Upgrade to Pro — share decks privately, control downloads, hide ads and more …

From Prompt to Prototype: How Conversational AI...

Avatar for Emery Fertic Emery Fertic
September 15, 2026
15

From Prompt to Prototype: How Conversational AI Made 3D Design Accessible to a Novice

Avatar for Emery Fertic

Emery Fertic

September 15, 2026

More Decks by Emery Fertic

Transcript

  1. By Emery Fertic ‘28 IT SLOAN SCHOO F MANAGEMENT From

    Prompt to Prototype: How Conversational AI Made 3D Design Accessible to a Novice
  2. How does conversational generative AI change the workflow and expertise

    required for a novice to design and fabricate a functional 3D-printed product? 5
  3. Approach: • Research Methodology • Attempted to design and fabricate

    a functional product using AI throughout the process Approached the project as a novice with limited 3D-printing and CAD experience Evidence collected: • • • • AI conversations and prompts Generated designs and STL versions Screenshots and physical prototypes Printing failures, modifications, and final test results AI tools for identifying sources and assisting with synthesis (ChatGPT, Claude, and NotebookLM)
  4. AI changed who performed the work—not the overall prototyping loop

    Create technical specification s Build CAD model Evaluate design Fabricate prototype Physically test and revise Troubleshoo t through conversation 3D-print prototype Physically test and revise 24% Develop specification s with AI AI generates the model 24% 24% 9
  5. AI changed who performed the work—not the overall prototyping loop

    Create technical specification s Build CAD model Evaluate design Fabricate prototype Physically test and revise Troubleshoo t through conversation 3D-print prototype Physically test and revise 24% Develop specification s with AI AI generates the model 24% 24% 10
  6. Specialized tools could not support the full prototyping workflow Example

    1: MeshyAI produced a concept but stopped there Prompt: - Create a foldable, 3D-printable phone stand that supports portrait and landscape orientation, requires no supports, and minimizes material. 11 Result: - Generated several visually promising concepts - Could not export a testable STL without upgrading - Could not submit a second prompt on the free tier
  7. Specialized tools could not support the full prototyping workflow Example

    2 & 3: PrintPal and Tripo3D required an input I did not have Prompt: - Create a foldable, 3D-printable phone stand that supports portrait and landscape orientation, requires no supports, and minimizes material. 12 Result: - Text-to-3D generation required a paid plan - Free access only supported image-to-3D - I had a written concept, not an existing image - The workflow stopped before I could generate, test, or revise a model
  8. Conversational AI Supported the Full Prototyping Workflow Example 3 &

    4: ChatGPT and Claude worked from the input I had Prompt: - Create a foldable, 3D-printable phone stand that supports portrait and landscape orientation, requires no supports, and minimizes material. 13 Result: - Accepted my written description - Generated a printable 3D model - Allowed continuous follow-up questions and revisions - Helped troubleshoot the design and printing process - Led to a successfully printed phone holder
  9. Conversation turned one-time generation into a continuous workflow Specialized 3D

    AI Tools ChatGPT and Claude - Produced objects that were difficult to control - Limited access to text generation and STL export - Offered little opportunity for precise revisions - Workflow stopped before printing and testing - Let me explain the product’s intended function - Generated and revised printable models - Supported follow-up and tangential questions - Helped with design, slicing, printing, and assembly
  10. The idea: A modular boat cup holder What I wanted:

    - Attach securely to a two-inch vertical pole Extend toward the person sitting behind the console Hold drinks upright without interfering with the boat Use one permanent base with interchangeable cup holders Fit a skinny can, standard can, and 32-ounce Owala Allow each holder to slide on and off easily I knew what I wanted the product to do but not how to engineer each component. 16
  11. Phase 1 Workflow: Designing the clamp and arm 1. Share

    Boat Context → 2. Specify the 2-Inch Pole → 3. Generate the Clamp and Arm → 4. Correct Orientation and Screw Access → 5. Add the Dovetail and Stopper → 6. Print and Test on the Pole 17
  12. Describe 1. Share boat context 2. Specify the 2-inch pole

    3. Generate the clamp and arm 4. Correct orientation and screw access 5. Add the dovetail and stopper ***This is an example boat 18 6. Print and test on the pole
  13. Specify 1. Share boat context 2. Specify the 2-inch pole

    3. Generate the clamp and arm 4. Correct orientation and screw access 5. Add the dovetail and stopper 19 6. Print and test on the pole
  14. Specify 1. Share boat context 2. Specify the 2-inch pole

    3. Generate the clamp and arm 4. Correct orientation and screw access 5. Add the dovetail and stopper 20 6. Print and test on the pole
  15. Specify 1. Share boat context 2. Specify the 2-inch pole

    3. Generate the clamp and arm 4. Correct orientation and screw access 5. Add the dovetail and stopper 21 6. Print and test on the pole
  16. Specify 1. Share boat context 2. Specify the 2-inch pole

    3. Generate the clamp and arm 4. Correct orientation and screw access 5. Add the dovetail and stopper 22 6. Print and test on the pole
  17. Generate “The boat pole is exactly two inches in diameter.

