Why Asking Artificial Intelligence Beats Telling It What to Do

Most professional services firms approach artificial intelligence the same way they approach every other piece of software: by writing detailed instructions, mapping every step, and expecting the system to follow the script exactly. That instinct made sense for spreadsheets and rule-based automation. It breaks down with AI agents. According to Rich Kent of TAINA Technology, the real unlock isn't tighter control — it's asking artificial intelligence better questions instead of handing it a rigid playbook.

This distinction matters more than it sounds. Teams that treat AI agents like traditional software often get exactly what they asked for and nothing more. Teams that shift toward asking artificial intelligence what it needs, what it recommends, and where the risks sit tend to uncover gaps in their own thinking — gaps a rigid workflow would have quietly buried.

The Old Habit: Instructing AI Like Conventional Software

AI is now woven into everyday professional workflows — tax teams use agents to research regulations, analyze large datasets, draft reports, and manage repetitive research tasks. But the tools behind these workflows aren't conventional automation. Traditional software executes fixed logic: if X happens, do Y. AI agents, by contrast, are built to reason through an objective, weigh dependencies, and propose their own approach to a problem.

The problem is that most organizations don't manage them that way. The instinct is to write a detailed process map: define every stage, specify every handoff, and dictate exactly how information should move from one step to the next. This produces consistency, but it also caps the agent's ability to flag inefficiencies, question flawed assumptions, or suggest a better path — the very things that make an AI agent worth using in the first place.

A Real-World Test: Detailed Workflow vs. a Single Objective

Kent tested both approaches while building an agentic workflow for software development. His first attempt was exhaustive: a full process map covering activities, checkpoints, documentation requirements, and how information should flow between stages. The agent followed it — but the results exposed the limits of that method. Some parts of the process were unnecessarily rigid. Others baked in assumptions about dependencies that weren't actually required. In short, the workflow reflected how a human happened to design the process, not how an AI system might have optimized it.

The second attempt looked very different. Instead of a script, Kent gave the agent a single objective and two open questions: what information would it need to complete the task effectively, and how would it recommend structuring the process? 

When asked what it needed, the agent identified gaps around stakeholder objectives, testing expectations, governance requirements, feedback loops, and deployment constraints — items the original process map had glossed over. It then proposed a workflow built around iterative validation cycles, automated quality checks, continuous documentation, and structured feedback loops, rather than a single linear sequence of steps.

What Changes When You Start Asking Artificial Intelligence Questions

This experience points to a broader shift in how organizations need to collaborate with intelligent systems. The role of the human moves from process designer to outcome owner. The objective, the oversight, and the accountability stay firmly with people — that doesn't change, particularly in areas like tax due diligence where accuracy, compliance, and defensibility are non-negotiable. What changes is how the process itself gets designed.

For tax and professional services teams specifically, this could reshape how AI gets applied across transaction reviews, risk identification, workpaper generation, and report preparation. Rather than opening an engagement with a fully scripted set of instructions, teams increasingly need to start by 

Why This Shift Reflects a Bigger Change in How Expertise Works

As AI systems become more capable, the advantage stops coming from whoever can write the most exhaustive set of instructions. It starts coming from whoever knows which questions to ask. That's a meaningful change for how expertise gets defined in professional services. Domain knowledge is still essential — but increasingly, it shows up in the quality of the questions a person poses to an agent, not just in the steps they hand it.

TAINA Technology's analysis lands on a simple conclusion: the value organizations get from AI agents depends less on how precisely they can instruct them and more on how effectively they can collaborate with them. Clear objectives, relevant context, and appropriate oversight let an AI agent contribute meaningfully to process design, while human judgment stays exactly where it belongs — at the center of every decision.

Practical Ways to Start Asking Artificial Intelligence Better Questions

Shifting from instruction-writing to question-asking doesn't require abandoning structure altogether — it requires changing what kind of structure you provide. A few practical starting points:

Everyday productivity tools follow the same underlying logic on a smaller scale. Whether someone is cleaning up text with a 

Where Rigid Instructions Quietly Cost Firms Value

It's worth pausing on why the instruction-heavy habit is so persistent. For decades, professional services firms have built their operating model around documented procedures — checklists, templates, and sign-off gates designed to reduce variability and protect against error. That discipline is valuable, and no one is suggesting firms abandon it. The issue is narrower: when that same documented-procedure mindset gets applied directly to an AI agent, it strips away the one capability that makes the agent worth deploying in the first place — its ability to reason about the task rather than just execute it.

A rigid workflow assumes the person who designed it already knew the best path through the problem. That's often true for stable, well-understood processes. It's much less true for tasks involving judgment, incomplete information, or shifting inputs — which describes a large share of the work in tax, audit, and advisory functions. When a firm hands an agent a fixed script for that kind of work, it isn't just limiting the agent. It's importing the blind spots of whoever wrote the script and locking them in place.

This is where asking artificial intelligence open-ended questions earns its keep. An agent that's allowed to say "I don't have enough information to do this reliably" or "here's a dependency you haven't accounted for" is providing exactly the kind of early warning that a fixed checklist cannot. Suppressing that capability in the name of consistency trades a small amount of predictability for a much larger amount of missed insight.

Building a Question-First Habit Across a Team

Shifting an entire team's default behavior from instructing to asking takes more than a one-off experiment. It tends to work best as a habit built in stages, starting small and expanding as trust in the agent's recommendations grows.

None of this removes the need for governance. If anything, it sharpens it — because the questions a firm asks an agent, and the assumptions that come back in response, need to be reviewed with the same rigor as any other piece of professional work product. The difference is that the review now happens on the agent's reasoning and recommended approach, rather than on whether it obeyed a script line by line.

Where This Matters Most in Tax and Advisory Work

Not every task benefits equally from a question-first approach. Highly regulated, well-defined calculations — where the correct method is fixed and known — are still good candidates for tight, scripted automation. The bigger payoff from asking artificial intelligence shows up in tasks that involve interpretation, incomplete data, or multiple plausible paths forward: scoping a due diligence review, structuring a risk assessment, or working through how to reconcile conflicting data sources ahead of a report deadline.

The common thread is that scripted automation still has a place — it just isn't the default anymore. Firms that keep both approaches in their toolkit, and know which situations call for which, tend to get more consistent value out of their AI investment than firms that pick one style and apply it everywhere.

A Note on Adoption Speed

One overlooked side effect of the question-first approach is how it changes adoption speed inside a firm. Detailed, scripted workflows take a long time to build correctly — every edge case has to be anticipated in advance, and the workflow often needs revisiting the moment the underlying process changes. A question-first approach front-loads far less design work. The team defines the objective, the constraints, and the guardrails, and lets the agent handle the parts of process design that used to require a lengthy internal mapping exercise.

That doesn't mean the approach is faster to trust. Teams still need time to validate an agent's recommended workflow against real outcomes before relying on it for anything client-facing. But the design cycle itself — the part that used to consume weeks of workshops and process documentation — tends to shrink considerably once a firm gets comfortable handing an agent an objective instead of a script.

The Bottom Line

AI agents aren't just faster versions of the automation tools professional services firms already use. They reason, weigh trade-offs, and can propose approaches a human process designer might never have considered. Treating them like rigid software caps their value. 

The organizations getting the most out of AI agents today aren't the ones with the most detailed instruction manuals. They're the ones that have learned when to step back, hand over an objective, and let the system show its work. For teams exploring how AI-powered utilities can support that shift in smaller, everyday tasks, the