From conversational AI to working AI, ChatGPT itself is being redesigned
Series Overview
Part 1 | Why Does AI Get Things Wrong So Convincingly?
Part 2 | Why the OXMV Loop and Evidence Ledger Are Necessary
Part 3 | Mapping Knowledge for AI with Ontology
Part 4 | OpenAI to End ChatGPT’s Custom GPT Era on December 11, 2026
Part 5 | Taking Aim at ChatGPT’s Chronic “Hallucination” Problem
Part 6 | Can AI Detect Financial Cyberattacks Before They Escalate?
Part 7 | GPT-6.1 Sol: From Conversational AI to a Work Platform
By Manripungchwi Jung Yeon-tae
ChatGPT is changing again. This time, however, the change is not simply about moving from one model number to another. Nor is it merely a redesigned interface or a few additional menu items. What is happening now is closer to a structural reorganization of the entire AI ecosystem.
Functions that were once mixed together inside a single ChatGPT experience are being separated into more specialized layers, with each layer becoming more powerful. The most important question is therefore no longer simply, “How much smarter is the new model?” A better question is, “What role will GPT, Plugins, dot, Work, Page, and Space each play in the new ChatGPT environment?” Once that structure becomes clear, the recent changes start to make much more sense.
1. GPT-6.1 is not just “a smarter model”
The significance of GPT-6.1 lies not only in improved reasoning or technical performance. The larger change is how ChatGPT itself is evolving. ChatGPT is moving away from being a system that simply produces text answers and toward one that can choose the most appropriate interface, tool, workflow, or visual representation for the task. This is where the idea of an Intelligent UI becomes important.
Depending on the question, ChatGPT may present text, comparison views, diagrams, calculators, interactive tools, visual summaries, or structured interfaces. The direction is clear. ChatGPT is evolving from “an AI that answers questions” into “an AI that creates the tools and interfaces needed to complete the work.”
2. Why are Custom GPTs moving toward Plugins?
Custom GPTs were among the most useful features of earlier ChatGPT generations. Users could create specialized AI systems by adding custom instructions, knowledge files, domain expertise, workflows, and external tools. This was powerful, but it also meant that several different functions were bundled together inside one GPT. Professional knowledge, instructions, tools, APIs, interfaces, and actions were all mixed into a single unit.
The new Plugin architecture separates those elements more clearly. When a Custom GPT is migrated, its main components can be reorganized into a Plugin structure: GPT instructions become a Plugin Skill, GPT knowledge files become Reference Files, and connected external services become Plugin Apps. This is not simply the disappearance of GPTs. It is better understood as the repackaging of specialized GPT capabilities into a larger and more modular Plugin architecture.
3. The new Plugin is very different from the old Plugin
In the earlier ChatGPT era, a Plugin was mostly a bridge to an external service. The structure was relatively simple: ChatGPT → External Service. A Plugin might call a travel service, calculator, search service, or external API. The new Plugin concept is much broader.
A modern Plugin can combine Skill, Reference Files, Apps, MCP Tools, and Extensions. A Skill contains professional instructions and workflows, while Reference Files contain detailed domain knowledge. Apps connect external services such as Gmail, Google Drive, or Calendar, and MCP Tools can connect private databases, APIs, company systems, and specialized engines. Extensions can provide custom interfaces such as forms, dashboards, file handlers, and interactive views.
This means that a Plugin is no longer merely an external connector. It can become a complete specialized AI application package.
4. What happens when a professional GPT becomes a Plugin?
In a professional GPT, the core value usually lies in its domain knowledge and reasoning rules. Consider a specialized employment and talent-matching GPT. It might contain industry and occupation ontologies, JOB, SHIFT, and TASK models, matching logic, trust rules, runtime rules, OXMV verification, and Evidence Ledger structures. When converted into a Plugin, those elements can be reorganized more systematically.
