ChatGPT Work: What If AI Were Not Just a Tool We Use, but Work We Could Delegate?
From prompting for answers to defining outcomes, supervising execution, and accomplishing meaningful work.
Lateral Thinking × PO × Generative AI × Human Judgment
An independent AIYING exploration | November 2026
1. Reality: We Have Learned to Prompt AI
For several years, the dominant pattern of working with generative AI has been conversational.
We ask for an explanation. AI explains.
We ask for a draft. AI drafts.
We ask for an analysis. AI proposes an analysis.
Then we carry the output into another tool, perform the next step, and return with another request.
This can be highly productive. But the human remains responsible for coordinating much of the workflow.
Consider a familiar assignment: prepare a comparative report about three AI engineering practices.
A conventional workflow might involve finding sources, extracting information, organizing notes, building a comparison table, drafting conclusions, and formatting the final document.
ChatGPT can help at every stage.
But what if the interaction could begin with the desired deliverable rather than a sequence of separate prompts?
2. Understand: What Is ChatGPT Work?
ChatGPT Work is an OpenAI experience designed for longer, multi-step tasks.
Rather than focusing only on a single conversational response, Work can coordinate supported tools and workflows to help produce completed deliverables.
Depending on access and permissions, its capabilities include:
- Researching and analyzing information.
- Working with uploaded files and connected services.
- Creating documents, spreadsheets, presentations, and reports.
- Interacting with supported websites through a cloud browser.
- Allowing users to inspect progress and redirect work.
- Requesting user input or approval when necessary.
- Supporting scheduled or recurring work where available.
The cloud browser operates separately from the user's normal browser session. Website compatibility and permitted actions vary.
Work is not a guarantee that every task can be completed autonomously. Some steps require user involvement, and some websites or operations may not be supported.
Official starting point: https://chatgpt.com/
3. Connect: From Prompt Engineering to Work Engineering
Prompt engineering asks how we should communicate a request to an AI model.
AI engineering asks how we design systems that use models, tools, evidence, and controls to accomplish tasks reliably.
ChatGPT Work brings a related question into everyday professional practice:
How should a person define, delegate, supervise, and verify a complete unit of work?
Consider the difference.
| Prompt-centered approach | Work-centered approach |
|---|---|
| Write a report outline. | Prepare a report with defined sections, evidence, and review criteria. |
| Find information about competitors. | Research selected competitors and produce a source-backed comparison. |
| Make a spreadsheet table. | Create a spreadsheet with specified columns, formulas, and validation checks. |
| Suggest a project plan. | Prepare a project plan, identify dependencies, and flag unresolved decisions. |
| Summarize this document. | Analyze the document and produce a decision memo identifying evidence and gaps. |
The important difference is not simply that AI performs more steps.
It is that the user begins to design the conditions for successful accomplishment.
4. The Generative View: Change the Unit of Interaction
The dominant assumption is:
Now introduce a deliberate Lateral Thinking provocation.
This is not a literal recommendation to transfer legal, professional, or ethical accountability to an AI system.
It is a provocation designed to move our attention from individual prompts to complete outcomes.
Movement 1 — Extract the principle
Instead of describing only what the AI should say, describe what successful completion should look like.
Define the output, the quality standard, the constraints, and the conditions requiring human review.
Movement 2 — Notice the difference
A prompt is often an instruction for the next response.
A work assignment is a specification for a result.
It may involve several actions, decisions, checks, and revisions.
Movement 3 — Generate a possibility
What if every delegated AI task had a small Work Contract?
Such a contract would define:
- Outcome: What must be accomplished?
- Inputs: What information may be used?
- Authority: What actions are permitted?
- Constraints: What must not happen?
- Evidence: How will claims be supported?
- Acceptance: How will success be judged?
- Escalation: When must the AI stop and ask?
A practical framework for defining a bounded, verifiable unit of AI-assisted work before execution begins.
The Work Contract is an AIYING design proposal, not an official ChatGPT Work feature or OpenAI terminology.
5. Explore: Five New Possibilities
Possibility 1 — Research becomes a deliverable
Instead of collecting links and asking for repeated summaries, define a research brief with source requirements, comparison criteria, and a finished report.
Possibility 2 — Learning becomes project completion
Instead of asking for lessons on spreadsheets, define a small spreadsheet project with an explanation of formulas and a verification checklist.
Possibility 3 — Meetings become traceable decisions
With authorized meeting notes, ask for a decision register distinguishing agreed actions, unresolved questions, owners, and evidence.
Possibility 4 — Engineering documentation becomes reviewable work
Ask for a technical design comparison that identifies trade-offs, assumptions, failure modes, and unanswered engineering questions.
Possibility 5 — Repeated work becomes a monitored workflow
Where scheduling is supported, recurring information-gathering tasks can be structured to report changes and flag issues requiring attention.
These possibilities depend on suitable access, available tools, and human review. They are examples of potential workflows, not claims of completed experiments.
6. Review: What Should Humans Continue to Own?
Delegating execution is different from delegating judgment.
Even when AI can carry out a workflow, important responsibilities remain with people.
| AI may assist with | Human responsibility |
|---|---|
| Gathering and organizing information | Assessing source quality and relevance |
| Drafting a recommendation | Choosing whether to adopt it |
| Preparing a spreadsheet | Checking formulas and business assumptions |
| Proposing external actions | Approving consequential actions |
| Following a workflow | Defining permissions and stopping conditions |
| Reporting task completion | Verifying the actual deliverable |
AI may misunderstand instructions, rely on incomplete information, encounter misleading website content, or produce artifacts that appear complete but contain errors.
