# AI product design concepts: 3 simulation cycles to prototype or kill

Charlotte Higgins · November 7, 2026

> Prototype or kill AI product concepts with 3 simulation cycles: geometry, physics, and DFM. All green and under 10% of first-article cost to proceed.

This guide explains when an AI-generated product concept should move on to prototyping, using three pre-registered simulation cycles: geometry/clearance, physics/performance, and DFM/tolerance.

A concept moves forward only if every cycle gets a green light and the total simulation cost stays below 10% of the first-article cost; any red light stops it immediately.

![This guide explains when an AI-generated product concept — AI product design concepts](https://static.mm-ais.com/article-images-ai/ai-product-design-concepts-3-simulation-ai-643152dd.jpg)

## How the three-cycle AI gate actually works

The three-cycle AI gate is a sequential go/no-go process: geometry/clearance, physics/performance, and DFM/tolerance must each pass their own defined criteria before the concept can move on to prototyping. The first cycle checks the bounding box, minimum wall thickness, draft angle, and tool access against a written set of requirements; if any check fails, the concept is stopped right away, and no physics or DFM work is done. Cycle 2 runs one nominal case plus one sensitivity case for load, deflection, thermal, or fatigue using specified material data, and it must show sufficient margin over the allowable limit before moving on. Cycle 3 reviews tool radius, undercuts, wall ratios, tolerance stack-up, and vendor availability; a failed DFM gate means the concept does not advance. Only when all three gates are green and the total simulation cost stays below ten percent of the first article does the concept move forward to prototyping.

Each gate is a separate checkpoint with its own pass/fail threshold, so a concept can fail in cycle 1 without using up engineering time on cycles 2 or 3. That order matters because simulation costs rise quickly once physics and DFM analysis begin.

d gate caps total simulation spending at ten percent of the first article's cost, which means the team can't rack up hidden solver fees across all three cycles without hitting the budget ceiling. A concept that passes the geometry check with room to spare but fails physics or DFM must not move on to a prototype build, because the gate system exists specifically to prevent that kind of waste. The project is killed at the cycle where it fails, before any prototype dollars are spent.

 The gate sheet itself is simple—three checks, each with a defined threshold—but the discipline required to enforce it consistently is what separates a disciplined simulation program from an undisciplined one.

The gate sheet uses concrete thresholds that an engineer can verify against the project's material data and vendor list. The geometry/clearance check reviews bounding box fit, minimum wall thickness, draft angle for the chosen process, and tool access clearance against a written method gate. The physics/performance check runs one nominal case and one sensitivity case for the dominant load path, deflection, thermal condition, or fatigue cycle, and requires the calculated margin to exceed the allowable limit by a pre-registered amount before Cycle 3 is authorized. The DFM/tolerance check reviews tool radius, undercut feasibility, wall ratio, tolerance stack-up, and vendor availability. The tolerance stack-up limit is a pre-registered fraction of process capability, set before the first solve, and any stack-up that approaches or exceeds it triggers a red gate. Together, these thresholds form a complete gate sheet that a single engineer can quickly audit once the method is documented.

The hybrid three-cycle gate is the recommended approach because it combines simulation discipline with selective physical validation. A simulation-only approach avoids most prototype failures, but it can miss manufacturing realities that only become apparent through a hands-on build. A prototyping-only approach skips gate discipline entirely and often leads to designs that fail on their very first physical build. The hybrid approach keeps the three-cycle gate as the release criterion, while allowing for a small number of prototype builds to validate the most critical assumptions identified during simulation. Sources such as WiredWhite's 2026 coverage of Simulation-Driven Engineering, the SimOps article on cost reduction, and the Eureka PatSnap comparison of simulation-driven design versus prototyping costs all reach the same conclusion: simulation-first iteration with gate discipline reduces prototype rework and shortens development cycles more reliably than either approach alone.

