Managing AI Agents: Why Clear Commands Beat Clever Conversations

Artificial intelligence is rapidly becoming part of the tax due diligence toolkit. From analysing transaction data and reviewing tax positions to identifying risk areas and drafting reports, AI agents are beginning to support professionals across the deal lifecycle. 

Yet many firms are discovering that success with AI is not simply about choosing the right technology. It is about learning how to manage it effectively. 

One of the most important lessons emerging from early adoption is surprisingly simple: when directing an AI agent, use imperative language. 

In other words, tell the agent what to do rather than describing what you would like it to think about. 

This may feel counterintuitive. After all, many of us have spent years perfecting polite, nuanced communication in professional settings. We are trained to ask questions, explore possibilities and leave room for interpretation. 

AI agents, however, tend to perform best when instructions are direct. 

Consider the difference between these two prompts: 

“I was wondering if you could take a look at the deferred tax calculations and see if there are any issues.” 

versus 

“Review the deferred tax calculations and identify any errors, inconsistencies or unusual assumptions.” 

Both requests are intended to achieve the same outcome. However, the second provides a clear instruction with a defined objective. The agent does not need to infer intent or interpret ambiguity. It simply executes the task. 

This approach is particularly valuable in tax due diligence, where ambiguity often creates inefficiency. Whether reviewing tax returns, analysing historical liabilities or assessing transaction structures, AI agents generally produce more reliable results when given explicit directions and measurable outcomes. 

Of course, this does not mean every prompt should be reduced to a single command. Context, objectives and constraints remain essential. In fact, the quality of an AI agent’s output is often directly related to the quality of the context it receives. 

However, once that context has been established, agents generally perform better when the task itself is expressed as a clear instruction rather than an open-ended conversation. 

Of course, there will be occasions when you are unsure about the exact methodology the agent should use. Perhaps you know the result you need but not the process required to get there. 

In these situations, many users revert to vague or exploratory language. A better approach is to remain imperative while signalling uncertainty. 

For example: 

“Identify the most effective method for testing FATCA compliance across the dataset.” 

or 

“Determine the appropriate analytical approach for reviewing our straight-through processing controls.” 

Notice that these instructions still direct the agent to perform an action. The user is not asking the agent whether it has an opinion. The user is assigning a task and asking the agent to apply its reasoning capabilities to determine the best approach. 

Think of it as the difference between assigning a piece of work and having an informal discussion about it. AI agents generally perform better when they receive a defined assignment, a clear objective and an expected outcome. 

As AI becomes more deeply embedded within tax due diligence workflows, professionals will need to develop a new skill set: agent management. 

The firms that gain the greatest value from AI are unlikely to be those with the most sophisticated technology alone. They will be the organisations that learn how to communicate with that technology effectively. 

The good news is that the adjustment is relatively straightforward. Be direct. Be specific. Use action verbs. 

And when you’re unsure how the work should be done, remain imperative and allow the agent to propose the approach. 

It may seem like a small change in communication style, but in the world of AI-assisted tax due diligence, clarity is often the difference between an agent that delivers meaningful value and one that simply produces more output. 

Managing AI effectively is not only about understanding what the technology can do. It is also about learning how to give it work to do in the first place.

If you’d like to see how TAINA can simplify and streamline your CARF and CRS compliance journey, we’d be delighted to request a demo.

To stay up to date with our latest insights on tax compliance, automation and regulatory change, sign up for our industry newsletter.