The Skill can contain integrated operating instructions, while Reference Files hold specialized knowledge. Apps can connect services such as Google Drive, Gmail, and Calendar. Custom MCP Tools can connect employment APIs, employer databases, talent databases, and mapping services. The result is more than a knowledge-based chatbot: it becomes a system combining professional intelligence, live data, and real-world action.
5. What is the difference between a migrated GPT Plugin and an external App Plugin?
This distinction is important. Some Plugins are built mainly around external applications such as Gmail, Calendar, Drive, Slack, and customer relationship management systems. Their primary value lies in accessing external data and carrying out actions. They can read, write, search, send, schedule, or update information.
By contrast, a professional GPT-derived Plugin is centered on domain knowledge and decision logic. Examples might include financial crime analysis, smart city planning, AI and semiconductor strategy, regional policy analysis, space science, and employment matching. The easiest way to understand the difference is that an external App Plugin provides the hands and feet, while a professional GPT-derived Plugin provides the specialized brain. The strongest architecture combines both.
6. How do you build a specialized Plugin?
The development process is more structured than before. The first step is to organize the domain knowledge by defining what the system must know, which ontology it uses, which rules it follows, what must be verified, and what types of decisions it can make. The second step is to create the Skill, which defines how user intent is interpreted, which knowledge files are used, how reasoning proceeds, which workflow is followed, and how output is controlled.
The third step is to add Reference Files. Detailed professional knowledge should be separated from the Skill, and these files may contain ontologies, rules, legal references, technical definitions, workflow specifications, and domain manuals. The fourth step is to add external Apps such as Google Drive, Gmail, Calendar, Slack, or enterprise tools.
The fifth step is to add MCP servers or custom tools when no existing App provides the required function. These may connect proprietary databases, government APIs, company systems, search services, or specialized engines. The sixth step is to add UI elements and Extensions, such as dashboards, candidate lists, forms, maps, file editors, and interactive result views. The development path can therefore be summarized as Knowledge → Skill → Reference Files → Apps → MCP → UI.
7. Has dot inherited the “agent role” of Custom GPTs?
Conceptually, yes—at least in part. It would be inaccurate to say that OpenAI literally split Custom GPTs into two pieces, with one half becoming Plugins and the other becoming dot. Functionally, however, that is close to what the new architecture looks like. A traditional GPT mainly behaved as “a specialist AI that responds when asked,” while dot is closer to “an AI agent that keeps working toward a goal.”
A dot can maintain context over time, use connected tools, monitor changes, and repeatedly act on the user’s behalf. For example, imagine an AI international standards Plugin containing expert knowledge. A dot could then be assigned the following task: “Check new international AI standards every week and tell me when they conflict with the standard I am developing.” In that structure, the Plugin is the specialist brain, while dot serves as an ongoing AI staff member.
8. What is Work, and how is it different from dot?
Work is also agentic, but its role is different. A useful distinction is that Work is an execution agent for completing a complex task, while dot is a persistent agent for managing an ongoing responsibility. For example, “Research the Korean construction industry over the last five years and prepare a full report” would be a suitable task for Work. By contrast, “Monitor construction policy changes every week and notify me when something important changes” would be more naturally suited to dot.
9. What is Page?
Page is one of the most important new concepts. The old Canvas experience was essentially an editable document or coding surface beside the conversation. Page extends that idea much further. A Page is a living work artifact that can continue to evolve with the user and AI.
Instead of producing a draft, a revised draft, another revision, and a final version as separate chat messages, users can continuously improve a single Page. They can ask the AI to shorten section three, add an OXMV section, leave a particular paragraph unchanged while revising the conclusion, or turn part of the document into a diagram. Page is therefore not simply another chat response. It is a persistent work product.
10. Is Space just a larger version of Canvas?
Not exactly. It is more accurate to say that Canvas evolved conceptually into Page, while Space became the larger workspace containing Pages, files, and collaborative assets. A Space can contain multiple Pages, folders, uploaded files, reference materials, team documents, and AI-generated outputs. It serves as the organizational layer above individual Pages.
