Empowering Workshops with Custom GPTs for GenAI Training

by Dinis Cruz and ChatGPT Deep Research, 2025/06/19

Introduction¶

GenAI (Generative AI) has rapidly emerged as a transformative technology, yet many executives and professionals struggle to find practical entry points for its adoption. Often, the barrier is not just understanding what GenAI can do, but experiencing it hands-on. Using OpenAI’s custom GPTs – essentially mini AI applications built on ChatGPT – in workshops offers a powerful solution to this challenge. Custom GPTs enable participants to learn by building, providing an immediate, interactive way to grasp GenAI’s capabilities. In this white paper, we explore why custom GPTs are so effective for workshops, how they evolved into a commoditized tool in the AI landscape, and how to leverage them for executive and technical training. We also incorporate a Wardley Mapping perspective to understand the strategic importance of GPTs in the evolution of AI technology. The goal is to equip organizations and educators with a blueprint for using GPTs to accelerate GenAI learning and innovation in a workshop setting.

The Rise of Custom GPTs¶

OpenAI introduced “GPTs” in late 2023 as a way for anyone to create a tailored version of ChatGPT for a specific purpose. A custom GPT combines a large language model with user-provided instructions, optional domain knowledge, and tool integrations, all packaged behind a shareable chatbot interface. Crucially, no coding is required – creating one is as simple as having a conversation with the GPT Builder, supplying guidance and data in plain language. In effect, OpenAI took the art of prompt engineering and productized it into an easy workflow, moving it from the realm of experts to the general public. This dramatically lowers the barrier for experimentation; a manager or student with a Plus account can spin up a custom AI assistant in minutes.

The impact was immediate. Within two months of launch, users created over 3 million custom GPTs. OpenAI rolled out the GPT Store to index and share these community-built bots, complete with categories and leaderboards for discovery. Early examples ranged from GPTs that can design presentations with Canva to AI tutors for coding and math. By making GPT creation trivially easy and shareable, OpenAI essentially turned ChatGPT into a platform for countless niche applications. As one analysis noted, this move was “the most obvious and inevitable thing” from a strategy standpoint, following a classic Innovate–Leverage–Commoditize pattern in tech. OpenAI commoditized its core product (the ChatGPT AI) by allowing anyone to build on it, thus spawning a wave of user innovation on their platform.

Key Features of GPTs that Enable Workshops¶

Several features of custom GPTs make them particularly powerful for training workshops and live exercises:

Pre-loaded Expert Instructions: Each GPT has a customizable system prompt (the “Instructions” or persona of the bot). This can be a detailed role or guidance that defines how the GPT behaves. Essentially, it allows the creator to bake in expert knowledge or specific processes up front. OpenAI effectively exposed the system prompt as a user-editable field, turning careful prompt engineering into a reusable asset. In workshops, this means participants can craft a GPT that behaves like, say, a marketing coach, a cybersecurity auditor, or any persona relevant to their domain, and see consistent behavior out of the box. It’s a hands-on lesson in how much prompt context matters.
Custom Knowledge Base: GPT builders can upload files (PDFs, docs, data) to give the GPT reference material. Unlike a generic ChatGPT, a custom GPT can have grounding in specific documents – an invaluable feature for enterprise use cases. In a workshop, participants might feed in a product manual, company policy, or research paper and immediately have a chatbot that “knows” that content. This is done without any model fine-tuning; the GPT simply indexes the data and retrieves it as needed (a lightweight RAG – Retrieval Augmented Generation approach). The MIT Sloan EdTech team highlights that this no-code customization via instructions and a knowledge base lets users tailor an AI to needs that standard models would miss. For trainees, it’s an eye-opening introduction to concepts like context windows and embedding knowledge into AI assistants.
Integrated Tools and Skills: When creating a GPT, builders can enable capabilities like web browsing, code execution, or image generation (DALL·E) with simple checkboxes. This means a custom GPT can go beyond just conversation – it can fetch real-time information, produce charts or artwork, or run Python code if those features are toggled on. Workshop participants can experiment with these “skills” to see how adding a tool plugin changes the AI’s utility. For example, a finance-focused GPT with web access could pull the latest stock prices during the session. OpenAI essentially commoditized agent plugins by making them one-click options, so users in a workshop can explore multimodal AI behaviors without needing to program API calls.
User-Friendly Creation Interface: Perhaps the most underrated innovation is the GPT Builder interface itself. It provides two modes – Configure and Create – to suit different user preferences. In Create mode, you literally chat with an assistant who helps you build the bot. You can say, “I want a GPT that helps me plan hiking trips,” and the builder will ask follow-up questions (desired name, tone, logo style, etc.) and generate initial settings for you. In Configure mode, you have direct form fields to set the name, description, system prompt, and to upload files or enable tools. This dual interface (conversational vs. form-based) means both non-technical users and power users are comfortable. Notably, the builder provides a live Preview pane where you can test your GPT on the right, while tweaking instructions on the left. This immediate feedback loop – you adjust something and instantly see how the bot’s answers change – is extremely powerful for learning. It embodies Bret Victor’s principle that “creators need an immediate connection to what they create”. In a workshop, this rapid iteration keeps participants engaged, as they don’t have to wait or run separate sessions to validate their changes – the impact of a prompt tweak is evident in seconds.
Auto-Generated Icons and Identity: A small but delightful feature is the ease of giving your GPT a custom identity. The GPT Builder can generate a profile picture (icon) for your bot using DALL·E 3 if you provide a concept. For example, if someone says “use a robot teacher holding a book as the icon,” the builder will create a few image options. Participants love this because it personalizes their AI creation – it’s their bot with a name and face, not just a generic ChatGPT. Psychologically, this makes the AI feel like a product they built rather than just an output of a black-box model. It’s a subtle reinforcement of the learning goal: they are designing an AI tool, not merely chatting.
Easy Sharing and Collaboration: Once a GPT is built, it can be shared via a link or published to the GPT Store for others to discover (if the creator chooses to make it public). Initially, OpenAI allowed creators to generate private shareable links so that even non-plus users (with the link) could use a GPT. Subsequently, the GPT Store provided a central hub for browsing community GPTs. For workshop settings, this distribution model is a boon. Trainers can prepare GPT templates ahead of time and give attendees the links, or, as an exercise, attendees can share the GPTs they built with each other to test and compare. The fact that a custom GPT runs in any ChatGPT interface (no installation needed) means deployment friction is near zero – a critical factor when dealing with groups of varying technical skill. Moreover, for companies concerned about privacy, OpenAI’s ChatGPT Team and Enterprise plans support publishing GPTs in a private, internal store visible only to the organization’s members. This way, workshops within a company can safely share GPTs containing proprietary knowledge or prompts without exposing them publicly.
Scalability and Iteration: A perhaps less obvious feature is how GPTs scale the learning beyond the workshop itself. Because these GPT “apps” live on as persistent agents, participants often continue to refine and use them after the session. They might build a simple tool during the workshop and later expand it as they find new needs. This encourages ongoing practice with GenAI. OpenAI anticipated this iterative innovation; in fact, they expect “the most incredible GPTs will come from the community” over time. By giving people a taste of creating GPTs, workshops plant seeds for longer-term engagement and creativity in the GenAI ecosystem.

A Wardley Map Perspective on GPTs¶

Figure: A Wardley Map illustrating OpenAI’s strategy in launching custom GPTs. In step (1) a successful product (ChatGPT) is commoditized by exposing it via APIs and a no-code builder, turning it into a platform. This enables a flurry of innovation on top of the platform as users build custom GPT applications (step 2). OpenAI can observe which use-cases gain traction, leveraging that insight (step 3) to improve their offerings or incorporate popular GPTs. Finally, the most successful new capabilities can be folded back into the core platform or offered as official features (step 4), completing the Innovate–Leverage–Commoditize cycle. In short, OpenAI’s GPT Store move was a textbook play to “build with the community” and accelerate AI adoption. As one commentator noted, if OpenAI hadn’t done it, someone else would have – but by doing it within ChatGPT, OpenAI ensured that innovation happened on their terms and infrastructure. For workshop organizers and enterprise strategists, this context is important. It means that custom GPTs are not a fad but a strategic shift towards commoditized AI capabilities. The barrier to creating AI-powered tools has been lowered to the point that domain experts (not just software developers) can innovate. When planning GenAI workshops, we’re tapping into this new dynamic: harnessing a commodity capability (LLM-as-a-service) to rapidly prototype solutions. Wardley Mapping helps convey to executives that GPT-building isn’t just a toy exercise – it’s training them in a new kind of agility where AI capabilities are building blocks readily available to solve business problems.

