Recently we hosted a webinar with the Sourcing Industry Group about AI Agents. Audience engagement was through the roof; so many questions came in, we didn't have time to answer them all during the show. We wanted to try answering all of them in one place.
The main difference between Tonkean’s AI Agents and ChatGPT lies in their purpose, context-awareness, enterprise integration capabilities, and dexterity. ChatGPT is a general-purpose conversational AI model designed to answer questions, draft content, or hold natural conversations across a wide range of topics in a human-like style. While powered by the same LLM technology that powers ChatGPT, Tonkean AI Agents are purpose-built for enterprise operations, designed to conduct complex work across domains like procurement, legal, and IT.
The differences manifest in several ways. ChatGPT does not execute workflows or integrate with backend systems out-of-the-box. Tonkean AI Agents orchestrate entire processes—connecting people, systems (like Coupa, SAP, CLMs), and data using no-code tooling. They can trigger workflows, assign tasks, use tools, collaborate with other agents, and ensure resolution based on enterprise policies.
Tonkean wraps around existing tech stacks with enterprise-grade security, governance, and auditability. It offers flexible deployment options (cloud, on-prem) and full audit trails. ChatGPT alone cannot meet these standards without significant customization and infrastructure layering.
Tonkean AI Agents deliver dynamic guided intake experiences tailored to the requester across technology environments, ensuring requests are compliant from the start (e.g., supplier onboarding, contract reviews). ChatGPT can help explain policies or answer questions, but it lacks the embedded context and enforcement logic needed to drive compliance and adoption within enterprise processes.
Tonkean’s AI Agents are AI-powered process orchestrators, not just conversational bots. They enable real business outcomes like increased adoption, reduced rogue spend, and faster request resolution—all embedded in enterprise workflows, not outside them.
Tonkean Agents do not require training. Instead of needing to train models with labeled data (as you would with traditional AI systems), Tonkean allows teams to:
That means “training” is really process design, which can be done by operations teams without engineering help.
Additionally, Tonkean’s AI Agents come with:
These capabilities mean that much of the agent’s logic and behavior can be applied immediately by configuring rather than training.
Further, Tonkean’s generative AI components (used for request understanding and drafting responses or documents) are already pretrained and require only light prompt customization. You can tune these agents to your team’s language and policies by updating prompt templates—no model fine-tuning necessary.
Most teams can launch their first AI-powered workflows in days or weeks—not months. As one positioning doc puts it: "If you can describe it, you can automate it." So overall: No extensive model training is required. Tonkean AI Agents are configurable, not coded or trained, and designed for business teams to implement autonomously.
Trust in the comparative analysis of multiple suppliers, performed by a Tonkean Agent, is based largely on the quality of the data source the agent has been given access to. If the supplier data is of good quality, up to date, and so on, you can generally trust that the comparison is a fair assessment. However, there’s not really a need for blind trust here—you can always verify by asking the agent where it got certain information from. If an agent’s determination is suspect for any reason, it’s always possible to both ask it for the data it based its comparison on.
Yes, absolutely. The accuracy and ability of an agent stems in large part from the context you establish for it. When configuring an agent in Tonkean, you are able to “fine tune” it. For example, when constructing an agent capable, among other things, of generating memos, you might provide it an exemplary memo as an example of what it should attempt to accomplish, with specific instructions on what about that example is good. This, and ensuring the agent has access to the relevant information, data sources, and processes to be as informed as possible, will all help to improve the performance of the agent.
Both approaches are possible. While leveraging multiple specialized agents tends to be more effective as a best practice, because the agents can be provided with specific examples and instructions, it’s certainly possible to build a more generalized agent that’s able to both identify contract types and perform assessments.
This is a difficult question to answer since it varies by data source type and the sort of agent you want to create, but in general, the more detailed and organized a data source is, the more helpful it will be to an agent and the more quickly you can get something up and running. Ultimately, though, it depends a lot on the anticipated goal of the agent. For example, you can’t expect even a simple Company Policy FAQ Agent to be very useful if it only has access to outdated or poorly organized policy information,
Yes.
Tonkean does not have a proprietary LLM. As a default, Tonkean uses the latest version of GPT via Microsoft Azure OpenAI Service, but our Agents are LLM-agnostic—they can work with whichever LLMs your organization uses.
Tonkean connects to customers’ various internal systems generally via RESTful API or a service account. Though the specific connection method depends on the system in question and how that system is being used by the customer. Some Ariba services, for example, are only accessible via SOAP API, so that’s the connection method Tonkean uses.
Tonkean can be rendered to orchestrate across every internal system your organization uses.
Tonkean orchestrates processes across these systems through its stateful architecture, monitoring each system for changes and responding to these as needed. For example, Tonkean will update the status of a Salesforce Opportunity when a particular condition is met, or send a reminder to the relevant stakeholder in Slack when a purchase review has been waiting for three days. Tonkean agents add another dimension to Tonkean’s stateful architecture in that they have this same visibility across systems they’ve been given access to and can proactively take action to accomplish their prescribed goals.
Yes, along with about every enterprise technology your organization might use.
Glad you enjoyed 🙂 We can put in touch with one of our account executives for information regarding pricing. If you’re interested, get in touch. We’d love to chat!
You can find information about our prebuilt agents here. You can also build agents yourself using our agent builder.
No. Tonkean is certified HIPAA compliant.
There are various ways Tonkean can integrate with SAP depending on which flavor of SAP you’re running and how it’s implemented/customized, etc. See our SAP Integration docs topic for more information here.
Once your SAP instance is connected to Tonkean, an Agent can access any of the entities and records you give it access to.
