I was on a call last week with a marketing director at an AmLaw 200 firm. Fifteen minutes in, she stopped me mid-sentence.
"Guy, I need you to slow down. What's the difference between a custom GPT and an agent? My CMO keeps asking about workflows, and I don't really know what to tell her."
She's not alone. The AI terminology explosion has left many legal marketing professionals feeling like they missed a meeting where everyone else got the decoder ring.
So here's the plain-English breakdown I gave her-and that I've now given to dozens of law firm marketing teams.
Custom GPTs are personalized versions of ChatGPT that you configure for specific tasks. Think of them as ChatGPT with a narrow job description.
OpenAI introduced these in late 2023. Creating them requires no coding.
The process involves providing instructions, uploading reference documents, and defining how you want the GPT to behave. The GPT then follows those guidelines every time you interact with it.
Law firm example: Imagine a custom GPT built specifically for drafting attorney bios. Start by uploading your firm's style guide, examples of approved bios, and instructions about tone and length.
When a marketing coordinator needs to draft a bio for a new partner, they feed the GPT the partner's resume and practice description. The GPT produces a first draft that already follows your house style.
Compare that to generic ChatGPT, where the coordinator would need to explain the style guidelines every single time. Still, it might get something that reads like it was written for a tech startup rather than a white-shoe firm.
Custom GPTs work well for repetitive tasks where consistency is important. They're ideal for writing client alerts, drafting social media posts in your firm's voice, and summarizing case developments for the website.
The limitation: Custom GPTs are conversational. You talk to them, they respond. They don't take independent action or connect to other systems.
If Custom GPTs are ChatGPT's specialized versions, Gems are Google's equivalent for Gemini. Google launched Gems in late 2024 as part of Gemini Advanced.
The functionality is similar. You give a Gem instructions and context, and it maintains that persona throughout your conversations.
Law firm example: A BD team could create a Gem focused on competitive intelligence. Upload your firm's practice descriptions, key differentiators, and information about major competitors.
When preparing for a pitch, the BD manager asks the Gem to compare your firm's capabilities in a specific practice area against two competitors. The Gem draws from its loaded context to provide relevant comparisons.
The main reason to choose Gems over Custom GPTs comes down to your firm's existing tech stack. For firms on Google Workspace, Gems integrate more naturally. Microsoft shops will probably find Custom GPTs make more sense.
Firms running on Microsoft 365 should pay attention to Copilot Studio. It's Microsoft's platform for building custom AI assistants, and it occupies interesting middle ground.
At the basic level, Copilot Studio works like Custom GPTs or Gems. You define instructions, upload knowledge sources, and shape how the copilot behaves. For firms already invested in the Microsoft ecosystem, this is the natural starting point.
Where Copilot Studio differs: it connects natively to Power Automate and other Microsoft services. Your custom copilot can actually take action: update SharePoint, send emails through Outlook, pull data from Dynamics CRM. That pushes it beyond simple conversation into workflow territory.
Law firm example: A marketing team could build a copilot that handles event follow-up. After a client seminar, the copilot could pull the attendee list from Dynamics, draft personalized follow-up emails based on each attendee's practice interests, and queue them for attorney review in Outlook. The attorneys approve or edit, and the emails go out.
The copilot coordinates across systems, handling more than simple question-and-answer interactions.
The trade-off: Copilot Studio requires more setup than a simple Custom GPT. The power comes from those integrations, but someone needs to configure them.
For firms with Power Platform expertise already on staff, this is manageable. For others, it may require outside help to get started.
This is where terminology gets confusing. "Skills" can mean different things depending on which platform you're using.
In the context of AI assistants like Claude, skills are modular capabilities-specialized instruction sets that help the AI perform certain tasks better. Think of them as expertise areas. An AI assistant might have a skill for creating presentations, another for writing in a specific style, and another for analyzing spreadsheets.
Law firm example: At InnovAItion Partners, we've built AI assistants with skills specifically designed for professional services. One skill analyzes RFP requirements and matches them against firm capabilities. Another formats proposals according to common bid standards.
When a BD team member needs proposal support, the assistant draws from both skills to produce something actually useful.
Skills are less visible to end users than GPTs or Gems. You don't typically create skills yourself unless you're building AI systems. But understanding that they exist helps explain why some AI tools perform better at specific tasks than others.
Workflows take us from conversation into automation. A workflow is a sequence of actions that the AI performs automatically, often connecting multiple tools or systems.
Law firm example: Consider your typical press coverage process. Today it probably looks like this: Someone monitors news, copies relevant articles into an email, sends to attorneys for review, waits for responses, then manually updates the firm's press page.
An AI workflow could handle most of that automatically. It would monitor news sources for firm mentions, pull relevant articles, summarize each one, draft the website blurb, and send it to the appropriate attorney for approval. Once approved, update the press page.
The attorney still makes the final call. But the workflow eliminates hours of manual coordination.
Workflow tools vary widely. Some require technical setup, others are designed for marketing teams to configure themselves. The common thread: workflows string together multiple actions that previously required human coordination.
Agents represent the frontier of practical AI in professional services. An agent is an AI that can plan, reason, and take independent action to accomplish goals.
Unlike a chatbot that waits for your next prompt, an agent can break down a complex objective into steps, execute those steps, and adjust its approach based on what it learns along the way.
Law firm example: Consider pitch preparation. An AI agent assigned to "Prepare a pitch for Target Company's IP litigation needs" could handle multiple tasks.
It would research Target Company's recent litigation history and IP portfolio, analyze your firm's relevant experience and wins, and identify which attorneys have handled similar matters. The agent would then draft initial pitch sections based on the research, compile everything into your firm's pitch template, and flag specific areas that need attorney input.
The agent handles the research and assembly. Attorneys review and add strategic insight where it counts most.
Companies are deploying agents like this now in real-world applications. The technology still requires oversight. Agents can make mistakes or pursue unhelpful paths. But their ability to handle multi-step work independently changes what's possible for lean marketing and BD teams.
The straightforward answer: probably multiple tools, matched to specific problems.
Start with Custom GPTs, Gems, or Copilot Studio for:
These tools work well for tasks your team does repeatedly, work that requires consistent voice or formatting, and situations where a human will always review the output. Choose based on your tech stack: OpenAI for flexibility, Gems for Google Workspace, Copilot Studio for Microsoft 365
Consider workflows (or Copilot Studio with Power Automate) for:
Workflows excel at multi-step processes that eat up coordination time, anything involving data moving between systems, and tasks where speed is important and the steps are predictable.
Explore agents for:
Agents shine with complex projects that currently require hours of research and assembly, work where the AI needs to gather information from multiple sources, and situations where you want the AI to think through an approach rather than just execute predefined steps.
Every tool I've described is exactly that-a tool. None of them replaces the judgment that makes great legal marketing work.
The attorneys' insights, the understanding of client relationships, and the creative instinct that turns a dry practice description into a compelling story all remain essential.
AI handles the tedious parts so your team can focus on work that actually requires human expertise.
Technology is moving fast. The terminology will probably shift again by the time you've finished reading this. What won't change is the underlying question: What repetitive work could AI handle so your team can do more strategic work?
Start there, and the tool selection gets easier.
Need assistance in building custom GPTs, Gems, Skills, Workflows, or Agents? InnovAItion Partners specializes in creating these for our clients. Contact me to learn more or for help with your next project.
Guy Alvarez is Co-Founder and Managing Partner of InnovAItion Partners, an AI consulting firm serving law firms, accounting firms, consulting firms, and marketing agencies. Before founding InnovAItion Partners, he built and sold Good2bSocial, a digital marketing agency for law firms.