Why GPT-Powered Workshops Are So Effective¶

Using GPTs in workshops flips the traditional training approach from passive learning to active creation. Here are several reasons this approach works particularly well for teaching GenAI concepts:

Immediate Practical Experience: Instead of merely hearing about prompts or reading case studies, participants get to build and interact with their own AI agent in real-time. This tangible outcome – “I made a chatbot that does X” – is far more impactful than slides or demos. It turns abstract concepts into something personal and concrete. Importantly, the feedback cycle is instantaneous: as they adjust the GPT’s instructions or knowledge, they see the bot’s responses change right away. This aligns with educational best practices of learning-by-doing and iterative improvement. The immediate connection between the creator and the creation keeps engagement high. Attendees often describe the experience as eye-opening, because within an hour they accomplish something that previously seemed like science fiction – creating a bespoke AI assistant.
Demystifying AI and Prompt Engineering: Many executives have heard of prompt engineering as a skill, but they haven’t had a systematic way to practice it. Crafting a custom GPT forces you to think about how to instruct an AI clearly and constrain its behavior. For instance, if a participant’s GPT returns answers that are too verbose or off-target, they learn to refine the system prompt or add a guiding example. They see cause and effect: when I add this line to the instructions, the output improves in the following way. This trial-and-error in a safe sandbox builds intuition about how large language models work. In effect, the workshop becomes a crash course in prompt design, but with the fun of creating a useful tool as the end reward.
Empowering Non-Technical Users: A key advantage of GPT workshops is that you don’t need to be a programmer. Business users, domain specialists, and decision-makers – people who might never write Python or fine-tune a model – can still create functional AI applications through natural language. This is enormously empowering. It shifts the mindset from “AI is a magical thing done by tech experts” to “AI is a toolkit I can leverage for my own domain problems.” We’ve observed in workshops that once an executive successfully builds a simple GPT (for example, a GPT that answers FAQs about their product using the product datasheet as the knowledge base), a lightbulb goes off. They start thinking of multiple use cases where a custom AI assistant could streamline their work. The hands-on creation overcomes the skepticism or intimidation that often surrounds AI.
Collaboration and Sharing: Workshops can be structured to include group collaboration using GPTs. Because GPTs can be easily shared, teams of participants might each build a component GPT and then use them collectively. For example, in a cybersecurity training workshop, one group could build a “Threat Briefing GPT” (fed with threat intel reports), another builds a “Compliance Advisor GPT” (fed with policy documents), and so on. In a final exercise, they chain these assistants together or simply have each team demo their GPT to others. This not only reinforces learning (teaching others what you built is a great way to solidify knowledge), but also showcases the diversity of applications. Participants often get inspired by each other’s creations – “I hadn’t thought you could do that with a GPT!”. The workshop thus cultivates a community of practice, with the GPT Store serving as a gallery of examples even after the event.
Safe-to-Fail Environment: During these sessions, it’s important to emphasize that not every attempt will work perfectly – and that’s okay. GPTs sometimes produce irrelevant or incorrect answers, especially if the instructions are unclear or the knowledge base has gaps. In a workshop, this becomes a learning moment rather than a failure. Participants can debug the issue by tweaking the prompt or adding missing info. The facilitator can guide a discussion: Why did the GPT respond that way? What could we change to fix it? Because the cost of an error is basically zero (you just try again), learners are encouraged to experiment. This playful, iterative exploration is one of the best ways to discover both the capabilities and limits of GenAI. It trains a mindset of continuous improvement when working with AI, which is valuable back on the job.
Instant Deployment and Realism: Traditional software workshops often suffer from setup complexity – installing environments, deploying to a server, etc. With GPTs, deployment is instant via the cloud. As soon as a GPT is built, it’s live at a URL. This means workshop scenarios can be very realistic. For instance, if training customer support staff, you can literally have them build a support assistant GPT and then role-play a customer chat with it during the workshop. The realism of interacting with an AI agent that knows your business context brings GenAI from theory into practice. Attendees leave not just with ideas, but with actual AI tools they can continue to use and refine. One workshop participant noted that using a GPT built on their company’s data felt like “having a colleague that instantly onboards with all our docs” – a glimpse of how AI might augment their team’s work.