![How the three-cycle AI gate actually works — AI product design concepts](https://static.mm-ais.com/article-images-ai/ai-product-design-concepts-3-simulation-ai-387b136e.jpg)

## Evidence and Limits: One Number to Verify

Every claim behind the three-cycle gate rests on evidence, and the honest truth is this: none of the published sources provide a transferable number. Each one supports a mechanism, a trend, or a claimed benefit, but none gives a specific cost, error rate, or percentage that you can plug directly into your own gate sheet. That isn't a weakness of the method—it's exactly why you need to pre-register your thresholds. Set them before running any simulations, based on your own first-article baseline, not on someone else's blog post.

| Source | Grounded claim | Figure status | Metric to extract |
| --- | --- | --- | --- |
| WiredWhite, Apr 8, 2026 | Simulation-driven engineering and digital twins are the standard for 2026; iterate in simulation rather than through physical builds | No cost or error figures | Mechanism only |
| SimOps 101, Sep 9, 2026 | Simulation lowers prototyping costs, shortens development cycles, and reduces rework; claims a measurable ROI | No percentage provided | Record the claimed baseline, not a specific number |
| Eureka/Patsnap, Mar 6, 2026 | Combining simulation with selective rapid prototyping maximizes cost benefits | No dollar figure | Record the criteria for deciding when to use prototyping |

Treat this table as a fact-checking exercise. For each row, the "Figure status" column is your checkpoint: if a vendor or consultant quotes you a savings percentage "backed by industry research," ask which of these sources actually contains that number. The answer is none of them. The SimOps claim of a measurable ROI is useful as a directional indicator—it justifies running the physics and performance cycle in the first place—but the actual ROI figure must come from your own logged simulation costs compared against your own first-article costs.

Apply the same level of scrutiny to the commonly cited minimum tool-radius figure used to debunk design myths. The claim that a minimum tool radius resolves most geometry and clearance failures is widely repeated, but none of the available sources publish a specific radius value. Treat any radius quoted to you as a hypothesis that you need to verify against your own CAD clearance study in cycle 0.

ne, not as a grounded constant.

The limits are equally concrete. The ScienceDirect work on automatic simulation model generation (2025) shows models can be generated and validated from layout descriptions, but it does not certify that generated models match physical behavior — that gap is exactly what the physics/performance gate exists to close. The MIT work on converting 2D designs to 3D models improves prototype readiness but says nothing about tolerance stacks, which is why the DFM/tolerance cycle remains separate.

Your one-number verification rule: before releasing any concept to prototype, confirm that every threshold on your gate sheet traces to your own pre-registered baseline, and that no threshold was copied from a source in the table above. If a number on the sheet has no internal provenance, the gate is red regardless of what the simulation says.

## Simulate, prototype, or hybrid: pick one

When an AI-generated concept lands on your desk in 2026, you have three ways to move it forward: simulate everything, prototype everything, or run a hybrid gate. Simulation-driven engineering has become the norm for iterating designs digitally before committing to hardware, as WiredWhite describes for 2026, but "the norm" is not the same as "the right gate." The comparison below is the decision this section owns: which option wins, and why.

| Option | Cycle 1 geometry | Cycle 2 physics | Cycle 3 DFM | Decision |
| --- | --- | --- | --- | --- |
| Simulation-only | Yes | Yes | Weak | Reject as sole gate |
| Prototype-first | Late | Late | Real but expensive | Reject for concept screening |
| Hybrid three-cycle gate | Gate | Gate | Kill before CAD | WINNER |

The hybrid three-cycle gate wins because it converts simulation from a report into a pass/fail checkpoint. A simulation-only approach runs geometry and physics well but stumbles on DFM: your models cannot see the vendor's tool access, draft angles, or fixture constraints unless someone feeds them in, and that feed usually happens too late. Prototype-first has the opposite failure mode — it learns DFM early and honestly, but it learns geometry and physics only after you have paid for parts, which is the most expensive possible sequencing for concept screening.

The hybrid fixes the sequencing problem. Cycle one clears geometry and clearance before any physics run is worth computing. Cycle two clears physics and performance only for concepts that survived cycle one. Cycle three applies DFM and tolerance checks while the concept is still cheap to kill — before detailed CAD, before tooling quotes, before anything that costs real money to change. Each cycle is a gate, not a memo: a concept that fails is dead at that line, and the surviving concept carries forward only cleared evidence.

Simulation still does the heavy lifting inside the hybrid. SimOps notes that simulation cuts prototype costs, shortens development cycles, and reduces rework, and the Eureka/PatSnap comparison of simulation-driven design versus physical prototyping points to the same conclusion: combining simulation-driven design with selective rapid prototyping creates a hybrid approach that maximizes cost efficiency. That is exactly the structure here — simulation as the filter, prototyping as the confirmation step reserved for concepts that already cleared all three gates.

