agent sidequestsIdeas worth trying

Get an AI-designed 3D print to fit your real object

Measure the object your print must fit, define how the pieces should move, and ask the agent for a small test of that interface before a full print. Keep three things separate: the real measurement, the size you asked the CAD model to make, and the size that came off the printer. A render cannot settle the last one. This guide uses a fictional game-token holder to show which number to change after a failed fit.

Start with the fit you want to observe

Choose a small desk organizer, game insert or similar low-consequence object. Write a physical acceptance check before the agent generates geometry: what goes where, from which direction, and what movement should feel easy? Keep this exercise away from protective, load-bearing, electrical or other safety-critical parts.

For our invented example, a flat rectangular game token should slide into a holder and come out by hand, without bending or forcing it. The holder must also fit inside its game box. Passing the token-width check alone will not establish the box fit, stability or durability. No object in this example was printed or tested.

  • Bring the actual object, measured dimensions with units, photos from useful angles and the locations of protrusions, cables or lips. Mark an unmeasured feature unknown; a photo is not a caliper reading.
  • Record the intended printer, material, orientation and slicer profile if known. Ask for help identifying missing details before planning a print.
  • Keep the measurements and acceptance check in a dated brief. Do not let the agent replace a requirement with a number it measured from its own generated model.

Keep the real measurement separate from the CAD setting

Use a short record for each important interface. A chosen clearance is the intended space between mating parts; it is not a measurement of printer accuracy. State whether a gap means the total difference between widths or the allowance on each side.

In the fictional example, the token width is recorded as 20.0 mm and the chosen total width gap is 0.4 mm. The desired physical opening is therefore 20.4 mm. If the token is centered, that would leave 0.2 mm on each side. These values explain the arithmetic; they are not recommended clearances for your printer.

Illustrative width record, set before generating the model
FieldExample valueWhere it comes from
Object width20.0 mmHuman measurement of the real token; repeat at relevant locations.
Wanted total gap0.4 mmChosen design intent for this example, pending a physical trial.
Desired printed opening20.4 mmObject width plus wanted total gap.
Initial CAD opening20.4 mmFirst design setting, with no process compensation assumed.
Actual printed openingUnknown until measuredA later human observation, never filled from the CAD file.

Check the exported geometry against the fixed brief

Ask the agent to expose named dimensions in editable CAD source or a parameter file. Keep object_width, wanted_total_gap and any later process_adjustment separate. That makes a correction traceable instead of hiding it in a global scale change.

Before printing, open the actual exported model in the available CAD viewer or slicer and compare its units, outer size and critical opening with the brief. Check that the version being sliced is the version you reviewed. If the exported opening differs from the requested one, repair the model or export first; a physical test should not be used to explain away a known digital mismatch.

Inspect how the object enters and leaves, including nearby walls and protrusions. Correct final-position geometry does not show that the object can reach that position. Ask for a view of the full insertion path, and keep unobserved obstructions on the question list.

Make the small test preserve the important interface

Markforged’s Composites Design Guide describes testing a small critical section and comparing variants before updating the complete CAD model. Its published settings are specific to its printing context; we are using the small-test principle, not transplanting its numerical fit ranges.

For the token holder, request a short section that retains the opening, lip, wall shape and direction of insertion. Label the version and changed dimension. If cropping removes the obstruction or makes the walls much more flexible, the sample may answer the wrong question.

  1. Have the person operating the printer review the proposed sample and print setup. Keep the intended material, orientation and relevant settings recorded.
  2. After preparation and cooling appropriate to the process, inspect and measure the sample. Note residue, visible deformation and where the measurement was taken.
  3. Try the actual token through the relevant movement without forcing it. Record both the measurement and the observed behavior, including where it catches.
  4. Use the result only for the interface and setup tested. The full holder still needs its own assembly, box-fit and use check.

Worked example: change the opening, not the whole holder

Imagine the exported opening measures 20.4 mm, but repeated measurements of the printed sample read about 20.1 mm and the token binds. The print is about 0.3 mm narrower than this CAD setting. That observation does not identify the cause or establish a universal printer correction.