Running a GPT-Building Workshop: A Practical Plan¶

How can you structure a workshop around building custom GPTs? Below is a typical blueprint, based on successful sessions with both technical and non-technical groups:

Prerequisites and Setup: Ensure participants have access to ChatGPT Plus or an Enterprise/Team account beforehand, as only paid accounts can create GPTs. It’s worth arranging trial accounts or a temporary team setup if needed for a large workshop. Kick off by explaining the goal: each person (or team) will create a custom AI assistant by the end of the session. No software installation is required beyond a web browser login to ChatGPT.
Introductory Demo: Start with a live demo. For example, build a simple GPT in front of the class: “Let’s make a GPT that is a Recipe Helper.” Walk through the steps: click Explore GPTs, then Create, then type a one-sentence instruction like “You are a chef bot that suggests recipes based on available ingredients.” Show how the GPT Builder suggests a name (“ChefMate”), an icon (perhaps a chef’s hat graphic via DALL·E), and a sample prompt. Then switch to the Preview and ask it a question (like “I have chicken and broccoli, what can I cook?”). The purpose of this demo is to remove the mystery and motivate the group – if they see a GPT built in 5 minutes, they’ll be eager to try it themselves.
Explain Key Concepts Briefly: Provide a short primer on the pieces that make a custom GPT: Instructions (system prompt), Knowledge base (uploaded files), and Capabilities (tools like web browsing). Keep this section concise – the aim is not to lecture, but to give enough understanding so participants can make informed choices when building. Emphasize that they can always modify these settings later; the process is iterative, not one-shot. If relevant, mention any corporate guidelines (e.g., don’t upload confidential data, or preferred use-cases to focus on).
Hands-on Building (Guided): Let them dive in. It helps to provide a few structured exercises or prompts for those who aren’t sure what to build. For instance, you might suggest: “If you’re in marketing, try making a GPT that drafts social media posts given bullet points.” Or “If you’re a developer, maybe create a GPT that explains segments of code from our codebase (you can upload a sample file).” Participants can also choose their own idea – creative freedom often leads to high engagement. As they build, roam the room (or breakout rooms, if virtual) to assist. Encourage them to use the chat-based Create tab if they prefer a conversational setup, or the Configure tab for direct control. Many will do a mix: chat to get a baseline, then fine-tune in the form fields. Remind them to test in the Preview panel frequently, trying both good and edge-case queries for their GPT. This phase is where peer learning happens too; allow people to share cool tricks they discover (e.g., one might figure out that adding a step-by-step example in the instructions yields better outputs – they can share that tip with the group).
Iteration and Tuning: After an initial building period (say 30 minutes), gather some quick reports. Ask a few volunteers to describe their GPT and any challenges they faced. Common early issues might be: “My GPT still refuses to answer some questions” (could be default safeguards kicking in) or “It gives too generic answers”. Facilitate a discussion on how to improve them: perhaps adjust the phrasing of instructions, add a more specific knowledge file, or enable a capability. Participants can then go back for a second round of refinement. This models the real-world process of AI development – you rarely get everything perfect on first try. It teaches patience and systematic troubleshooting with AI outputs.
Sharing and Testing: Once the GPTs are functional, have participants share them. If privacy is a concern or not everyone has Plus, pair people up or do a round-robin where they try each other’s GPTs on the facilitator’s account projected on screen. This is usually the most fun part: people love to “stump test” each other’s creations and see how the different GPTs behave. It often sparks ideas: “Oh, you uploaded our product catalog and now the bot can answer pricing questions – I want to add that to mine too!” If using ChatGPT Team, you might have an internal GPT workspace where everyone in the workshop can publish their GPT for others to see. Alternatively, simply clicking Share and distributing the link in a chat channel works if all are Plus users. The act of presenting their GPT to peers also reinforces participants’ understanding and pride in what they built.
Use-Case Discussion: Bring the group back together for a debrief. Discuss potential real-life applications of the GPTs they built or ones they can imagine now. This connects the workshop exercise to business value. For example, someone might say, “Our HR team could really use a GPT to answer common employee questions from the handbook”, or “I realized I can build a GPT to summarize client reports before meetings.” List these ideas – it shows a tangible return on the time invested in training. Also discuss limitations: where would you not trust the GPT without human oversight? What about data privacy considerations when using internal knowledge? These questions ensure that while enthusiasm is high, it is tempered with a realistic understanding of responsible AI deployment.
Next Steps: Conclude by empowering the participants to continue exploring. Encourage them to keep refining their GPTs or create new ones, and to share exceptionally useful GPTs with the wider organization (perhaps through an internal newsletter or a team GPT store). If your company has a center of excellence for AI, invite the attendees to join and keep the knowledge exchange going. It’s also worth mentioning that OpenAI has been piloting a GPT builder revenue program (starting in the US) where creators may eventually earn based on usage of their public GPTs. While monetization for GPT builders is still emerging and initially limited, it underscores that building good GPTs is a valued skill – potentially even an entrepreneurial opportunity in the future. This can further motivate technically inclined participants to polish their GPTs beyond the workshop.

