AI Stopped Talking. Now It’s Working.

Enterprise AI is moving beyond its early chapters.

The first phase was experimentation. Organisations tested generative AI in isolated pilots: chat interfaces, search tools, content generation, coding assistants, and internal productivity use cases.

The second phase was feature inflation. Every SaaS vendor attached copilots, assistants, summaries, or “AI-powered” labels to existing products.

The third phase is beginning to emerge, unevenly, and in early form. It is not about adding AI to software. It is about rebuilding software around agents that can act.

Most organisations are still navigating phases one and two. But at the platform and infrastructure level, the architectural bets are already being placed.

Two recent developments illustrate the direction of travel.

monday.com has repositioned itself from a work management platform into an AI Work Platform - the company’s largest strategic change in its history, with agents built natively into the platform and designed to operate across workflows, live business data, permissions, security, and governance.

Anthropic, meanwhile, has launched ten finance-specific agents for workflows including pitchbooks, KYC screening, credit memos, financial modelling, statement review, and month-end close, while expanding Claude’s integration into Microsoft 365 and regulated financial data ecosystems.

Individually, these are product announcements. Strategically, they point to something larger: the beginning of a reorganisation of the enterprise software stack around systems of action.

The Old Model Was Interface-Centric

Traditional SaaS was built around the human. A worker opened an application, entered data, clicked through workflows, created reports, triggered approvals, and coordinated with colleagues.

The software organised work, but the human remained the execution engine.

That model created enormous value. It produced systems of record, systems of engagement, and systems of workflow.

Agentic AI challenges the boundary between software that supports work and software that performs work.

The emerging model looks different: intent → agent → tools → data → governed execution → human oversight

This does not eliminate people. It changes where they sit in the workflow, shifting from manual execution toward supervision, exception handling, judgement, approval, and design.

Why monday.com’s Move Matters

monday.com’s announcement should not be read only as a product upgrade. It is a defensive and offensive platform strategy.

The defensive logic is clear: if external AI agents become the primary interface through which employees coordinate work, monday.com risks becoming a background database ; valuable but invisible.

The offensive logic is equally clear: monday.com already sits inside operational workflows. It holds customer data, project context, task structures, permissions, automations, and cross-team visibility. That makes it a natural candidate to become an agentic work execution layer.

This is why the distinction between “AI feature” and “AI-native architecture” matters. A chatbot attached to a legacy SaaS product can answer questions. But an agent-native platform must understand business context, execute tasks, respect permissions, coordinate workflows, integrate with external systems, and maintain audit trails.

That is not a user-interface challenge. It is an architecture challenge.

Why Anthropic’s Finance and Legal Agents Matter

Anthropic’s finance announcement is strategically different but directionally aligned. Anthropic is not simply offering a general-purpose model. It is packaging Claude into domain-specific workflows: finance agents, Microsoft 365 integration, governed data connectors, and partnerships with financial infrastructure players.

Finance is not a casual AI adoption environment. It is regulated, data-intensive, documentation-heavy, compliance-sensitive, and workflow-rich. The FIS partnership is especially significant because it moves agentic AI into financial crime workflows, beginning with AML-related use cases with human oversight built in. This is not productivity software. It is AI entering regulated operational infrastructure.

Around the same period, Anthropic also introduced specialised agents for legal professionals, supporting research, document review, and other knowledge-intensive workflows.

Taken together, these initiatives signal something broader than a product expansion.

Anthropic appears to be moving beyond the business of providing intelligence toward the business of operationalising expertise.

Generating an answer is one part of the process. Embedding that answer into enterprise operations - with permissions, governance, accountability, and workflow context - is an entirely different challenge, and one where specialised agents begin to show their real value.

This is, in one sense, vertical SaaS by another name: a well-established model of packaging capability into domain-specific workflows.

What is different is the speed, the breadth of domains being targeted simultaneously, and the degree to which the underlying capability (the model) can be rapidly adapted across sectors.

The move is not unprecedented in structure; it may be unprecedented in scale and speed of execution.

The Data Vendor Dilemma

The market reaction to Anthropic’s finance announcement was immediate. Bloomberg reported that FactSet fell as much as 8.1%, Morningstar lost more than 3%, and S&P Global and Moody’s came under selling pressure.

