Article Series: An Introduction to AI in context of business

  1. What are AI Agents ↗
  2. AI 'Brains': Large Language Models ↗
  3. Prompt Engineering for Business ↗
  4. Tool-Enabled AI Systems

Large Language Models (LLMs) have evolved from being purely conversational systems into intelligent agents capable of taking actions in the real world. At the heart of this transformation is the ability to use 'tools' – functions that allow AI to interact with external systems, databases, and applications.
The Model Context Protocol (MCP), introduced by Anthropic, represents the latest evolution in standardising how AI systems connect to and utilise external tools, much like USB-C standardised device connectivity.

Tool-enabled AI systems combine the reasoning capabilities of LLMs with the ability to execute specific functions – from querying databases to booking meetings, processing payments, or updating CRM systems.
Rather than simply suggesting actions, these systems can now perform them.

Why It Matters

Traditional AI assistants could only provide information and recommendations. Tool-enabled systems fundamentally change this dynamic by allowing AI to become an active participant in business processes. When a customer service AI can not only answer questions but also process refunds, update accounts, and schedule callbacks, it transforms from a support tool into an operational system.

The introduction of MCP addresses a critical pain point: the fragmentation of AI integrations. Previously, each AI system required custom integrations for every tool or data source. MCP provides a universal protocol, enabling AI systems to connect to any MCP-compatible service through a standardised interface. This is particularly significant as businesses increasingly adopt multi-model strategies, using different AI systems for different tasks.

Business Context

The business landscape is experiencing a shift from 'AI as advisor' to 'AI as operator'. Companies are moving beyond chatbots that answer questions to implementing agentic systems that complete end-to-end workflows. This transition is driven by the need for operational efficiency and the desire to automate increasingly complex processes.

Tool-enabled AI systems are appearing across multiple business functions: customer service agents that resolve issues autonomously, financial analysts that pull real-time data and generate reports, HR assistants that schedule interviews and process documentation, and sales systems that qualify leads and update pipelines. The common thread is AI systems that don't just think – they do.

The MCP standard, being open-source and vendor-neutral, is particularly relevant in the current market where organisations are cautious about vendor lock-in. It enables businesses to build integrations once and use them across multiple AI systems, significantly reducing development overhead and future-proofing AI investments.

Benefits

  • Operational Efficiency: AI systems can complete tasks in seconds that would take humans minutes or hours, operating 24/7 without fatigue
  • Reduced Integration Complexity: MCP's standardised approach means businesses can create connectors to their internal systems that work across multiple AI platforms
  • Enhanced Accuracy: Tools provide AI systems with real-time, authoritative data rather than relying on potentially outdated training information
  • Consistency: Tool-based actions follow defined processes, ensuring compliance and reducing variation in outcomes

Challenges

  • Security and Access Control: Granting AI systems the ability to take actions requires robust authentication, authorisation, and audit mechanisms. A misconfigured AI with database access could cause significant damage
  • Error Handling: AI systems must gracefully handle tool failures, API timeouts, and unexpected responses without degrading user experience
  • Cost Management: Each tool call typically requires additional LLM processing to determine which tool to use, increasing operational costs
  • Governance: Organisations must establish clear policies about which AI systems can take which actions, and under what circumstances

Examples

Klarna: The Swedish fintech implemented an AI assistant using OpenAI's function calling capabilities that handles customer service enquiries whilst also performing actions like processing refunds and updating account information. The system reportedly handles the equivalent work of 700 full-time agents whilst reducing resolution time from 11 minutes to less than 2 minutes.

Shopify: Uses tool-enabled AI to help merchants manage their stores. The AI can query sales data, update inventory, modify product listings, and analyse customer behaviour – all through natural language interactions backed by function calls to Shopify's APIs.

Microsoft 365 Copilot: Integrates with the entire Microsoft ecosystem through a plugin architecture. It can read and send emails, schedule meetings, create documents, and analyse spreadsheets by calling functions across different Microsoft services, demonstrating enterprise-scale tool integration.

How Companies Can Incorporate Tool-Enabled AI

Start with Read-Only Tools: Begin by giving AI systems the ability to query internal data sources without modification capabilities. This builds confidence whilst minimising risk.

Implement Clear Boundaries: Define which actions require human approval versus full automation. Use confirmation steps for high-impact operations like financial transactions or customer communications.

Adopt MCP for Future-Proofing: When building new integrations, consider implementing MCP servers for your internal systems. This creates reusable connectors that work across different AI platforms.

Create a Tool Library: Develop a curated set of approved tools with clear documentation, security guidelines, and usage examples. This accelerates development whilst maintaining governance.

Test Rigorously: Implement sandbox environments where AI systems can be tested with tools before production deployment. Include edge cases and failure scenarios in testing protocols.

Summary and Next Steps

Tool-enabled AI systems represent a fundamental shift in how businesses can leverage artificial intelligence, moving from advisory to operational capabilities. The standardisation brought by protocols like MCP is accelerating adoption by reducing integration complexity and vendor lock-in concerns.

For businesses considering this technology, the path forward involves: assessing which business processes would benefit from AI-driven automation, identifying the tools and systems AI would need to access, implementing robust security and governance frameworks, and starting with low-risk, high-value use cases to build organisational confidence.

The technology is mature enough for production use, with major platforms like OpenAI, Anthropic, and Google providing robust function calling capabilities. However, success requires careful planning, strong governance, and a clear understanding of both the opportunities and risks involved in giving AI systems the ability to take action.

Links

** Notes

  • The article was drafted by an AI Agent.
  • Then reviewed by a human and published - (Us here at Siris!)
  • The topic research and guidance on style, tone and audience were built into the agent.