First check the version, units, measurement method, visible residue and deformation. If those checks support trying a dimensional adjustment, one bounded experiment is to increase only the CAD opening by 0.3 mm and repeat the representative sample under the same recorded setup. Keep the desired physical opening at 20.4 mm. The extra CAD allowance is a proposed process adjustment, not a new claim about the token or the desired gap.

  • Do not enlarge the entire model to fix one opening: the outside dimensions may already be correct.
  • Do not predict that version B will print at exactly 20.4 mm. Measure it and record what happened.
  • If the token enters but catches farther down, investigate that location or the insertion path before adding more clearance everywhere.
  • If the new sample still fails, bring the measurements and photos to a capable maker or print service. Avoid an unlimited chain of blind reprints.
Fictional revision log: an observed failure and an untested next step
VersionCAD openingPrinted openingDecision
A20.4 mmAbout 20.1 mmToken binds; retain failure and measurements.
B, proposed20.7 mmPendingChange only the opening parameter; print and check a new sample.
Full holderUse the accepted sample settingPendingProceed only after the relevant sample passes; then check the complete object.

Classify the mismatch before asking for another version

Makelab’s fit troubleshooting distinguishes scaling, inadequate clearance, material effects, residue and deformation. Use those as possible explanations to investigate, not a diagnosis from one photo. Keep an unchanged reference file and ask the agent to explain which evidence supports the next edit.

What the observation suggests checking next
ObservationNext checkAvoid
Export and brief disagreeNamed parameters, units and the actual exported versionChanging printer settings to hide a model error.
Print differs from the exportMeasurements, profile, orientation, residue and deformationCalling every mismatch shrinkage.
Object enters, then catchesThe complete insertion path and the exact contact pointWidening an opening that already passed.
Sample works; full object does notWhat changed in geometry, support, flexibility or setupCalling the small sample proof of the full assembly.

Give your agent a bounded handoff

Start with the game-insert or phone-stand prompt below to describe the object, constraints and a paper concept. Those prompts produce plans; they do not already provide a tested CAD-to-print pipeline. Extend the brief only if your assistant has the required modeling and file tools.

A useful follow-up is: “Use my measured dimensions and preserve the agreed fit behavior. Return editable geometry, an export, a record separating measured size from desired gap and CAD settings, and one representative fit-test section. Compare the exported dimensions with my brief. Stop before submitting or starting a print. After I return measurements and a fit observation, propose one justified revision and keep the old version.”

For Muse, Instinct, Dots or Grok Bot, verify access to a suitable CAD or code environment, export tools and a way to inspect the resulting model. This guide does not establish compatibility in any particular account. If those tools are unavailable, keep the measurement sheet and dimensioned concept for a human maker; do not relabel a sketch as a printable file.

What the source report establishes

In an October 27 account, すー|AIと設計 describes using Claude Code for household CAD projects and finding physical problems after digital checks looked satisfactory. The author is a professional mechanical designer selling a paid article and skill bundle. We read the free portion, not the paid instructions, and did not reproduce the projects.

The token-holder record and numerical revision above are our own illustrative exercise. The source does not validate those dimensions, the proposed correction or any of our featured assistants. A model check, a printed sample and a finished object remain separate results.

Copy a prompt and try it

Each idea has a complete starter prompt, the inputs to bring, and the tools it needs.

Common questions

Can AI infer the dimensions from a photo?

A photo can help identify what to measure, but perspective, hidden faces and an uncertain scale can leave important dimensions unresolved. Supply real measurements for mating features and keep any inferred values labeled until checked.

What clearance should I tell the agent to use?

Define the behavior you need, identify total versus per-side allowance, and use relevant guidance for your actual process as a starting point. Then check a representative sample. The numbers in this article are fictional arithmetic, not a universal fit prescription.

Is a valid STL proof that the part will fit?

No. A usable file can still describe the wrong dimensions, print differently from the model or omit an obstruction in the real assembly. Check the export against the brief, then check the physical result against the actual object.

What should I save after a successful fit test?

Keep the measured object dimensions, accepted source and export, sample version, printer/material/profile/orientation, measured result and the fit behavior observed. Treat a new material, orientation or important geometry change as a reason to review what that earlier test still proves.

Sources and further reading

These are practical editorial examples. The prompts have not been independently run-tested across every assistant.

Keep exploring

For platform setup, read the Muse, Instinct, Dots and Grok Bot guides.