    Help me create a two-piece clamp and an arm that extends toward the seat.” 23 1. Share boat context 2. Specify the 2-inch pole 3. Generate the clamp and arm 4. Correct orientation and screw access 5. Add the dovetail and stopper 6. Print and test on the pole
  18. Generate 1. Share boat context 2. Specify the 2-inch pole

    “I won’t be able to access the screws.” 3. Generate the clamp and arm 4. Correct orientation and screw access 5. Add the dovetail and stopper 24 6. Print and test on the pole
  19. Troubleshoot “How is that going to fit around the pole?”

    “The screw mounts aren’t connected to the arm.” 1. Share boat context 2. Specify the 2-inch pole 3. Generate the clamp and arm 4. Correct orientation and screw access 5. Add the dovetail and stopper 25 6. Print and test on the pole
  20. Troubleshoot “The screw mounts aren’t connected to the actual arm.”

    “This will not close around the pole. Fuse the screw mounts and clamp body into one continuous solid so they print as a single part.” 26 1. Share boat context 2. Specify the 2-inch pole 3. Generate the clamp and arm 4. Correct orientation and screw access 5. Add the dovetail and stopper 6. Print and test on the pole
  21. Troubleshoot Finally, got the correct clamp! As problems continued, my

    conversational feedback became more technically precise. 27 1. Share boat context 2. Specify the 2-inch pole 3. Generate the clamp and arm 4. Correct orientation and screw access 5. Add the dovetail and stopper 6. Print and test on the pole
  22. Finalize 1. Share boat context 2. Specify the 2-inch pole

    3. Generate the clamp and arm 4. Correct orientation and screw access 5. Add the dovetail and stopper 28 6. Print and test on the pole
  23. Print & Test 1. Share boat context 2. Specify the

    2-inch pole 3. Generate the clamp and arm 4. Correct orientation and screw access 5. Add the dovetail and stopper 29 6. Print and test on the pole
  24. Print & Test *Rubber added for grip 1. Share boat

    context 2. Specify the 2-inch pole 3. Generate the clamp and arm 4. Correct orientation and screw access 5. Add the dovetail and stopper 30 6. Print and test on the pole
  25. Print & Test 1. Share boat context 2. Specify the

    2-inch pole 3. Generate the clamp and arm 4. Correct orientation and screw access 5. Add the dovetail and stopper 31 6. Print and test on the pole
  26. Print & Test 1. Share boat context 2. Specify the

    2-inch pole 3. Generate the clamp and arm 4. Correct orientation and screw access 5. Add the dovetail and stopper 32 6. Print and test on the pole
  27. Conversation translated my instructions into technical specifications Conversational input 33

    Technical specification “It needs to extend toward the person sitting there.” Arm direction and length “The pole part looks too big.” 2-inch internal diameter “I won’t be able to reach the screws.” Fastener placement and tool clearance “I want it to slide out.” Dovetail connection with a stopper
  28. Key Definitions Jagged Frontier – the uneven and shifting boundary

    between tasks AI can perform effectively and tasks it cannot, even when those tasks appear similarly difficult to a person. Dell’Acqua, McFowland, Mollick, Lifshitz-Assaf, Kellogg, et al, “Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality” working paper (rev 17 Mar 2026) Mollick,“Centaurs and Cyborgs on the Jagged Frontier” (16 Sep 2023) https://www.oneusefulthing.org/p/centaurs-and-cyborgs-on-the-jagged
  29. Phase 2 Workflow: Designing the interchangeable cup holders 1. Resolve

    threads and fastening hardware → 2. Define the three drink sizes → 3. Generate the interchangeable holders → 4. Check drink fit and clearance → 5. Adjust the dovetail for printable Tolerance → 6. Evaluate weight distribution and arm support 36
  30. Screw Assembly The Design Included Fastener Holes—but No Assembly Instructions