Your check when choosing among the three options: ask where each option first detects a DFM failure. Simulation-only detects it late or not at all. Prototype-first detects it after the spend. The hybrid detects it in cycle three, before CAD, at the lowest possible cost of being wrong. If a proposed workflow cannot answer that question with a specific cycle, it is not a gate — it is a report, and reports do not kill concepts.

## Cost math that forces a kill decision

The kill decision lives in the cost math, not in opinions about the design. The method formula for each cycle is C_cycle = H_eng × R_loaded + C_solver + C_data, and the total is C_total = Σ C_cycle across all three cycles. The release rule is a ceiling: if C_total exceeds 10% of the first-article quote, you defer the prototype until simulation spend drops—regardless of how green the gates look. Simulation-driven methods are widely credited with cutting prototype costs and rework (SimOps), which is exactly why the spend cap is the honest constraint: it keeps the simulation budget from becoming its own sink.

The gates carry numeric floors, but none of them are transferable constants — as the evidence section above shows, no published source in this article's grounding supplies a margin percentage or a stack ceiling you can lift. The method therefore requires you to pre-register your own margin floor for Cycle 2 and your own tolerance-stack ceiling for Cycle 3, derived from your first-article baseline and process capability, before the first solve. Once written down, a red result is a kill rather than a negotiation; the specific values are yours to set and defend, not to quote from an external source.

When a gate goes red, the decision is a comparison of two burn paths, not a judgment call. An early wrong prototype costs the first-article quote plus your fixture and rework estimate. A late wrong simulation costs C_total plus one re-run. Pick whichever expected burn is lower; the sources on simulation versus prototyping costs (Patsnap/Eureka) frame this trade-off as the core economic argument for hybrid approaches, but the method makes it concrete per-concept.

One check: write the two competing cost estimates on the gate sheet *before* opening the solver, so a red gate triggers arithmetic instead of hope.

A worked structure makes this audit-ready. Compute C_cycle for each cycle from your loaded engineering rate, solver cost, and data cost; sum them; divide by the first-article quote; and compare to the 10% ceiling. If the ratio clears, the gates are green, and both the 20% margin and 80% stack checks pass, the concept releases to prototype. Anything else is a kill at the first red gate—no "one more simulation," no partial release. Frameworks for automatic simulation model generation (ScienceDirect) are reducing the human setup labor that drives H_eng, which is how C_total stays under the cap as concepts iterate.

## Gate sheet with 3 checks

The gate sheet is where the three-cycle rule stops being a principle and becomes a checklist you can fail. Before any AI-generated concept moves toward prototype, fill in one row per cycle using the cost formula from the cost-math section: C_cycle = H_eng × R_loaded + C_solver + C_data, summed across all three cycles. Then divide that total by your own first-article quote and compare against the 10% method cost gate. Do not borrow a worked dollar example from any external source — none of the grounded sources publishes one. If your own quotes push total simulation spend past the 10% line, the method itself is too expensive for the part, regardless of how the concepts score.

Run the checks in order and record each result on the sheet the moment it lands, against the gate values you pre-registered for your own part. Cycle 1, geometry/clearance: compare measured clearance and draft angle to your written gate. Cycle 2, physics/performance: compare deflection and safety factor to your pre-registered allowable and margin floor. Cycle 3, DFM/tolerance: compare wall ratio and tolerance stack to your method floor and stack ceiling. A red result at any cycle kills the concept at that line, subject only to the bounded exceptions defined in the if/then rules — one sensitivity case for a marginal Cycle 2 failure, or one written vendor confirmation and re-check for Cycle 3. There is no fourth cycle and no open-ended appeal.

The sheet's discipline is that every cell is pre-registered: gate values are written down before the solver runs, not after. This mirrors the direction the field is already moving — simulation-driven engineering with AI-generated models is now standard practice, as WiredWhite described in April 2026, and frameworks for automatic simulation model generation documented in ScienceDirect work convert layout descriptions into validated, structured simulation inputs. Pre-registration is what keeps that automation honest: the machine generates the model, but the gate values were human commitments made before any result existed.

Two rules make the sheet actionable on your next concept. First, kill on the first red gate even when earlier cycles passed with margin — a strong Cycle 2 result does not buy forgiveness for a failed Cycle 3, and the simulation spend already incurred is the cost of finding out cheaply. Second, release to prototype only when both conditions hold: all three rows green and total simulation spend below the 10% method cost gate. The economics behind that pairing are priced out in the cost-math section above, so here it is simply the release rule you apply.

