operations, and consumer-facing product commitments. The frontier labs are racing into regulated workflows, while Apple just learned that marketing AI before it works can carry a real price tag.

• Anthropic is making its biggest Wall Street push yet. Anthropic announced ten ready-to-run agent templates for financial services and insurance, aimed at work such as building pitchbooks, preparing meetings, reviewing earnings, building models, conducting market research, screening KYC files, reviewing valuations, reconciling general ledgers, closing the books, and reviewing financial reports. Each agent ships through Claude Cowork, Claude Code, and Claude Managed Agents, which means Anthropic is trying to reduce financial-services deployment from months to days. Axios framed the move as Anthropic working to cement itself as Wall Street’s go-to AI lab. (anthropic.com)
• The real move is not “Claude for banks”; it is Claude as financial workflow infrastructure. Anthropic says the new finance agents are available through a financial services marketplace and are paired with Claude Opus 4.7, which Anthropic says leads the Vals AI Finance Agent benchmark at 64.37%. The strategic point is that Anthropic is not selling a generic assistant into finance; it is packaging repeatable finance work into deployable agent patterns. (anthropic.com)
• OpenAI answered with the CFO office. PwC announced an expanded collaboration with OpenAI to create an AI-native finance function at enterprise scale, including agents for forecasting, planning, reporting, procurement, payments, treasury, tax, and accounting close. PwC and OpenAI are also building a procurement agent inside OpenAI’s own finance organization, using OpenAI as “customer zero” before applying those learnings to enterprise clients. (prnewswire.com)
• This turns the OpenAI–Anthropic enterprise race into a vertical race. Yesterday’s story was that both labs are building enterprise deployment vehicles. Today’s story is that financial services and finance operations are becoming the first major proving ground. Anthropic is targeting banks, insurers, asset managers, compliance workflows, and financial data ecosystems. OpenAI is targeting the CFO stack through PwC. The shared thesis is clear: the money is in high-value, repeatable, regulated workflows where AI can be measured against time saved, risk reduced, and throughput increased.
• The market is already repricing financial-data moats. FactSet shares reportedly fell after Anthropic launched the ten finance-focused agent templates, and separate coverage noted pressure on other financial-data names as investors considered whether AI agents could move closer to the analyst workflow than traditional data terminals and research platforms. The point is not that FactSet or Morningstar disappear. The point is that investors are beginning to distinguish between owning the data layer and owning the workflow layer above the data. (investing.com)
• Apple agreed to pay $250 million over delayed AI Siri features. Apple agreed to a $250 million settlement over claims that it misled iPhone buyers by promoting Apple Intelligence and upgraded Siri capabilities that were not delivered on the advertised timeline. The settlement, which still requires court approval and includes no admission of wrongdoing, reportedly covers eligible U.S. purchases of iPhone 16, iPhone 15 Pro, and iPhone 15 Pro Max devices during the relevant purchase window. (theguardian.com)
• The Apple story is the consumer-side version of the same enterprise lesson: AI promises now need operational proof. Apple’s settlement is not just an Apple story; it is an AI marketing accountability story. The industry has spent two years selling roadmaps as if they were features. That gets riskier as AI moves from demo videos and keynote promises into purchase decisions, enterprise budgets, compliance workflows, and user expectations. For Orthogonal, the lesson is straightforward: do not overclaim capability. Show the workflow, show the source, show the confidence level, and make the output verifiable.

Orthogonal Take

The market is narrowing around a more serious AI thesis:

1. Financial services is becoming the enterprise AI proving ground — because the workflows are high-value, repetitive, regulated, data-rich, and measurable.
2. The winning AI systems are becoming workflow-native — they live inside banking investigations, finance operations, pitchbooks, reporting cycles, procurement processes, spreadsheets, decks, and audit trails.
3. AI marketing is becoming a liability if the product cannot deliver — Apple’s Siri settlement is a reminder that the distance between “coming soon” and “available now” matters when customers buy hardware, software, or services based on AI claims.

That combination matters. The next phase of AI is not just better models. It is models plus workflows plus evidence plus governance plus delivery discipline.