Landing-page copy: "AI-native, not AI-featured"¶
Status: copy ready to implement. Models on frontend/src/components/Landing/FeaturesSection.tsx
/ PhasesSection.tsx; intended mount point is between FeaturesSection and
CTASection on the landing page. Strategy source:
ai-native-vs-ai-featured-cad.md,
ai-native-cad-pick-a-wedge.md,
positioning: ai-native-positioning.md.
Implementation note: this copy is shipped as
frontend/src/components/Landing/ComparisonSection.tsx(the component names this doc as its copy source). Edit both together when the copy changes.
Section 1 — The comparison (primary)¶
Eyebrow: AI-NATIVE, NOT AI-FEATURED
Headline:
The agent asks the solver. The solver never guesses.¶
Subhead: Every CAD vendor bolted a chat panel onto a 1990s file format. We built it the other way around: an agent that reads your intent, and real solvers that compute every number. No hallucinated impedances. No invented part numbers.
Three questions to ask any "AI-powered" CAD tool¶
A three-column comparison. Left = the question; middle = "CAD with an AI panel"; right =
MagicON. Use a muted/slate treatment on the middle column and a cyan-accent treatment on
the right (matching the hero's cyan-400 accent).
| Ask your tool | CAD with an AI panel | MagicON AI |
|---|---|---|
| Does the AI have my design intent — or is it guessing from geometry? | Infers from a netlist and a polyline. The "why" lives in slides and someone's head. | Captures intent natively — frequency, power target, loss budget, cost ceiling — as first-class data. |
| Can I delegate the goal, or only click faster? | Accelerates the next click. The goal stays your problem. | Delegate the goal. "5.8 GHz, 1 W satellite uplink" → a full module: chain, BOM, stackup, thermal, PDN, export. |
| Where do the numbers come from? | The chatbot. It will happily invent a Df at 28 GHz. | Deterministic solvers. The agent never originates a value — it routes to impedance, thermal, and PDN engines and reports what they compute. |
Caption under the table: If the answer to all three is "the AI panel," you're not using AI-native CAD. You're using CAD with an AI panel — and the difference shows up the moment a design gets ambitious.
CTA row:
- Primary button: Describe your radio → (→ /demo/agent once the public demo ships; until then /app/agent)
- Secondary link: Read the architecture argument (→ the AI-native essay / blog)
Section 2 — The trust strip (secondary, optional)¶
A single horizontal band reinforcing the deterministic boundary. Three short stats or claims with icons (model on the hero's stats row).
Band headline: Probabilistic where it helps. Deterministic where it counts.
| The agent | interprets intent, ranks tradeoffs, drives the workflow. |
| The solvers | compute impedance, thermal margin, and PDN target impedance. Verifiable, reproducible. |
| The boundary | the agent can't invent a number — only a solver can answer one. That's what kills the hallucinations engineers fear. |
Section 3 — Delegate or drive (secondary, optional)¶
Handles the Agent/Guided distinction without confusing the single-pitch hero. Frame it as one tool, two ways in — never as "modes."
Eyebrow: ONE ENGINE, TWO WAYS IN
Headline:
Delegate it to the agent — or drive the solvers yourself.¶
Body: Tell the agent what you're building and it designs the module end to end. Want to see every input? Open the full editor and drive the same solvers by hand — every impedance, thermal, and PDN number is identical, because it's the same engine underneath. Start in the agent, lock it down in the editor, switch back anytime. Nothing is lost.
Two cards (equal width):
- 💬 Agent — "Describe your radio; it designs the module." Best for: first-pass exploration, systems architects, getting from idea to a credible module fast.
- 🧭 Full editor — "Drive the same solvers; audit every input." Best for: locking a design for fabrication, reviewing every decision, learning what the agent did.
Footer line: Same project data. Same solvers. Switch anytime.
Microcopy bank (reusable lines)¶
- "Spec to KiCad in five minutes."
- "300k+ verified parts. Real solvers. Direct KiCad, ODB++, and ADS export."
- "The agent never invents a part number — it searches a verified catalog and reports what it finds."
- "Feature breadth is a stock. Architecture is a flow. We bet on the flow."
- "Own the workflow incumbents do badly: RF module design, end to end."
Do / don't (consistency with positioning)¶
- ✅ Lead every comparison with the deterministic-boundary proof, not feature counts.
- ✅ Keep the hero single-pitch and agent-first; this section is the argument below it.
- ❌ Don't put "Guided mode" on the hero as a co-equal CTA — it reads as AI-featured.
- ❌ Don't claim a closed-loop outcomes flywheel as shipped — say "built for it."
- ❌ Don't point mobile creative at the agent (desktop-first;
<640 pxfalls back to Guided).