Successful delegation therefore requires clear boundaries and independent checks.
7. The AIYING GENERATE Method Applied to Work
Define the actual task, inputs, constraints, and desired outcome.
Identify assumptions about who must perform each step.
Use PO to challenge the existing division of work.
Generate alternative workflows and delegation boundaries.
Evaluate risks, permissions, evidence, and quality criteria.
Delegate a small, bounded task and inspect the result.
Record outcomes and improve the Work Contract.
8. Hands-On Lab: Delegate Your First Bounded Work Assignment
Objective
Use ChatGPT Work to create a short, source-backed comparison report on three AI engineering practices.
Tool: ChatGPT — https://chatgpt.com/
Feature: Select Work when available in your account.
Time: Allow approximately 20–30 minutes for setup, execution, and review. Actual execution time varies.
Prerequisites: An eligible ChatGPT account with Work access. No external account connections are required for this proposed research task.
Step 1 — Open Work
Open ChatGPT on the web or in the app and select Work. If Work is unavailable, you can still use the Work Contract as a planning exercise in regular ChatGPT, but that is not equivalent to testing Work's multi-step execution capabilities.
Step 2 — Define the outcome
The deliverable is a concise comparison report covering:
- Retrieval-Augmented Generation (RAG).
- AI evaluation.
- AI observability.
For each topic, require a definition, engineering purpose, example, limitation, and supporting source.
Step 3 — Define the boundaries
Use public sources only. Do not connect private accounts. Do not publish or send anything externally.
Step 4 — Give Work the assignment
Copy and paste the prompt below.
WORK ASSIGNMENT: AI Engineering Practices Comparison OUTCOME: Create a concise, professional report comparing RAG, AI evaluation, and AI observability. DELIVERABLE: A report containing: 1. Executive summary. 2. Definition of each practice. 3. Its role in AI engineering. 4. One practical use case per practice. 5. Key limitations and risks. 6. A comparison table. 7. Five questions an AI engineering team should ask before adoption. 8. Source links and publication dates when available. SOURCES: Prefer official technical documentation and credible primary sources. Clearly distinguish verified facts from interpretations or suggestions. BOUNDARIES: Use public information only. Do not access private accounts. Do not publish, email, purchase, or modify external resources. QUALITY CRITERIA: Make the report understandable to an aspiring AI engineer. Avoid unsupported product claims. Identify any missing evidence. HUMAN REVIEW: Flag uncertain findings. Ask before taking any external action that changes data or communicates with another party. FINAL OUTPUT: Provide the finished report and a short verification checklist. Do not claim to have checked a source unless you actually did.
Step 5 — Observe the workflow
Inspect the steps Work takes. If it requests clarification or approval, review the request rather than automatically accepting it.
Step 6 — Verify the deliverable
Open at least three cited sources independently.
Check whether the report accurately represents their content. Verify the comparison table and examine whether limitations have been disclosed.
Step 7 — Record the accomplishment
Save the report and your review notes. Record what was completed, what required correction, and what you would change in the Work Contract next time.
Expected output
- One comparative AI engineering report.
- One table comparing the three practices.
- Source links for verification.
- One completed review checklist.
- One revised Work Contract based on observations.
These are intended outputs. Their completion and quality must be confirmed by the person performing the exercise.
9. Today's Accomplishment
The practical accomplishment is not simply obtaining a report.
It is learning to define a bounded work assignment with explicit inputs, permissions, evidence requirements, and acceptance criteria—and then evaluating the result.
10. Discoveries to Record
- Which parts of the task could be delegated effectively?
- Which parts still required human interpretation or verification?
- Did the final report satisfy the original Work Contract?
- What errors or omissions appeared?
- Which additional permissions would have been unnecessary?
- What would make the next assignment easier to supervise?
11. The Bigger Insight: The Unit of AI Productivity Is Changing
The first wave of conversational AI encouraged us to become better at asking.
Agentic work systems invite us to become better at defining outcomes, delegating execution, setting boundaries, and evaluating results.
That shift creates a new professional capability:
The statement is intentionally provocative. Good prompts still matter. But the larger opportunity is to connect them to a complete, accountable process.
In the Generative View, AI does not merely accelerate existing tasks.
It challenges how we divide work between people and machines.
AI can help execute the work. Humans must remain responsible for defining what matters, deciding what is acceptable, and verifying what has been accomplished.
12. References and Further Reading
- OpenAI — ChatGPT for Your Most Ambitious Work
- OpenAI Help Center — ChatGPT Work and Codex
- OpenAI Help Center — Using Cloud Browser in ChatGPT
- OpenAI Help Center — Built-in Desktop Browser
- NIST — AI Risk Management Framework
- Edward de Bono, Serious Creativity: Using the Power of Lateral Thinking to Create New Ideas (1992).
Editorial note: This independent AIYING article combines documented ChatGPT Work capabilities with original Lateral Thinking interpretations. The Work Contract and proposed workflow are AIYING concepts, not official OpenAI features. Product availability and behavior may change.
No comments:
Post a Comment