## If/then rules for the 2026 gate

The 2026 gate runs on five if/then rules, and each rule exists to stop exactly one way a bad concept sneaks through. Run them in order, because an out-of-order exception is how simulation budgets get burned on concepts that were already dead.

Rule 1: If Cycle 1 (geometry/clearance) is red on clearance interference or tool access, kill immediately — do not run Cycle 2, and do not spend solver time averaging or re-meshing to soften the failure. A geometric blocker does not improve with more physics runs; it only consumes budget that a viable concept could have used. The kill is the decision; there is no "promising but red" category.

Rule 2: If Cycle 2 (physics/performance) is red on performance targets, or if the run used generic rather than verified material data, you get exactly one sensitivity case to test whether the failure is marginal or structural. If the sensitivity case is still red, kill. This is the only appeal in the method, and it is deliberately limited to one run — as simulation-driven engineering sources note, simulation's cost advantage comes from early kills, not from endless refinement of failing designs (WiredWhite, 2026).

Rule 3: If Cycle 3 (DFM/tolerance) is red, kill — unless a named vendor confirms the exact cutter geometry and the full tolerance stack in writing. With written confirmation, run exactly one re-check at the corrected parameters; if it clears, the gate flips green. Without written confirmation, the gate stays red no matter how confident the vendor sounded on a call.

Rule 4: If all three gates are green but total simulation spend exceeds the method cost gate of 10% of first-article cost, stop before prototype release and renegotiate the simulation scope — not the gates. The gates are the quality check; the cost gate is the economic check, and simulation's ROI depends on holding both (SimOps, 2026; PatSnap Eureka, 2026). A green-gate concept that cost too much to validate is still a cost problem, and the fix is scope reduction on the next revision, not waiving the gate.

Rule 5: Release to prototype only when all three gates are green with margin and the cost gate passes. Marginal greens — a clearance that clears by a rounding error, a tolerance stack with no headroom — should be treated as a hybrid-flag: the concept releases, but the prototype must specifically probe the thin margin, per the hybrid simulation-plus-selective-prototyping logic described by PatSnap Eureka (2026). The check to run before release is simple: list the margin on each gate; if any margin is within measurement uncertainty of red, write that number into the prototype test plan.

Every exception above is named and bounded — one sensitivity case, one written vendor confirmation, one re-check. If you find yourself needing a second exception on the same concept, that is not an exception; that is the kill signal arriving late.

## What to do next

| Step | Action | Why it matters |
| --- | --- | --- |
| 1 | Pre-register the three simulation cycles in fixed order—geometry/clearance, physics/performance, then DFM/tolerance—and obtain a first-article quote before running any analysis. | Locking the sequence and the first-article cost baseline ensures every gate decision is objective and comparable against the method cost gate. |
| 2 | Run the geometry/clearance cycle and record a clear pass or fail against its gate criteria. | This first cycle confirms the AI-generated concept can physically fit and clear; a red result stops further work immediately. |
| 3 | If the geometry/clearance gate is green, run the physics/performance cycle; if it is red, kill the concept and do not proceed. | Enforcing the kill-on-first-red rule prevents wasted effort on later simulations or prototyping for a concept that has already failed. |
| 4 | If the physics/performance gate is green, run the DFM/tolerance cycle; if it is red, kill the concept and do not proceed. | DFM/tolerance validation confirms the concept can be manufactured with margin, completing the evidence needed for a prototype decision. |
| 5 | Compare cumulative simulation spend to the method cost gate relative to the first-article quote, then decide: release to prototype only when all three gates are green and spend is below the gate; otherwise defer or kill. | This final check ensures the concept is technically ready and that simulation investment remains justified before committing to physical prototyping. |

## Frequently Asked Questions

**What happens to later simulation cycles if a concept fails the geometry and clearance checks?**

If any geometry or clearance check fails, the concept is stopped right away and no physics or DFM work is done.

**What specific design features are evaluated during the geometry and clearance cycle?**

The first cycle checks the bounding box, minimum wall thickness, draft angle, and tool access against a written set of requirements.

**What types of cases must be run during the physics and performance cycle?**

Cycle 2 runs one nominal case plus one sensitivity case for load, deflection, thermal, or fatigue using specified material data.

