Two homepage concepts for Marr Labs' voice agent. Both unsolicited, both live.

Why this one

Marr Labs makes Vox, a voice AI agent that answers phones for businesses in regulated industries: healthcare, lending, insurance, banking. The company is YC W24, founded by Dave Grannan and Han Shu, who built the speech stack behind the first Siri app together at Vlingo before Nuance acquired it for $225M. Han went on to co-found Wyth (acquired by Airbnb), then ran ML and AI groups at Airbnb and DoorDash. This is not a first-time team guessing at voice AI. It's the second act of two people who built the category.

I picked Marr Labs for one specific reason: the product is the hardest kind to show on a webpage. A password manager can screenshot its dashboard. A voice agent's entire value is a sound nobody visiting the homepage can hear. Every competitor in this category, Bland, Retell, Vapi, ElevenLabs, converges on the same visual answer: dark mode, a neon accent, an animated waveform. None of it actually demonstrates anything. It's decoration standing in for proof.

So instead of one redesign, I built two, each solving the "you can't screenshot a voice" problem a different way. Both are concepts. Neither is commissioned. Marr Labs didn't ask for either.

Concept 1: The Directory

The idea. If the product is a voice on a phone line, the site should behave like the object that has always organized phone numbers: a phone directory. The hero isn't a screenshot, it's a listing, and the "demo" is a real phone number you can dial. A Vox agent answers. That's the whole pitch, no video, no sandbox account required.

What it looks like. White-pages paper, Yellow Pages yellow, Bell System blue ink. Barlow Condensed for the directory-listing voice (standing in for Bell Centennial, the actual typeface AT&T commissioned for phone books), IBM Plex Mono for phone numbers and tariffs. Every listing, nav item, and price row uses a dot leader, the little dotted line that walks your eye from a name to a number. It's a structural device lifted straight from the artifact instead of invented from scratch.

What's on the page. A live-feeling "General Directory" listing for Vox with a dialable number. A "What Vox answers for" section laid out as directory listings with extensions (scheduling, lead qualification, customer service, servicing and follow-up). A "Regulated Lines" section in Yellow Pages yellow covering HIPAA, TCPA/CFPB, DOI, and SOC 2, because Marr's actual customers are healthcare, lending, insurance, and banking, not generic SMBs. An "Operators on duty" section crediting Grannan and Shu by their real track record. A published tariff: $1,000 setup, $1,000/month for 4,000 minutes, $0.25/minute after that, with a plain comparison against a $57K/year human phone agent.

Why this works. Every rival in the category tells you their agent "sounds human" and asks you to trust that. This concept doesn't ask for trust, it hands you a phone number in the first five seconds. The 24/7 line-open status indicator and the "answers in under 1 second" claim in the hero aren't marketing copy, they're the same claim every competitor makes, just made falsifiable. You can check it yourself before you've scrolled past the fold.

See "The Directory" live →

Concept 2: The Screenplay

The idea. Marr's real claim is that Vox is indistinguishable from a human on the phone. The only honest way to argue that is to put an actual conversation on the page and dare a visitor to spot which side is the machine. So the homepage opens as a cold open: sluglines, character names, a scene, dialogue that types itself out line by line, exactly like a script.

What it looks like. Script-paper white, Courier ink, and goldenrod, the color of a Hollywood revision page. Courier Prime carries the whole script. A vertical red margin line runs down the page, borrowed straight from a real shooting script. Compliance content sits inside a "goldenrod revision" band, a literal reference to how physical scripts get revised on colored paper when legal has notes on a scene.

What's on the page. A cold-open scene: a dental office at 2:47 AM, a patient calling about a broken crown, a "receptionist" who books the appointment in four lines of dialogue, then a title card asking which character was the machine. A "Cast" section that plays Vox's three main jobs (scheduler, qualifier, service rep) as acting credits, each with a real line of dialogue as its "audition tape." A proof band styled as studio title cards: 1,000,000+ calls performed, under 1 second to answer, 99% task completion, 24/7. The same regulated-industry compliance content as Concept 1, reframed as "every line is delivered to code." The same published tariff, reframed as a production budget.

Why this works. Showing the dialogue and asking "which one was the machine" is a stronger proof than any stat band, because it makes the visitor do the work of trying to catch the AI and fail. That's a more convincing experience than reading a claim. The reveal, that the receptionist answered before the second ring at 2:47 AM on the fortieth call of the hour, does the selling: no human desk operates like that, and now you've watched it happen in the transcript.

See "The Screenplay" live →

What both concepts reject, on purpose

Neither one uses dark mode, a neon accent, or a waveform animation, the three things every voice AI competitor's homepage currently has in common. That convergence is exactly the opening. When an entire category looks the same, looking different is the highest-leverage design decision available, before a single word of copy gets written.

Neither one leads with an AI disclaimer either. Both put the product's actual behavior (a working phone line, a real transcript) ahead of any explanation of how the AI works. Founders in this category tend to over-explain the technology. Buyers don't care how the trick works. They care whether it holds up on a call at 2 AM with their business's name on it.

What this would cost to actually ship

If Marr Labs engaged Plinth to build either concept as a real homepage:

• Discovery + strategy: $5K, 1 week
• Design (hero, sections, mobile, design system): $10K, 2 weeks
• Build (Next.js + Vercel, live call integration, CMS for content): $12K, 2 weeks
• Total: ~$27K, 5 weeks end-to-end

That's an Authority-tier project at Plinth pricing. For a YC-backed company selling into regulated enterprise buyers, a homepage that actually demonstrates the product instead of describing it is the kind of change that shows up in a sales team's close rate, not just a bounce-rate report.

Why I share concept redesigns

Same three reasons as every other unsolicited redesign on this site:

1. It's how design studios get hired. Concept redesigns of recognizable companies have driven more inbound for design studios in the last five years than almost any other content format.
2. It's the best portfolio piece I can make without a client. Real client work is gated by NDAs and time. Concept work is unconstrained, which is why I could ship two takes on the same brief instead of one.
3. It teaches me things. Building two different answers to "how do you show a voice on a page" forced me to articulate two entirely different design systems in one sitting. That's a harder exercise than picking one direction and defending it.

If you work at Marr Labs and want to talk about either concept, I'd genuinely love to. Book a call.

---

Ready to build a site that converts? See Plinth's transparent pricing, four tiers starting at $3,500. Or read how much a website actually costs in 2026.