Recoleta
GitHub
Source note

Anthropic and OpenAI race to embed engineers inside Wall Street workflows

Authors
dr_dshiv
Source
Hacker News
Published
Collection
Software Intelligence
Enterprise AI AgentsFinancial Services AIForward Deployed EngineeringCode IntelligenceAI Governance
Open originalMarkdown source

Summary

Anthropic and OpenAI are moving engineers into enterprise and Wall Street workflows to turn AI agents into shipped systems. The article says finance is the early test case because many tasks use documents, policies, approvals, and measurable outcomes.

Problem

  • Enterprises can access advanced models, but many have not put them into production workflows with governance, data access, and human review.
  • Financial firms need AI systems that can handle regulated document work such as underwriting, KYC, reconciliation, reporting, and procurement without changing liability rules.
  • Developers face changed expectations: Claude Code and Codex may speed coding, coaching, modernization, and review, while raising new governance and compliance work.

Approach

  • Anthropic is backing a services firm that embeds applied AI engineers with clients to find workflows, build Claude-based systems, and support deployments over time.
  • OpenAI’s DeployCo targets large enterprises with forward-deployed engineers, including about 150 FDEs from Tomoro, more than $4 billion in initial investment, and partners including McKinsey, Bain, and Capgemini.
  • PwC and OpenAI are building finance agents for CFO-office tasks such as planning, forecasting, procurement, payments, treasury, tax, reporting, and accounting close.
  • Anthropic released 10 finance agent templates for tasks including pitch building, KYC screening, month-end close, GL reconciliation, earnings review, and underwriting, with connectors to Dun & Bradstreet, Verisk, SS&C Intralinks, Third Bridge, and Moody’s.
  • The core mechanism is direct workflow engineering: map a real business process, connect the model to trusted data, test against past outcomes, let agents draft first-pass work, and keep humans at review checkpoints.

Results

  • OpenAI says its own finance team processed 5x as many contracts with the same headcount using Codex.
  • OpenAI says an internal IR-GPT tool handled more than 200 investor interactions during a fundraise.
  • Anthropic says Claude Opus 4.7 scored 64.37% on Vals AI’s Finance Agent benchmark, leading that benchmark.
  • PwC cited an insurance deployment where the underwriting cycle dropped from 10 weeks to 10 days through backtesting, co-delivery with human oversight, and agent-produced first drafts.
  • OpenAI’s DeployCo starts with roughly 150 forward-deployed engineers from Tomoro and more than $4 billion in initial investment.
  • Anthropic shipped 10 ready-to-run finance agent templates across Claude Cowork, Claude Code, and Claude Managed Agents cookbooks.

Link