    1. 2. 3. 4. - AI generated the clamp The model contained smooth through-holes (not threaded) Printing revealed a missing question “What screw am I using to put all of this together?” I showed AI the hardware available AI explained that I needed a bolt through both pieces A washer and nut would secure the opposite side The bolt needed to be long enough to span the entire assembly I physically checked the recommendation Compared the available bolts with the printed parts AI could explain the fastening system once asked, but it did not proactively communicate the hardware assumptions built into its design. 1. Resolve threads and fastening hardware 2. Define the three drink sizes 3. Generate the interchangeable holders 4. Check drink fit and clearance 5. Adjust the dovetail for printable tolerance 6. Evaluate weight distribution and arm support
  31. Screw Assembly 1. Resolve threads and fastening hardware 2. Define

    the three drink sizes 3. Generate the interchangeable holders 4. Check drink fit and clearance 5. Adjust the dovetail for printable tolerance 6. Evaluate weight distribution and arm support
  32. Screw Assembly 1. Resolve threads and fastening hardware 2. Define

    the three drink sizes 3. Generate the interchangeable holders 4. Check drink fit and clearance 5. Adjust the dovetail for printable tolerance 6. Evaluate weight distribution and arm support
  33. Sizing 1. Resolve threads and fastening hardware 2. Define the

    three drink sizes 3. Generate the interchangeable holders 4. Check drink fit and clearance 2.60 in × 4.81 in tall 40 2.28 in × 6.17 in tall 3.43 in × 10.66 in tall **Pictured drinks are examples. Measurements were based on common dimensions for a standard soda can, slim can, and 32-ounce Owala 5. Adjust the dovetail for printable tolerance 6. Evaluate weight distribution and arm support
  34. Soda can Generation Skinny can 1. Resolve threads and fastening

    hardware 2. Define the three drink sizes 3. Generate the interchangeable holders 4. Check drink fit and clearance 5. Adjust the dovetail for printable tolerance 6. Evaluate weight distribution and arm support 41 Owala
  35. Fit “Are you sure a soda can will fit?” “Does

    this give enough space to pick up the drink?” 1. Resolve threads and fastening hardware 2. Define the three drink sizes 3. Generate the interchangeable holders 4. Check drink fit and clearance 5. Adjust the dovetail for printable tolerance “Would an Olipop or Coke fit?” 42 6. Evaluate weight distribution and arm support
  36. Tolerance & Print “Make sure the two dovetails are not

    exactly the same size so they can actually slide in and out.” “Is the cup holder part going to be sticking up out above the base?” “There is a slight gap where the two parts connect.” “Can it still slide out when needed?” 43 1. Resolve threads and fastening hardware 2. Define the three drink sizes 3. Generate the interchangeable holders 4. Check drink fit and clearance 5. Adjust the dovetail for printable tolerance 6. Evaluate weight distribution and arm support
  37. Tolerance & Print 1. Resolve threads and fastening hardware 2.

    Define the three drink sizes 3. Generate the interchangeable holders Soda can 4. Check drink fit and clearance Owala Skinny can 44 5. Adjust the dovetail for printable tolerance 6. Evaluate weight distribution and arm support
  38. Assembly & Testing 1. Resolve threads and fastening hardware 2.

    Define the three drink sizes 3. Generate the interchangeable holders 4. Check drink fit and clearance 5. Adjust the dovetail for printable tolerance 45 6. Evaluate weight distribution and arm support
  39. Assembly & Testing 1. Resolve threads and fastening hardware 2.

    Define the three drink sizes 3. Generate the interchangeable holders 4. Check drink fit and clearance 5. Adjust the dovetail for printable tolerance 46 6. Evaluate weight distribution and arm support
  40. Assembly & Testing 1. Resolve threads and fastening hardware 2.

    Define the three drink sizes 3. Generate the interchangeable holders 4. Check drink fit and clearance 5. Adjust the dovetail for printable tolerance 47 6. Evaluate weight distribution and arm support
  41. 3 Key Takeaways Specificity Accessibility • • • 49 Conversation

    supported design, revision, printing, and assembly within one continuous workflow. Specialized tools were limited by isolated generation, paywalls, and restricted iteration. Ongoing guidance lowered the technical barrier for a novice. • • • I began with functional goals, not complete technical specifications. Images, measurements, and failed designs revealed which constraints needed clarification. Each correction translated conversational instructions into precise design requirements. Jagged Frontier • • • AI handled complex geometry while sometimes missing basic physical and assembly constraints. Printing exposed problems involving fit, friction, threads, and fastener length. Success still required human judgment, follow-up questions, physical testing, and persistence.
  42. Future research directives 1. How can a self-leveling gimbal be

    incorporated to keep the drink upright as the boat pitches and rolls? 2. What dovetail clearance provides easy interchangeability without creating excessive gaps or movement? 3. How do different drink sizes and weights affect the stability of the arm and gimbal? 4. How does the complete design perform under realistic boat movement, vibration, and repeated use? 5. As AI improves, will it proactively identify stability, tolerance, and assembly requirements before physical testing? 51
  43. Melissa Webster My research group: Alma Erro and Alice Lee

    Ben Fertic MIT Metropolis and its staff, for providing the space and equipment used to print my designs IT SLOAN SCHOO F MANAGEMENT Acknowledgements : 52
  44. 2 Findings ➔ Conversation made design more accessible than specialized

    tools ➔ Specificity was developed through conversation during the arm design ➔ The interchangeable system revealed AI’s moving “jagged frontier”