Over the past two years, most organizations have been experimenting with AI assistants.
These systems summarize documents, generate drafts, answer questions, and analyze information on demand. They respond to prompts and provide useful outputs to whoever is interacting with them.
For many professionals, assistants have already become genuinely valuable – a real shift in how individual tasks get done.
But a new phase of AI is beginning to emerge. The conversation across the technology industry is shifting - from assistants to agents. And that shift could significantly change how work is performed inside professional services.
What's the difference?
At a high level, the distinction is simple.
Assistants respond. Agents act.
An AI assistant helps a professional complete a task - summarizing a client document, drafting a memo, analyzing a dataset, suggesting language for a report.
In each case, the assistant provides an output in response to a prompt, and the professional decides what to do with it.
An AI agent operates differently.
Rather than waiting for instructions one prompt at a time, agents are designed to perform sequences of actions across a process.
An agent supporting an engagement team might retrieve relevant documents from prior engagements, analyze current-year financial data, identify potential anomalies, generate supporting documentation, and route findings for professional review – all as part of a single workflow.
Instead of assisting with a single task, the agent helps move work forward through an entire process - a workflow. That's a meaningful difference.
Why agents matter for professional work
Many professional services rely on structured processes. Whether it's an audit, a tax filing, or a compliance review, the work follows a defined sequence and sign off with professional judgement. Assistants can improve individual steps within these processes. Agents have the potential to support the entire workflow.
They can gather information, perform analysis, generate documentation, and route tasks for review - helping teams move work forward while keeping professionals in control of key decisions.
In this sense, agents begin to behave less like tools and more like participants in the workflow itself.
That's a significant shift in how we should think about AI's role in professional practice.
Agents need structure
But there's an important caveat.
Agents can only operate effectively when they genuinely understand the environment they are working within. Professional workflows contain a great deal of structure: engagement methodologies, firm-specific practices, regulatory requirements, review hierarchies, documentation standards.
For agents to operate in professional environments, they must work within systems that provide the workflow where work occurs, the context that informs professional judgment, and the governance that ensures accountability.
Without that structure, an agent may generate outputs - but it cannot reliably participate in the process itself.
This is why the rise of agents is closely connected to the broader concept of the AI Trust Stack.
When these layers are in place, agents can operate in ways that support trusted professional outcomes.
From automation to collaboration
Another way I find it useful to think about agents is this: they represent a shift from automation towards collaboration.
Traditional automation attempts to remove humans from processes entirely.
Professional work rarely allows for that and nor should it.
Judgment, oversight, and accountability remain essential to why professional outputs can be trusted in the first place.
Agents offer something different. They allow AI to collaborate with professionals inside structured workflows - performing analysis, gathering evidence, and generating documentation - while the professional applies judgment, reviews outputs, and makes final decisions.
This creates a working environment where the power of human expertise exploits machine capabilities to coexist while each does what each does best.
The implications for professional software
As agents become more capable, the platforms where professional work takes place will become increasingly important.
Agents need an operating environment where they can access relevant data, understand the structure of the workflow, interact with other systems, and record their actions and outputs.
In other words, agents need to operate where professional work is already happening.
This is why I believe the next generation of professional software will evolve from task-based tools into AI-enabled workflow platforms - environments where agents can operate safely and effectively within the processes professionals rely on. The platform isn't just a place to work. It becomes the system that makes trusted AI possible.
The next layer of the AI Trust Stack
The shift from assistants to agents represents an important step in the evolution of AI.
But it also raises a new question:
Where do agents actually perform their work?
For agents to move tasks forward across an engagement, they need to operate inside the workflows that structure professional work.
That's where the next layer of the AI Trust Stack comes in.
In my next article, I'll explore why workflow platforms are becoming the operating environment for AI in professional services - and why the structure of those workflows will play such an important role in shaping how AI systems evolve.
Because as intelligence becomes more powerful and agents become more capable, the systems where professional work happens will increasingly define how that intelligence is applied. That makes the platform question one of the most important in the sector right now.