**What result must the physics and performance cycle demonstrate before a concept can proceed?**

The physics and performance cycle must show sufficient margin over the allowable limit before moving on.

**Which manufacturing-related factors are evaluated in the DFM and tolerance cycle?**

Cycle 3 reviews tool radius, undercuts, wall ratios, tolerance stack-up, and vendor availability.

**Does a failure in the DFM and tolerance gate allow the concept to move forward?**

A failed DFM gate means the concept does not advance.

## Quick answers

| What are the three pre-registered simulation cycles used to decide whether an AI-generated product concept should move to prototyping? | The three cycles are geometry/clearance, physics/performance, and DFM/tolerance. |
| --- | --- |
| Under what conditions does a concept move forward to prototyping? | A concept moves forward only if every cycle gets a green light and the total simulation cost stays below the specified percentage of the first-article cost. |
| What happens if a concept fails the first geometry/clearance cycle? | If any check in the first cycle fails, the concept is stopped right away, and no physics or DFM work is done. |
| What does the physics/performance cycle evaluate? | Cycle 2 runs one nominal case plus one sensitivity case for load, deflection, thermal, or fatigue using specified material data, and it must show sufficient margin over the allowable limit before moving on. |
| What aspects are reviewed during the DFM/tolerance cycle? | Cycle 3 reviews tool radius, undercuts, wall ratios, tolerance stack-up, and vendor availability, and a failed DFM gate means the concept does not advance. |

Also worth reading: **How Enterprise Teams Can Accelerate Product Innovation With AI Concept Generation**: [How Enterprise Teams Can Accelerate](https://graftconcepts.com/blog/how_enterprise_teams_can_accelerate_product_innovation_with_ai_concept_generation.php) · **Validate Artificial Intelligence Product Concepts: 3-Gate Review — Build or Kill**: [Validate Artificial Intelligence Product Concepts:](https://graftconcepts.com/blog/validate-artificial-intelligence-product-concepts-3-gate-review-build-or-kill.php) · **Why Concept-Stage Simulation Cuts Engineering Change Orders 40%**: [Why Concept-Stage Simulation Cuts Engineering](https://graftconcepts.com/blog/why-concept-stage-simulation-cuts-engineering-change-orders-40.php)

### Related reading

- [How Simulation-Driven CAD Compresses Prototype Development](https://graftconcepts.com/blog/how-simulation-driven-cad-compresses-prototype-development.php)
- [Using AI to explore product concepts your competitors haven't thought of](https://graftconcepts.com/blog/using-ai-to-explore-product-concepts-your-competitors-havent-thought-of.php)
- [Measuring ROI on AI Product Concepts](https://graftconcepts.com/blog/measuring_roi_on_ai_product_concepts.php)
- [2026 Generative Design: 40% Faster Iteration Cycles](https://graftconcepts.com/blog/2026-generative-design-40-faster-iteration-cycles.php)
- [AI product ideas: 7-Point DFM Rubric vs 65% Scrap — Build or Kill](https://graftconcepts.com/blog/ai-product-ideas-7-point-dfm-rubric-vs-65-scrap-build-or-kill.php)
- [Why Concept-Stage Simulation Cuts Engineering Change Orders 40%](https://graftconcepts.com/blog/why-concept-stage-simulation-cuts-engineering-change-orders-40.php)

### Latest

- [AI product ideas: 7-Point DFM Rubric vs 65% Scrap — Build or Kill](https://graftconcepts.com/blog/ai-product-ideas-7-point-dfm-rubric-vs-65-scrap-build-or-kill.php)
- [How to judge AI part designs: 3-point DFM scorecard with ISO 2768-m — Build or...](https://graftconcepts.com/blog/how-to-judge-ai-part-designs-3-point-dfm-scorecard-with-iso-2768-m-build-or-kill.php)
- [AI Concept Generation vs Human Design: A Verify-Before-You-Commit Guide](https://graftconcepts.com/blog/ai-concept-generation-vs-human-design-a-verify-before-you-commit-guide.php)

Canonical: https://graftconcepts.com/blog/ai-product-design-concepts-3-simulation-cycles-to-prototype-or-kill.php
Markdown: https://graftconcepts.com/blog/ai-product-design-concepts-3-simulation-cycles-to-prototype-or-kill.php/index.md