For example, a single professional AI project might include a System Architecture Page, User Manual Page, OXMV Design Page, API Design Page, Release Audit Page, and Knowledge Files folder. Chat becomes the place where thinking happens. Page becomes the place where results are developed. Space becomes the place where the work is organized and preserved.
11. Why do Page and Space matter?
One long-standing weakness of conversational AI was that valuable results disappeared inside lengthy chat histories. Page and Space address that problem. They shift ChatGPT from a conversation-centric system toward a knowledge- and workspace-centric system. This is particularly important for long-term research, policy work, technical documentation, knowledge-base development, professional AI systems, and collaborative projects.
Instead of repeatedly recreating documents from chat history, users can maintain durable project artifacts. This changes ChatGPT from a temporary conversation tool into an environment where knowledge and work can accumulate over time.
12. Intelligent UI is another major change
The new generation of ChatGPT is also becoming more adaptive in how it presents results. Rather than returning only text, the system can increasingly decide whether a task would be better served by a visual comparison, chart, calculator, structured panel, diagram, or interactive control. This represents another important transition. ChatGPT is no longer only generating information; it is beginning to generate the interface required to use that information effectively.
13. Audio and meeting workflows are also expanding
Modern ChatGPT increasingly supports audio-based workflows as well. Users can bring in recorded discussions, meetings, or spoken content and turn them into transcripts, summaries, decisions, action items, and follow-up documents. Combined with meeting-oriented Plugins and Space and Page workflows, this creates a new pattern: conversation → transcription → analysis → action items → permanent project record. This is a much more operational model of AI use.
14. The new ChatGPT architecture in one view
The current structure can be understood as follows:
ChatGPT → general AI intelligence and conversation
Plugin → specialized knowledge, Skills, workflows, tools, and external Apps
dot → persistent personal AI agent
Work → execution agent for complex tasks
Page → evolving AI-assisted document
Space → workspace for Pages, files, people, and AI
Intelligent UI → adaptive interface that creates the right view or tool for the task
This division of roles is the key to understanding the new ChatGPT ecosystem.
15. What has really changed?
The biggest change is not simply a more powerful model. The deeper change is the transition from Question → Answer to Goal → Knowledge → Tools → Agent → Execution → Artifact → Ongoing Management. This represents a fundamentally different way of thinking about AI. AI no longer exists only to explain something.
AI can perform work, use tools, coordinate Apps, create artifacts, monitor ongoing tasks, maintain project context, and support repeated decision-making. Its role is expanding from answering individual questions to participating in a continuing workflow.
ChatGPT is becoming a working environment, not just a chatbot
Recent changes may feel confusing because the ChatGPT we once knew is being reorganized. Professional GPT capabilities are moving toward Plugins, while domain knowledge is being separated into Skills and Reference Files. External services are being connected through Apps and MCP. Persistent responsibilities are increasingly suited to dot, while complex one-time tasks are increasingly suited to Work.
Long-lived results are being maintained in Page, while multiple Pages and files can be organized in Space. The result is that ChatGPT is moving toward something much larger than a chatbot. It is becoming an AI brain, specialized knowledge, agents, applications, and workspaces integrated into one environment.
The question for users is therefore changing. It is no longer only, “How should I ask ChatGPT a better question?” The new question is, “Which work should be given to a Plugin, which responsibility should be assigned to dot, which task should be delegated to Work, which result should live in Page, and which project should be organized in Space?”
We are entering an era in which the real skill is no longer merely using AI. It is designing how humans and AI work together.
About the Author
Manripungchwi Jung Yeon-tae is a professional-knowledge AI planner, designer, and GPT development specialist. He currently serves as President of the Construction Industry Information Institute and develops professional GPTs across construction, law, taxation, finance, management, medicine, science, history, and philosophy. His research focuses on verified professional-knowledge AI combining ontology, XAI, the OXMV Loop, and the Evidence Ledger.
