The great promise of AI in the enterprise is to make internal processes more efficient and to make human employees more productive. AI Agents that can work autonomously and creatively to achieve outcomes on employees’ behalf, operating alone or in collaboration with other relevant agents in a multi-agent architecture, represent the most promising way to do that. 

The potential of AI agents that can be orchestrated across your company is particularly immense for internal teams like procurement, legal, and IT, whose key processes are inherently cross-functional and touch many different technology environments. 

Note: Tonkean’s AI agents give employees access to a full staff of specialised AI experts, surfaceable directly from within whatever application environment they work in—email, Slack, Teams, etc. Humans can set guardrails for what the agents can and cannot do, and escalate critical decisions to the right people for review, when necessary, while at the same time empowering each agent to determine proactively how to meet its responsibilities and achieve the goals set for it—all without ever compromising on governance, auditability, or compliance. Here are just a few Tonkean AI agents our enterprise clients are already making use of: 

• NDA Requests Agent. 
• Sourcing Agent. 
• Contract Manager Agent.
• IT Service Triage Agent.
• Policy Support Agent.
• Meeting Prep Agent. 

Yet many enterprises are still unsure exactly how to make the most effective use of agents, or about what truly enterprise-grade agents should look like, or what kind of internal infrastructure integrating them into your operations effectively ultimately requires. 

This Handbook gives you a place to start. Here are the essential steps for getting started with AI Agents in the enterprise, and the technical requirements for doing so successfully. 

1) Configure AI agents to orchestrate complex work across all systems, data sources, and departments so they can deliver on long-term initiatives

• What does “good” look like? Agents are integrated with every piece of technology your organization uses, so that they can perform actions and query information across all your organization’s systems and teams; agents are able to operate alone or in collaboration with other relevant agents in a multi-agent architecture; agents are able to perform actions and query information across all of your systems; the data sources AI agents are connected to are up-to-date.
• What do you need to achieve that? A back-end process orchestration engine to facilitate integrations and help agents interact with a full tech stack at an enterprise level; agentic orchestration capabilities that allow specialized agents to collaborate with each other.

Dive Deeper: In order to derive truly transformational value out of AI agents, agents must be able to carry out processes across many applications and data sources. To do that, agents must be integrated with every piece of enterprise technology your organization might use, from Slack to SAP, cloud applications to on-prem databases and in-house tools. This allows you to surface intelligent, specialized AI agents directly to employees in the environments where they already work in accordance with how they like to work, and for your agents to complete work that spans many different data systems and departments. Agents do little good if they can only help employees navigate one walled garden. Beware vendors selling agents that are glorified chatbots that can only connect with one platform. 

Agents must also be able to collaborate with each other. This is important because the best agents are highly specialized—calibrated to perform specific actions and access only specific data sources. If you ask an agent to do something it wasn’t built to do, it must be able to orchestrate the work with other agents, that is, find an agent that was built to do that thing. (Search the web, for example.) To do that, you need agentic orchestration capabilities. 

But how do you integrate agents across your tech stack and enable them to collaborate? To do that, you need an orchestration engine supporting the agents on the backend. That’s where something like Tonkean comes into play. Unlike other recent agentic and process orchestration offerings, which generally lack the level of control, accessibility, and interoperability enterprises need to derive true transformational change out of AI, Tonkean can integrate with every kind of enterprise technology your organization uses. Tonkean also comes with over 200 prebuilt integrations, and is adding more all the time. 

2) Scaffold agents with deterministic, rules-based automation to keep humans in control, and assign them clearly defined roles

• What does “good” look like? AI agents operating within clearly defined boundaries and goals, ensuring reliability and control, enabling employees to automate complex workflows while maintaining governance, and bridging the gap for agents between structured processes and intelligent decision-making; agents constructed in such a way that they can employ specific skills and access connected resources to perform relevant tasks—but always within the boundaries of human-established rules; agents trained to escalate critical decisions to the right people for review, ensuring workflows always include a trusted human touch.
• What do you need to achieve that? An agent builder that allows employees to strategically design agents with non-deterministic capabilities in deterministic, rules-based governance structures. 

