For law firm patent attorneys navigating the surge of AI tools in legal practice, it’s tempting to view any automation as either a shortcut or a threat. But in patent drafting, where quality, nuance, and alignment with firm expectations are non-negotiable, the key question isn’t whether to use AI—it’s how.
That “how” often hinges on the nature of the loop: AI-in-the-loop versus human-in-the-loop. These phrases are often tossed around loosely, but they reflect fundamentally different models of work—and more importantly, different roles for the attorney.
In AI-in-the-loop systems—sometimes called copilot models—the human user drives the process. The AI sits passively in the passenger seat, offering real-time suggestions or completing snippets of text. Think autocomplete on steroids.
These tools can be useful for brainstorming or speeding up routine drafting tasks. But the quality of the output heavily depends on the skill and attentiveness of the user. An attorney/agent can easily accept boilerplate suggestions without sufficient scrutiny, leading to inconsistencies or errors. Worse, copilots typically lack the contextual memory to maintain alignment across a full application, let alone across a firm’s portfolio.
In practice, AI-in-the-loop systems are often best for individual productivity boosts, not for driving quality or consistency across a team or firm.
By contrast, human-in-the-loop workflows invert the relationship. The AI takes on the role of a junior associate—capable of executing a structured drafting task end-to-end—while the attorney oversees and guides the work. These systems, sometimes called agentic AI, operate with a mission orientation. Instead of responding to reactive prompts, they interpret objectives, apply firm-specific standards, and produce coherent work product with internal consistency.
The attorney remains in control but directs their focus on review, refinement, and strategic judgment rather than mechanical authorship. In this model, quality is not only preserved—it can be institutionalized. Agentic systems can be configured to emulate the preferred drafting styles of a firm or client, reuse best-in-class phrasing, and adapt to different technical domains without losing structural fidelity.
In practice, human-in-the-loop systems enable centralized control over the drafting process while freeing attorneys from the grind work that rarely benefits from their highest-value expertise.
The “loop” metaphor only works if we clarify who or what is leading the process. In AI-in-the-loop systems, the attorney is doing the heavy lifting of prompting, editing, and steering output on the fly. The AI is reactive. In human-in-the-loop systems, the AI is doing the heavy lifting of synthesizing, structuring, and drafting autonomously, while the attorney supervises and elevates the work.
From a workflow perspective, this distinction has enormous consequences:
The most promising model is not simply AI-in-the-loop or human-in-the-loop; it’s expert-in-the-loop. This approach combines the scalability of agentic AI with the oversight and judgment of experienced practitioners. By configuring AI agents to handle the mechanical aspects of drafting and placing attorneys in a supervisory role, firms can deliver high-quality, firm-aligned applications faster and at lower cost.
The result is a redefined workflow: one that preserves attorney judgment where it matters most, leverages automation where it’s most effective, and ultimately raises the bar on both efficiency and quality.
For law firms seeking to adapt without compromising standards, the goal shouldn’t be to write less, but to write smarter. And that starts by putting the right expertise in the loop.
