After discussing NotebookLM’s infographic feature earlier today, I decided to create my own on the difference between generative AI vs. agentic AI.
To prepare for a webinar last week where we discussed agentic AI and how it differed from generative AI, I asked GPT-5 to discuss the differences between the two. I also asked for eDiscovery specific examples of each of them.
The results are below. I then saved that output, loaded it into Google’s NotebookLM and asked it to generate an infographic based on that content. Here is the infographic:
And below is the content from GPT-5 on which it was based. This example (in addition to the one I covered earlier today from Joe Bartolo) illustrates just how powerful these tools are today and how you can use multiple tools to create intuitive and informative content quickly.
Generative AI vs. Agentic AI (Simple Explanation)
Generative AI: Makes content when you ask for it
Generative AI produces something—text, images, code, summaries, answers—in response to a prompt.
It does not act on its own. It waits for instructions.
Think of it like:
A very smart assistant who answers questions and drafts documents but only when you ask.
Examples
- Ask ChatGPT to draft an email → it writes one.
- Ask Midjourney to create an image → it generates it.
- Ask Copilot to summarize a meeting → it produces a summary.
Legal technology examples
- Generate a first draft of a legal hold notice.
- Produce a summary of a custodian interview, long email thread, or deposition transcript.
- Suggest search terms or privilege log descriptions based on provided data.
Agentic AI: Acts on its own to achieve a goal
Agentic AI doesn’t just generate content—it can take actions, make decisions, use tools, and operate across systems to accomplish tasks.
It breaks a goal into steps, takes those steps, and adjusts based on results.
It behaves more like an autonomous worker than a passive content generator.
Think of it like:
A junior associate who not only drafts something but also
- gathers needed files,
- checks systems,
- performs research, and
- reports back—
all without being micromanaged.
Examples
- An AI that notices your inbox is overwhelming and automatically organizes messages, drafts replies, and schedules meetings.
- An AI that books travel by comparing prices, selecting flights, and confirming reservations.
Legal technology examples
- An eDiscovery “agent” that:
- collects relevant Slack channels or M365 sources using APIs,
- runs ingestion workflows,
- applies data normalization,
- launches processing jobs,
- monitors errors,
- and delivers a ready-to-review dataset.
- An AI case-assessment assistant that:
- identifies key custodians,
- pulls their data from multiple repositories,
- clusters communications,
- surfaces key facts,
- and updates an ongoing case timeline as new data arrives.
In One Sentence
- Generative AI = creates content.
- Agentic AI = creates and acts.
Generative AI in eDiscovery (Creates Content, Doesn’t Act Autonomously)
1. AI-Generated Document Summaries
The model reviews long emails, Slack threads, chat logs, or contractual documents and produces:
- concise summaries,
- issue-based highlights, or
- narrative timelines.
Example:
“Summarize this 400-message Slack thread and highlight anything related to potential antitrust behavior.”
2. Drafting eDiscovery Work Product
Generative AI can produce written artifacts such as:
- privilege log descriptions,
- custodian interview questionnaires,
- ESI protocol drafting language,
- deposition prep packets.
Example:
“Draft a privilege log entry for this email chain and suggest the likely basis for privilege.”
3. Suggesting Search Terms or Concepts
The AI analyzes exemplar documents and proposes terms, synonyms, and conceptual expansions to improve recall.
Example:
“Based on these hot documents, suggest 10 additional conceptual clusters or search terms.”
Agentic AI in eDiscovery (Acts Autonomously Across Steps and Systems)
1. Autonomous Data Ingestion & Processing Workflow
The agent—not just a model—executes real tasks:
- connects to M365, Slack, Google Workspace, or file shares,
- identifies and collects specified custodian data,
- launches processing in RelativityOne/Reveal/Nuix,
- monitors job status,
- retries failures,
- notifies the team when the dataset is review-ready.
What makes it “agentic”?
It takes action across systems, handles errors, and completes multi-step workflows without being told each step.
2. Continuous Case Intelligence Agent
An agent monitors the matter and updates work product in real time:
- evaluates new incoming documents,
- automatically re-clusters themes,
- updates a case chronology or fact matrix,
- identifies emerging issues or new custodians,
- triggers tasks such as further enrichment (OCR, translation) when needed.
What makes it “agentic”?
It doesn’t wait for a prompt — it proactively performs tasks when it detects new data or patterns.
3. Automated QC & Validation Assistant
An agent can perform:
- privilege QC sweeps,
- elusion sampling,
- coding-consistency checks,
- identification of likely miscoded documents,
- routing inconsistencies to reviewers.
Example:
“Run nightly QC on all CAL-high-confidence ‘not responsive’ docs. Flag anomalies. Produce a validation report.”
4. Review Workflow Orchestration
The agent coordinates reviewers:
- assigns batches based on progress,
- monitors reviewer performance for consistency,
- escalates unclear documents,
- dynamically tunes CAL classifiers based on reviewed data.
What makes it agentic?
It takes operational actions rather than just generating content.
Quick Side-By-Side
| Generative AI | Agentic AI |
| Creates content (summaries, drafts, suggestions) | Performs multi-step actions across tools and platforms |
| Needs a prompt | Self-initiates based on goals or triggers |
| Helps humans think faster | Helps systems operate autonomously |
| Example: Draft a privilege log entry | Example: Collect Slack data → process → load to review platform |
So, what do you think? Have you tried the infographic capability in NotebookLM yet? Please share any comments you might have or if you’d like to know more about a particular topic.
Image created using Microsoft Designer, using the term “robot lawyers engaged in tug of war”.
Disclaimer: The views represented herein are exclusively the views of the author, and do not necessarily represent the views held by my employer, my partners or my clients. eDiscovery Today is made available solely for educational purposes to provide general information about general eDiscovery principles and not to provide specific legal advice applicable to any particular circumstance. eDiscovery Today should not be used as a substitute for competent legal advice from a lawyer you have retained and who has agreed to represent you.