Dive Deeper: To make AI work in the enterprise, you need structure: process, policies, etc. If you put a gaggle of genius-level AI tools to work inside an enterprise organization and simply tell them to get to work, without structure or guardrails, what you’ll get is chaos. What you need instead, in order to leverage that genius strategically, is to harness it with a mix of deterministic and non-deterministic properties. 

That means you need the ability, when building agents, to define clearly what they can and cannot do. So, too, should you use an orchestration platform that provides you with full visibility into and detailed audit trails for every action your agents take. This is how you ensure compliance—by configuring agents to follow company-specific workflows and guardrails. It’s also how you deliver intelligent automation to employees with guaranteed safety, accountability, and governance.

The skills with which you equip agents, as well as the purposes you design them to serve, are also extremely important in ensuring agents are able to effectively and safely collaborate with other agents via orchestration. If you have multiple agents that have possibly overlapping purposes, the engine may get confused and select the wrong one. To avoid this potential confusion, ensure that each agent has a clear purpose. Avoid giving your agent access to data sources or other resources it doesn’t need to perform its job, as this extra context risks simply extending processing times and increasing the odds of incorrect responses.

You should also, when designing agents, ensure they’re instructed to keep humans in the loop at those critical moments when important decisions are called for. Such human touchpoints can be defined in Tonkean’s no-code process editor (more on that below), but the agent itself also has discretion to ask the human operator for clarification, direction, and decisions.

The best AI agents operate alongside employees to coordinate workflows, execute tasks, and drive outcomes—but they do so while following configurable policies and guardrails. This is the balance of determinism and non-determinism in action. Agents must be able to act on their own.They must also follow rules-based workflows to carry out processes.

3) Ensure agents are always easily accessible for employees

• What does “good” look like? Employees are able to access or “call” agents using a Front Door that can be opened from inside the tools employees already use—like Microsoft Teams, Slack, and email—and that’ll guide employees through appropriate next steps.
• What do you need to achieve that? An AI Front Door that gives employees a single, intuitive entry point for all requests and agent interactions and that generates user experiences on the fly to seamlessly guide employees through any workflow.

Dive Deeper: 

For AI agents to be truly useful to an enterprise, they must be able to execute work across your entire organization and autonomously drive business outcomes. They must also be easily accessible and always easy to use. 

In Tonkean, for example, when an employee wants to buy something but they’re not sure how, or if they want to call an agent for some other kind of purpose, all they need to do is call the Tonkean AI Front Door inside Slack and ask. 

Tonkean will tap whichever agent is most appropriate, depending on the request, and that agent—or whichever other agents that agent decides it needs to collaborate with to give the employee what they need—will guide them to resolution. And the employee will be able to interact with these agents as they would a human administrator, because agents communicate like us. Orchestration and automation handle all the data structuring—the toggling between systems employees used to have to do manually—seamlessly in the background.

4) Give internal teams a way to build agents on their own—and help them get started fast with pre-built agents.

• What does “good” look like? Internal teams like legal and procurement are empowered to build, deploy, and orchestrate agents themselves; internal teams also enjoy access to a library of ready-to-use agents that make it even easier to start automating complex processes right away.
• What do you need to achieve that? A 100% no-code agent builder; a library of pre-built agents.

Dive Deeper: Agents do little good for internal teams if they’re not in charge of their use—if, say, they always need to rely on IT to make changes to them or manage their use ongoing. Transformational gains in efficiency, experience, and performance can only come with agency and agility. 

That’s why whatever agent-builder you invest in should be 100% no-code. It must be configured to use any LLM provider or model based on your business needs. Internal teams must be able to get started using agents quickly and easily, and they must be able to manage and orchestrate them ongoing on their own. 

Internal teams should be able to distribute the deployment and administration of specialised enterprise agents throughout the organisation on their own, putting control into the hands of the people with the right subject-matter expertise and authority.

This is how we designed the Tonkean no-code agent and process editor, and it’s why we offer, alongside the builder, a full library of off-the-shelf agents, including Sourcing Specialist Agents, Buyer Agents, Contract Manager Agents, Purchase Intake Agents, AP Specialist Agents, Market Analysis Agents, Compliance Officer Agents, and many others.