Generative AI produces content when you ask for it: a draft clause, a summary, an answer to a question. Agentic AI takes a goal and carries out the steps to reach it: it reads the documents, decides what to do next, uses tools along the way, and hands back finished work for a lawyer to review. With generative AI, you run the process and the AI helps at each step. With an AI agent, the agent runs the process and you review the result.
Generative AI: one request, one output
Most lawyers' first experience of AI is generative: you type a request and get text back. It is useful for work that fits in a single request:
- Summarizing a deposition transcript or a long email thread.
- Drafting a first version of a clause or a client update.
- Rewriting a paragraph in plainer language.
- Answering a question about a document you've provided.
The limit is that the process lives in your head. You decide what to ask, in what order, what to check and what to do with the answer. Each request starts more or less from scratch, and the quality depends heavily on how the request is written.
Agentic AI: one goal, many steps
An AI agent is given a goal and works through the steps itself. What makes it agentic is that it can plan, act and check: it breaks the task into steps, uses tools such as searching your files, opening documents or comparing text against a playbook, looks at the results, and decides what to do next. A well-built agent also knows when to stop and ask a lawyer.
For example, an in-house agent for incoming vendor contracts might:
- read the contract and identify its type and parties;
- compare each clause against your company's positions;
- mark every deviation and propose a redline in Word;
- draft a short cover note explaining the changes;
- stop and hand the package to a lawyer for review before anything is sent.
In a law firm, agents can draft the first version of a transaction document from the firm's own precedent, prepare a response to a USPTO office action following the decision tree the firm's attorneys already use, or turn a matter file into a status answer sourced from the documents.
The difference at a glance
- What you give it: generative AI takes a prompt; an agent takes a goal.
- What you get back: generative AI returns one piece of content; an agent returns completed work, often several documents.
- Who runs the process: you, with generative AI; the agent, with agentic AI.
- Best for: generative AI suits one-off drafting and questions; agents suit repeat work with clear steps and standards.
- Main risk: with generative AI, a wrong answer to one question; with agents, an error carried through several steps if review points are missing.
Which does your team need?
Usually both. Generative AI is the right tool for the varied, one-off requests that fill a lawyer's day. Agents pay off on work that repeats: the same type of contract reviewed every week, the same questionnaire filled every quarter, the same status question asked on every matter. A useful test is simple: if the work repeats and you could write down the steps, it can be an agent.
Supervising AI agents
Agents don't change a lawyer's ethical duties. ABA Formal Opinion 512 applies the duties to supervise lawyers and nonlawyer assistance (Model Rules 5.1 and 5.3) to AI, and the duty to verify output applies however many steps the AI took. In practice, supervising an agent means:
- a defined scope: which tasks the agent does and which it never does;
- access limited to the matters, documents and systems the task needs;
- a lawyer's approval before anything leaves the firm or commits a client;
- a record of what the agent did and who reviewed it.
Our AI policy template includes a section on AI agents you can adopt, and our guide to avoiding AI hallucinations covers where to put review points.
Why agents are harder to get right
A generic agent follows a vendor's idea of how the work should be done. But every team's process is different: different positions, different precedents, different decision points. An agent that doesn't follow your process produces work you have to redo. Agents also need maintenance. Models change, your standards change, and the work changes, and an agent that isn't kept up drifts.
That is the problem Gist was built for. Gist's deployment team of lawyers and AI engineers maps how your team actually does the work, builds agents that run that exact process inside Word, Outlook or the Gist platform, and keeps them running as models and standards change. See how Gist works or book a demo.
More from Gist
- Free AI Policy Template for Law Firms
- How to Avoid AI Hallucinations in Legal Work
- Will AI Replace Lawyers?
- Which AI Tools Review Contracts Inside Microsoft Word?
- Can Legal AI Draft Using My Firm's Own Precedents and Templates?
- Who Builds and Maintains Custom AI Agents for Law Firms?
- Gist Legal AI Evaluation: Contract Processing and Legal Research