By the end of such a workshop, each participant has not only learned about GenAI – they have a working AI artifact to show for it. The session demystifies AI through direct engagement, leaving people more confident and curious to apply GenAI in their work. One can literally observe the shift in mindset: instead of asking “Could AI ever do XYZ?” they start asking “How might I get a GPT to do XYZ?”, which is exactly the innovation culture we want to foster.

Real-World Example: Executive Training with GPTs¶

To illustrate the impact, consider a real-world scenario drawn from our experience. We ran a workshop for senior executives at a financial services firm focused on using GenAI for decision support. Initially, some attendees were skeptical – AI felt abstract and risky to them. We guided them to build a custom GPT we nicknamed “Market Analyst GPT.” Each executive uploaded one of their company’s quarterly market outlook PDFs into the GPT’s knowledge base, and we gave them a starting system prompt about being a helpful financial analyst. Within 45 minutes, these executives – most of whom had never used ChatGPT beyond trying ChatGPT’s default mode – had a chatbot that could answer questions about their reports, compare current data with the previous quarter, and even draft a summary in a friendly tone that could be shared with non-technical stakeholders. The aha moment was palpable when one executive asked his GPT, “What were the main economic risks highlighted this quarter compared to last quarter?” and it responded with a cogent, referenced answer drawn from the two PDF reports. He exclaimed, “This would have taken my team a day to cross-check – and I got it in seconds!” That realization – seeing a GenAI agent rapidly synthesize their own institutional knowledge – made the power of GenAI concrete. By the end, instead of worrying that AI might be too “black box” or uncontrolled, the executives were proposing where to apply it, from compliance checklists to client Q\&A assistants. The key was that they built it themselves, fostering both understanding and buy-in.

Another example comes from a technical staff workshop at a cybersecurity company. Participants created GPTs like an “Incident Response Advisor” and a “Secure Code Reviewer”. In doing so, they learned how to give the model boundaries (one team cleverly set their GPT’s instructions to refuse answering outside its domain, to avoid overconfidence) and how to feed it sanitized incident data for analysis. This double effect of learning AI features while simultaneously creating domain value is what makes GPT workshops so high-yield. It turns training time into prototype development time. In fact, a few GPTs born in workshops have gone on to become internal tools used daily by those teams – essentially workshop output transitioning directly into operational utility.

Best Practices and Considerations¶

While GPT-powered workshops are powerful, there are some best practices and considerations to keep in mind:

Ensure Ethical and Policy Compliance: Remind participants about data privacy when using knowledge files. They should not upload sensitive personal data or anything that violates company policy. If the workshop involves company data, consider using ChatGPT Enterprise, which ensures data won’t be used to train OpenAI’s models and offers admin controls. It’s also wise to cover the basics of AI ethics – for instance, that the GPT may sometimes make up answers (AI hallucination) and thus critical decisions shouldn’t be made solely on its output without verification. Setting these expectations upfront maintains trust.
Define Scope for Non-Technical Audiences: For less tech-savvy groups, it’s helpful to frame what GPTs can and cannot do. Use analogies if needed: “Think of your GPT as a very smart intern: it’s read a lot (general training), you’re giving it your company manual (knowledge base) and some instructions on how to behave. It will try its best to help, but it doesn’t actually understand truth – so you have to double-check its work.” This kind of framing manages excitement to realistic levels and positions the GPT as an assistant, not a magic oracle.
Time Management: Building a GPT is fun – sometimes participants get so engrossed in fine-tuning that they might lose track of time. As a facilitator, allocate time blocks for each activity and gently nudge the group to move on to sharing or discussion phases so that everyone benefits from collective learning. It can help to set a timer or use a slide like “10 minutes left to refine your GPT!” to keep momentum.
Leverage Teams and Competition: If appropriate, introduce a friendly competition – e.g., a prize for the “most creative GPT” or “most helpful GPT” as judged by the group. This can spur teams to think outside the box. We’ve seen quirky but instructive entries this way, like a GPT that speaks in Shakespearean English while explaining cloud architecture (which actually taught everyone prompts can affect style drastically). The key is to ensure competition doesn’t sacrifice collaboration; it should serve as motivation, but the atmosphere should remain one of shared exploration.
Document and Debrief: Have a way to capture what was built (even if just a list of GPT names and what they do). This serves as a reference after the workshop. If possible, gather feedback: What did participants enjoy most? What do they plan to do next with what they learned? This not only validates the workshop’s impact but can also surface success stories or interest in follow-up sessions. Some organizations have even set up an internal forum or channel for “GPT Builders” to continue exchanging ideas post-workshop, effectively seeding an internal GenAI champions community.
Stay Updated: The landscape of custom GPTs is evolving. OpenAI continues to refine the GPT builder (for instance, model upgrades or new tool integrations might roll out). Keep an eye on release notes and community forums. Also, watch for the official monetization program developments and any emerging marketplaces. As of early 2025, the GPT builder revenue pilot has launched in limited form (paying U.S. creators based on engagement), but a full-fledged marketplace with pricing models may follow. These changes could influence how GPTs are used or perceived (for example, a marketplace might spur higher-quality GPT development, which you can showcase as examples in future workshops). Being informed enables you to provide the latest and greatest to workshop attendees and to anticipate how GenAI training might incorporate new features (such as fine-tuning or more advanced agent behaviors if they become available in the no-code interface).

Conclusion¶

Custom GPTs represent a significant step in the democratization of AI – they allow virtually anyone to mold a powerful language model into a bespoke assistant or mini-application. For organizations looking to upskill their workforce in AI literacy, or for consultants and educators teaching about GenAI, GPT-building workshops have proven to be a game-changer. They transform AI from an abstract concept into a hands-on craft. By creating and deploying GPTs within a single session, participants gain not only knowledge but also confidence in working with AI systems. They witness first-hand why GPTs are “so powerful to give workshops”: the technology is accessible, immediate, and rewarding to use, and it bridges the gap between learning and real-world application.

From a broader perspective, the rise of custom GPTs fits into the ongoing evolution of AI as a ubiquitous utility. Just as spreadsheets made computing power accessible to every finance worker, or websites made publishing information instant for anyone, GPTs are making conversational AI a tool that individuals can wield for their own purposes. Today, an executive can craft an AI assistant tailored to their workflow as easily as they might write a document. Tomorrow, this capability will likely be even more integrated – we might see marketplaces of corporate GPTs, or AI assistants becoming standard “team members” in organizations. By engaging with GPTs now through workshops, teams position themselves ahead of that curve, cultivating a mindset of solution-building and continuous learning with AI.

In summary, using GPTs for workshops creates a virtuous cycle: it accelerates the adoption of GenAI practices in the short term (through practical skill-building and prototypes), and it lays the cultural groundwork for innovation in the long term (by showing people that they can actively shape how AI is used in their domain). The experience tends to linger in participants’ minds – after all, it’s not every day you get to create a working AI agent from scratch. As one workshop attendee succinctly put it, “I came in curious about AI. I left with an AI I created.” That is the kind of transformational outcome that makes GPT-driven workshops a highly effective strategy for any organization navigating the new AI era.

Sources: