As AI tools become more intertwined in our workflows, one thing is clear: it’s not just about what you use, but how you work with it.
AI isn’t a plug-and-play solution—unlike the deterministic software of yesteryear, it’s a new kind of collaborator. To truly unlock its potential, we have to be mindful of how we communicate, how we manage our time, and how we build in continuous experimentation.
Here are three ways I’ve changed how I work to get the most out of AI—and what I’ve learned along the way.
1. Let’s start with the basics: Prompt engineering
Many have heard of prompt engineering at this point. I mention this still because in our work with CPA firms, we regularly come across team members who are intimidated or confused by the term.
I had the privilege of speaking at Accounting Today’s AI Summit 2025 panel with Samantha Bowling as a fellow panelist, and she said something that stuck with me:
“Prompt engineering is simply the way you communicate with AI.”
This distills the concept of prompting to its core. Just as you wouldn’t talk to your friend the same way you talk to your client or your parents, your communication needs to adapt to the context and audience—except this time, the “audience” is a large language model (LLM) in Copilot or ChatGPT.
In our AI Foundations course for accountants, we teach the GCSE framework (Goal, Context, Sources, Expectations), but truthfully, many frameworks work well. The key is picking one and applying the core concepts consistently.
Pro tip: Be intentional about how you communicate with AI. A simple framework—like a checklist for client onboarding—can help you clarify your prompts. The more deliberate and consistent your inputs, the more accurate and helpful your outputs will be.
2. Act like a manager, not an individual contributor
Congrats, you’ve been promoted to a manager! A manager of AI assistants and agents, that is.
With the rise of agents and assistants, your role shifts from doer to orchestrator, and your work (from personal to business) is no longer a solo act.
Instead of doing everything sequentially, I’ve learned to reframe tasks into discrete units and optimize my entire slate of work for AI assistants to help in parallel. For example, if I know a data entry / formatting project using ChatGPT Agent Mode will take a few minutes, I’ll tee that up and immediately pivot to clearing emails—tasks I don’t need deep focus for, but that still move the day forward. By the time the scrape finishes, I’m ready to pick back up on the next step of that project.
This is classic task batching—the AI-edition.
And don’t limit yourself to one AI assistant. I sometimes use a few (across work and personal, from free to paid tools) to work on projects in parallel. You’re not just working faster—you’re working smarter.
Pro tip: Set up a few standing workflows where AI can do its thing while you shift to something else. Use those AI “processing” windows as intentional micro-sprints for low-focus tasks—like inbox triage or quick approvals.
3. Design for iteration, not perfection—because AI never stands still
A trap I see is that sometimes people find an AI workflow that finally clicks for them, then get upset when for no rhyme or reason it no longer works. This is totally understandable as it takes effort to land the right prompt, the ideal formatting, the best agent setup.
Like it or not, we’re still in the Wild West phase of AI—models evolve fast, and nothing stays static for long. Models update. Interfaces change. Output quality fluctuates. What serves you today may feel obsolete tomorrow.
That’s why it’s important to approach your AI workflows as living systems—things that breathe, shift, and benefit from regular experimentation. You need to build resilience into your routines. That means not just being OK with things breaking, but planning for it.
I try to systematize this experimentation. For example, my two go-to assistants right now are ChatGPT and Copilot—but I always keep two “play assistants” in rotation. This month it’s Grok and Gemini. Next month? Could be Claude or Perplexity. The tools change, but the habit stays.
Pro tip: Create a recurring time block or a habit system to test new tools, features, or prompt strategies. You’re not just troubleshooting—you’re future-proofing.
Final thought: AI isn’t just a tool—it’s a relationship
If you want AI to be a force multiplier, you have to treat it like an entirely different entity that’s constantly evolving. That means changing your communication style, adjusting how you manage your time and queue your tasks, and being open to a bit of chaos in service of long-term clarity.
And maybe most importantly? It means staying curious and flexible. Because the people who learn to work with AI—not just use it—will be the ones who stay ahead of the cur