This does not mean financial data vendors are obsolete , and the market reaction probably overstates the near-term risk. Bloomberg, FactSet, and Moody’s are not simply UI providers. They hold licensed, regulated, and legally defensible data that AI agents still need to source from somewhere credible. That is not easily replicated. Trusted data becomes more important, not less, as AI agents perform higher-stakes work.

But the value chain may shift in ways that matter structurally, even if data vendors remain essential. The question is not whether they survive. It is whether they retain the user relationship.

Historically, data vendors controlled both the data and much of the workflow environment around it. Agentic AI threatens to separate those layers.

In the new model, a financial professional may not begin inside a data terminal. They may begin inside an AI agent embedded in Excel, PowerPoint, Word, or an enterprise workflow platform. The agent retrieves data, synthesises it, drafts outputs, validates assumptions, and escalates exceptions. The data provider remains essential — but the user relationship may migrate toward the agentic operating layer.

That is the real strategic risk for incumbent vendors.

The Governance Challenge

Agentic AI is powerful because it can act. That is also what makes it dangerous.

A chatbot can give a bad answer. An agent can trigger a bad process, move bad data, generate a flawed financial model, escalate the wrong case, approve the wrong workflow, or create compliance exposure.

Agentic AI therefore requires a new governance architecture.

Enterprises will need:

• Identity and access management for agents

• Workflow-level permissions

• Audit logs and data lineage

• Model-risk controls

• Human approval thresholds

• Cost monitoring and explainability

• Incident response protocols

• Clear accountability for agentic decisions

The strategic maturity question is no longer “do we have AI?” It is: Can we safely delegate work to AI inside our operating model?

What Organisations Should Be Paying Attention To

The most important strategic questions for enterprises may have surprisingly little to do with model selection. The harder questions concern control, governance, and dependency.

Where will organisational memory reside? Who controls the orchestration layer connecting agents to business processes? How is accountability maintained when decisions become increasingly agent-assisted? Which capabilities remain genuinely differentiating when analytical work becomes widely accessible?

These are not technology procurement questions. They are questions about future operating models.

Enterprises should avoid two specific mistakes.

The first is treating agentic AI as a chatbot procurement exercise. The second is automating broken workflows.

Agentic AI should begin with workflow diagnosis. Which processes are repetitive, data-rich, rules-based, and currently constrained by manual coordination? Which require judgement, escalation, and human accountability? Which systems contain reliable data? Which workflows are too fragmented to automate safely?

The first agentic use cases should not be chosen because they are impressive. They should be chosen because they are operationally mature enough to govern.

What are the Implications for Public Institutions

The same transformation will reach the public sector, and in some ways the stakes are higher there than in financial services.

Procurement, tax administration, social protection, customs, infrastructure monitoring, public financial management, permitting, and regulatory supervision all contain genuine agentic AI opportunities. The efficiency case is real. But so are the risks, and they compound in ways that are distinct from the private sector.

Public institutions face a sovereignty question that private enterprises do not.

When a bank adopts an AI agent embedded in a vendor’s workflow platform, the strategic risk is competitive. When a government does, the risk is constitutional.

If the agents, models, cloud infrastructure, data connectors, and workflow logic are controlled by a small number of foreign commercial entities, governments may gain short-term operational efficiency while quietly ceding long-term institutional control over how public decisions are made, recorded, and reviewed.

This is not a hypothetical concern. It is already embedded in procurement decisions being made today about cloud infrastructure, data residency, and AI platform contracts.

The governance requirements for public-sector agentic AI are also more demanding than in commercial settings. Auditability, explainability, and accountability are not optional features - they are legal and democratic requirements.

An AI agent that approves a benefit claim, flags a tax return, or triggers an enforcement action must be interrogable in ways that a commercial workflow tool is not.

For multilateral development banks and development finance institutions, the key issue is not simply funding digital transformation. It is helping countries build governed AI operating capacity from the ground up: data architecture, procurement standards, auditability frameworks, human oversight protocols, cybersecurity, and the institutional capability to manage and contest agentic decisions over time.

Funding infrastructure without building that capacity risks creating AI-dependent systems that governments cannot meaningfully control or adapt.

