
A Deeper Look at Agentic AI and the Future of Finance

What is Agentic AI?
Agentic AI isn’t just another advanced chatbot or a smarter version of generative AI with a few plugins. It represents a fundamentally different way of structuring workflows—one built around systems that can initiate actions, plan intelligently, adapt to changes, and execute multi-step processes without constant human guidance.
Think of it like the difference between using a checklist with an intern and delegating an outcome to a trusted team lead. The intern needs precise, step-by-step instructions. The team lead knows how to interpret the objective, resolve issues along the way, and deliver the result without needing constant oversight. Agentic AI behaves more like the team lead—it understands the goal, manages the process, and only loops you in when it really needs to.
This shift isn’t about bigger or faster AI models. It’s about autonomy, goal orientation, and intelligent workflow execution.
Get the Latest Insights
A Framework for Finance: Categorising AI Agents
To understand agentic AI clearly, it helps to think about it on a spectrum of capability:
1. Task Agents: Reliable Specialists
Task agents perform single, clear-cut actions triggered by specific events. They’re fast, accurate, and cost-effective.
- Automatically run document checks for reporting accuracy.
- Tag journal entries as soon as they're posted.
- Automatically classify and route uploaded invoices.
2. Autonomous Agents: Self-Guided Problem Solvers
Autonomous agents handle more complex tasks that involve multiple steps or ambiguity, adjusting their approach as they go.
- Reconcile ledger accounts and proactively flag anomalies with detailed explanations.
- Generate and update a dynamic draft checklist for the month-end close.
- Analyse and summarise variances across reporting periods.
3. Orchestrator Agents: Digital Project Managers
Orchestrator agents coordinate several task and autonomous agents, sequencing actions efficiently and reliably.
- Drive the full month-end workflow from data refresh, through variance checks, all the way to the generation of reporting packs and commentary.
- Coordinate financial planning inputs across multiple teams or systems.
4. Interacting Agents: Collaborators Beyond Your Walls
Interacting agents communicate across different systems and even across different organisations. Initially, this might start with simpler interactions, like:
- Verifying documentation with suppliers
- Sharing supporting data and reports with auditors
- Negotiating pricing and contract terms with customers
Over time, we should expect these interactions to evolve into more sophisticated negotiations and collaborations, streamlining processes further.
The Evolution of AI Agents: From Single Solutions to Interconnected Ecosystems
Today: Narrow Agents in Everyday Tools
Task agents and autonomous agents are already automating specific, repeatable actions inside enterprise platforms. Salesforce Agentforce, for example, includes service agents that can handle billing inquiries by retrieving customer data, reviewing billing history, and drafting a summary response. These agents go beyond simple chatbots. They operate within well-defined parameters, but when used smartly, they can eliminate the need for manual case review—delivering real productivity gains with minimal disruption.
Next: Development Frameworks That Enable True Orchestration
Google is setting the pace with its recently announced Agent Builder, which gives developers the tools to build and orchestrate multi-agent systems within enterprise environments. Its Agent Development Kit (ADK) allows organisations to coordinate agents through defined workflows—assigning tasks, tracking outcomes, adapting in real-time, and escalating where needed.
This architecture is built on two critical open standards:
- Model Context Protocol (MCP) acts a bit like a USB-C standard for AI agents. Just as USB-C enables a wide range of devices to plug into chargers, screens, and accessories without custom cables, MCP aims to give AI agents a universal way to connect to enterprise data (ERPs, CRMs etc.), and tools, regardless of platform or model.
- Agent2Agent (A2A) is a communication protocol that lets agents talk to one another across frameworks and systems, much like APIs do for applications.
Together, MCP and A2A form the infrastructure backbone for interoperable agent ecosystems. This open approach means developers can plug agents into any system—whether it's an ERP, a CRM, or a custom internal application—and have them work together seamlessly.
This matters because it unlocks a new kind of flexibility: choose the best agent for each task, link them across systems, and create your own high-performing ecosystem
This matters because it opens the door to assembling your own ecosystem of agents, selecting the best-performing option for each task, and enabling them to collaborate across systems. And if adoption of open protocols like MCP and A2A continues, it’s realistic to expect agents to begin interacting directly with agents from customers, vendors and other stakeholders.
Google has already launched an AI Agent Marketplace featuring ready-to-deploy agents from partners like Deloitte (initially focused on marketing and data use cases). The purchasing experience will feel familiar, as it’s part of Google’s broader marketplace, and the software itself will likely resemble traditional tools—just more adaptive and intelligent. Most initial agents are expected to complement—rather than replace—existing platforms, helping teams automate targeted workflows with greater flexibility and precision.
We’re still early—but the infrastructure is here, and the shift is underway.
Practical Steps for Finance Teams Today
To prepare your finance team without overinvesting, focus on low-risk foundational moves:
- Modularise workflows: Structure processes so individual steps can be automated by agents.
- Optimise your data: Ensure data cleanliness and clarity to maximise an agent’s ability to act effectively.
- Pilot simple task agents: Use tools like report accuracy checkers such as Accurate Digits to build familiarity and confidence.
- Monitor your current software stack: Track how your existing platforms like Salesforce, NetSuite, or Workday are integrating agent-based capabilities.
- Stay alert to agent-driven SaaS alternatives: Generic AI agents are already replicating certain functions of legacy SaaS tools—like invoice processing or expense management. Although not yet fully enterprise-ready, these agents may begin to pressure traditional software providers on cost and flexibility. In essence, what you buy today might be built in-house tomorrow, faster, cheaper, and with greater control.
- Keep it practical: The best agent isn’t always the most advanced—it’s the one that reliably gets the job done. In many cases, that might still be a non-agent solution. Choose based on fit, not novelty.
AI Agents in Finance: Closer Than You Think
You won’t need to deploy a dozen agents tomorrow—but you may interact with them sooner than you expect. The best thing finance teams can do now is to stay informed, stay modular, and stay ready to act when the opportunity fits.
The shift is already underway—and the teams that move steadily, not suddenly, will be best placed to benefit.
Next Steps
If you haven’t read our foundational article yet, start here: Finance and Agentic AI.
In our upcoming piece, we’ll share a practical roundup of AI agents already available to finance teams—from Salesforce Agentforce and Google’s Agentspace to Accurate Digits and n8n. Subscribe for updates, and keep an eye out for that guide.