Dublin-based CRM provider Onpipeline has added an AI agent that can turn a natural-language request into a prepared sales quote, moving generative AI away from drafting assistance and into transactional work inside the customer-management system.

A salesperson can ask the agent to locate an existing deal, create a quotation for a specified amount, generate the PDF, and prepare the document for electronic signature. Onpipeline says the system asks the user to confirm critical actions before they are carried out.

The workflow is narrow compared with broad claims about autonomous enterprise agents, although its specificity makes it easier to judge. The system is not merely suggesting wording: it is locating customer information, invoking existing CRM functions, creating a commercial document, and preparing the next formal step in the sale.

Enterprise AI moves from answers to actions

The first wave of generative features inside business software largely produced text, summaries, or recommendations while employees continued carrying out the underlying work.

Agentic functions change that arrangement by giving the model access to application tools. Once software can retrieve a record, create a document, change a field, or trigger a workflow, reliability begins to matter more than conversational fluency.

Onpipeline already combines lead and deal management with quotes, electronic signatures, invoices, recurring revenue, and document workflows. Giving an agent access across those functions allows several steps to be compressed without forcing the salesperson through each screen manually.

The quote example is also deliberately constrained. The customer and deal already exist, the amount comes from the user, and the action uses established quote and e-signature functions inside the CRM. That gives the agent less room to improvise than an open-ended system spanning unrelated tools.

Control becomes part of usability

Confirmation matters because a quote is not disposable generated content. Prices, quantities, terms, currencies, and customer details have direct commercial consequences once a document is sent outside the organisation.

A system that prepares the work while retaining an approval checkpoint can reduce navigation and data entry without handing away every decision.

That pattern is likely to appear across CRM and ERP software. Low-risk retrieval and administration can run with limited intervention, while financial, contractual, security-sensitive, or externally visible actions justify stronger controls.

Auditability follows the same logic. If an incorrect quotation is produced, the business needs to establish which request initiated it, which deal record was selected, what information was used, and who approved the result.

Existing enterprise applications have an advantage because they already contain identities, permissions, records, and histories around the process the AI is being asked to automate.

For smaller software vendors, agentic interfaces may also reduce the need to expand application menus indefinitely. Employees can describe an intended outcome while software maps the request onto existing functions underneath.

The risk is that natural language makes ambiguity less visible. Names can match several records, deal information may have changed, and a user can omit important constraints. Reliable systems therefore need validation and exception handling rather than simply a model capable of understanding the sentence.

Onpipeline’s feature is a useful example precisely because the task has a defined beginning and end. The agent removes several mechanical steps while remaining inside a familiar workflow and retaining confirmation before the consequential action.

CRM software has spent decades recording the work sales teams carry out. Agentic features are beginning to perform some of that work themselves, with the pace of adoption likely to depend on whether companies can preserve enough permissions, review, and auditability to trust the result.