Implications for SaaS Vendors

SaaS vendors now face a genuine strategic fork.

They can add AI features to existing products, or they can redesign their platforms around agentic execution.

The first path protects the current product narrative. The second challenges the product architecture, pricing model, customer success model, security posture, and possibly the company’s identity.

That is why monday.com’s move matters. It signals that some SaaS vendors understand the risk of becoming passive systems of record sitting behind someone else’s agent.

The future SaaS winners will likely share four advantages: proprietary workflow context; trusted enterprise data access; governance and permission infrastructure; and distribution inside daily work.

The Quiet Compression of the Enterprise Technology Stack

Historically, enterprise technology operated through a stable division of labour. Infrastructure providers supplied compute and storage. Software vendors built applications. Consulting firms embedded expertise. Enterprises retained ownership of their processes and institutional knowledge.

Generative AI is blurring those boundaries.

When a frontier AI company packages financial expertise into an operational agent, it is no longer simply supplying a model. It is moving closer to a function traditionally performed by specialised software, consultants, analysts, or internal teams. The same applies to legal agents.

This does not mean bankers, lawyers, or consultants disappear. Such conclusions are usually premature. But it does suggest that parts of professional expertise are becoming codified, reusable, and deployable through software, that specialised knowledge is gradually being transformed into digital infrastructure.

That may become one of the defining economic shifts of the AI era.

The Real Competitive Frontier

For much of the past two years, enterprise AI discussions have revolved around models. Which performs best? Which provider is winning? Which is cheapest to run?

These questions remain relevant, but they increasingly resemble discussions about processors during the early days of personal computing - important, yet not where the majority of future value will be created.

History suggests that control migrates toward layers that coordinate ecosystems rather than individual components.

Operating systems became more influential than processors. Cloud platforms became more influential than individual servers. Mobile ecosystems became more influential than individual applications.

A similar dynamic may emerge in enterprise AI -though it is worth being honest that this remains a thesis, not an established outcome.

The orchestration layer has not yet consolidated. The enterprise AI market is still early and genuinely contested. What we can say with confidence is that the companies investing most heavily right now are not just investing in model capability. They are investing in workflow integration, enterprise memory, and governance infrastructure - which suggests they believe the value will migrate in this direction, even if the timeline is uncertain.

When intelligence becomes abundant, scarcity moves elsewhere. The strategic assets of the future may not be the ability to generate information. They may be the ability to govern, validate, and operationalise it.

This is why monday.com and Anthropic are strategically connected. One is a SaaS platform becoming agent-native. The other is an AI lab moving upward into vertical workflows. Both are converging on the same destination: control over the operating layer of work.

Is SaaS Dead?

The question is too simplistic. SaaS is not dead. But passive SaaS is under serious pressure.

The future will not belong to software that merely stores records or displays workflows. It will belong to platforms that can execute work safely, contextually, and audibly.

The enterprise software stack is being rewritten — not around chatbots, but around systems of action.

Anthropic’s finance and legal agents matter not because they automate a particular task.

They matter because they provide a glimpse into the next phase: a world where value concentrates not around who provides the best model, but around who controls orchestration, workflow ownership, governance, and the operationalisation of expertise.

The strategic question for organisations is no longer whether AI will become part of the enterprise. It is which parts of the future operating model they are willing to entrust to it - and who they are willing to trust to run it.

⸻

SOURCES & FURTHER READING

* Reuters — Analysis of Anthropic’s financial AI agent strategy and enterprise expansion https://www.reuters.com/business/finance/anthropic-deepens-finance-push-with-10-new-ai-agents-banks-insurers-2026-05-05/

* Anthropic Engineering — Claude Managed Agents architecture and orchestration framework https://www.anthropic.com/engineering/managed-agents

* Anthropic Product Announcement — Managed Agents launch and deployment strategy https://claude.com/blog/claude-managed-agents

* Wall Street Journal — Anthropic’s expansion into financial services workflows https://www.wsj.com/tech/ai/anthropic-releases-new-ai-agents-for-financial-services-firms-e2829b37

* Axios — Anthropic’s positioning inside Wall Street AI transformation https://www.axios.com/2026/05/05/anthropic-wall-street-dimon-amodei