AI News Feed
Market watch
NPR News Lists Medicine Cabinet Essentials for Cold and Flu SeasonProofpoint Adds Microsoft 365 Risk Dashboard, Teams Training and AI Assistant for MSPsPHANES AI and Tinno Mobile Launch SeePhys Hub for Full-Body Multimodal Data CollectionMind Savers Pursue Six Speculative Paths to Preserve the Self Beyond DeathPig kidneys frozen for months rewarmed and shown to work after transplantCook Confirms Apple Intelligence China Beta for Late October, Siri AI Excluded
NPR News Lists Medicine Cabinet Essentials for Cold and Flu SeasonProofpoint Adds Microsoft 365 Risk Dashboard, Teams Training and AI Assistant for MSPsPHANES AI and Tinno Mobile Launch SeePhys Hub for Full-Body Multimodal Data CollectionMind Savers Pursue Six Speculative Paths to Preserve the Self Beyond DeathPig kidneys frozen for months rewarmed and shown to work after transplantCook Confirms Apple Intelligence China Beta for Late October, Siri AI Excluded
Products & Applications

Enterprise AI Needs System Connections to Move Beyond Chatbot, TechRadar Says

Enterprise AI investments often fall short because LLMs lack access to core business systems, TechRadar reported. The article points to the Model Context Protocol and accounts payable automation as examples of how connected AI agents can act within business guardrails.

AI News Feed · 2026-10-15 19:10

TechRadar reported that an LLM can summarize information, answer questions and suggest next steps, but on its own it cannot act on that information. Without connections to key business systems, an LLM is essentially a chatbot with a polished writing style, according to TechRadar. A finance team processing invoices illustrates the problem: an isolated LLM might accurately summarize a supplier invoice, but it cannot match the invoice to a purchase order, determine whether it has been paid or see whether it is coded to the correct general ledger account. That information exists outside the AI tool.

For AI to take an active role in business processes, it needs access to the systems where context lives, TechRadar reported. MCP can provide the infrastructure for secure connections. Connecting AI to business context has not always been straightforward. Different applications may require different integrations for the same systems, which becomes complicated as businesses add more AI tools. MCP offers a standardized way for AI applications to connect with external systems and data sources. When AI needs access to a new source, developers can use MCP rather than creating a custom connection for every application and system, according to TechRadar.

TechRadar reported that MCP changes what AI can actually do. With the right connections, AI agents can pull relevant data, use that information to make decisions and take action within boundaries defined by the business. MCP can facilitate bidirectional connections, so AI can also send information back to core business systems. For example, an AI agent might update a record, route something for approval or trigger another step in a process.

The report said MCP does not mean agents should be given free rein. The level of autonomy AI agents can reasonably handle today still varies considerably by task and implementation. Meng Liu, principal analyst at Forrester, said in the report that "agentic branding is widespread, but real autonomy is not." Even so, TechRadar reported that AI agents do not need complete autonomy to take on a more active role in business processes. Accounts payable, or AP, is a useful example of what connected AI looks like in practice.

In AP, the business needs to confirm that item prices match what was agreed to in the contract, that products were delivered, that the invoice has not already been paid and that each line item is coded to the right account, according to TechRadar. If those checks do not happen, there are more opportunities for errors and avoidable losses. Traditionally, those checks depended on people. Even today, when automation is widespread, an exception often ends up in a queue, and an AP employee must dig through purchase orders, receiving records, contracts and past invoices to sort it out.

That manual work comes at a cost. According to Ardent Partners' State of ePayables 2025, cited by TechRadar, the average cost of processing an invoice is $9.84, compared with $2.65 for best-in-class organizations.

TechRadar described how things could change if an AI agent had access to the same data sources as an AP employee. In one example, a restaurant receives an invoice for 20 cases of chicken at $42 per case. The AI agent can compare the invoice with the purchase order and receiving records and determine that only 18 cases were delivered. It can also see that the invoice cost per case is $5 more than the agreed-upon contract price. When the numbers do not match, the invoice will not automatically end up in an exception queue because the AI agent has the context to investigate further and take action.

Depending on the agent's findings and the business's policies and guardrails, TechRadar reported, it could request a credit from the vendor, route the discrepancy to the appropriate employee with recommended next steps and supporting evidence, or clear the invoice for payment if everything checks out. The report said the same concept can be applied to other AP decisions. For example, an agent with access to vendor history data can identify a possible duplicate invoice before payment goes out. One connected to the general ledger and historical coding can determine the appropriate GL codes for a new invoice.

Editor's Summary

TechRadar reported that enterprise AI investments often fall short because LLMs cannot act on business data without access to core systems. The article highlighted MCP as a standardized way to connect AI agents to external systems and used accounts payable to show how connected agents can investigate mismatches and take routine actions. It also cautioned that real autonomy remains limited and varies by task and implementation.