← Human viewllms.txt
# Prizmstack Studio
Prizmstack Studio (Prizmstack LLC) is an AI-native product studio: product, design, engineering, marketing and conversion, and support for founders.
Founded in California in 2019; the team of about 15 people works across the Americas, in the client's time zone, in English.
Founder-led: Bryan Gaurano is the client partner on every engagement and still ships.
Development is done exclusively with Claude: Claude Code writes the code on every engagement, from a ticket in Tengu Studio OS to a pull request in the client's own repo, and a person reviews every change before it ships.
Contact: bryan@prizmstack.io. Book 30 minutes with Bryan: https://calendar.app.google/zvGVWBbVkoMkDk6YA. Clutch: https://clutch.co/profile/prizmstack (5.0 from 2 verified reviews).
Work is priced in credits, not hours: every ticket is sized before work starts (XS 1, S 2, M 4, L 8 credits) and its credits are drawn only once it's live in production. Nothing is bigger than an L. Bigger work is split so every piece ships on its own. The price per credit is not published: Bryan gives a straight number on the first call, and https://prizmstack.io/pricing turns a first month into credits.
The studio takes on a few new clients a year and will look at products earlier than people expect (Magai was $10k a month when the work started).
## What we do
- Product: What to build next, decided from your users and your numbers.
- Design: Brand, interface, and the moments that make people stay.
- Engineering: Web, mobile, AI, and whatever your product was built on before we arrived. All of it built with Claude Code.
- Marketing & conversion: Ads, landing pages and funnels, tested until more visitors become customers.
- Support: Your customers, in Intercom, answered by the same team that fixes the bug.
## How we work
- Dedicated: One team on your product, the same people every week, for years. We take on a few new clients a year so that stays true.
- Shipping: Something goes live every week. You see it in a demo.
- Founder-led: Bryan is the client partner on every engagement. He still ships.
- Any starting point: Bubble, Lovable, Replit, or a codebase from three agencies ago. We keep it running while we change it.
## How credits work
1. Ask: In Slack, on a call or in a two-minute screen recording. Tengu turns it into a ticket.
2. Size: An AI model in Tengu sizes it, XS to L, and you approve the credits before anyone starts.
3. Ship: Built with Claude Code, reviewed by a person, QA'd on a live preview, released to production.
4. Pay: Credits are drawn the moment it's live. Not when it starts, not when it's “almost done”.
Ticket sizes:
- XS, 1 credit: A copy change, a broken link, a tweak to an email
- S, 2 credits: A new field or filter, a bug fix, an ad set refreshed
- M, 4 credits: A settings page, an onboarding step, a landing page
- L, 8 credits: Team invites with roles, a Stripe billing flow, a new integration
Nothing is bigger than an L. Bigger work is split so every piece ships on its own.
Hours compared with credits:
- What you pay for: hourly, time someone spent; with credits, work that's live in production.
- When you know the cost: hourly, after the work, on the invoice; with credits, before the work starts.
- When an estimate runs over: hourly, you pay for the overrun; with credits, we do — the size was agreed up front.
- A slow week: hourly, billed anyway; with credits, nothing ships, nothing is drawn.
- What the invoice shows: hourly, hours, by person; with credits, releases, each with a link.
- What the studio is paid to do: hourly, stay busy; with credits, ship.
Questions about credits:
- What counts as shipped? Live in production: a release on Vercel, a build in the App Store or Google Play, a campaign that's running. A preview doesn't count.
- Who decides a ticket's size? An AI model in Tengu sizes it from what the ticket asks for, not from who's asking or who's building. You approve it before work starts. If you disagree, we talk before anyone writes a line.
- What if a ticket turns out bigger? Its size only changes if what it asks for changes, and you approve the new size first. If we misjudged it, that's on us.
- What if something we shipped breaks? Fixing a bug we introduced costs no credits.
- Are calls, planning and QA extra? No. They're part of shipping the ticket. Your statement only shows releases.
- Why no price list? Every product is different, so every plan is too. Bryan gives you a straight number on the first call.
## What every month includes
- The same team: The people on your first call are the people on your product a year later.
- A live demo: Every week you see what shipped, running — not a status report.
- Your code, your accounts: Your GitHub, your Vercel, your stores. You own all of it from day one.
- A shared Slack channel: Talk to the people doing the work, not an account manager.
## After someone writes in
1. Bryan reads it: Within one business day, with notes on your product.
2. A 30-minute call: What it is, what's not working, and whether we're the right team.
3. Your first month, sized: The tickets we'd ship first, each with its size and credits, and who'd be on it.
## Work
### Magai (AI platform) — https://prizmstack.io/work/magai
Client work.
Summary: An all-in-one AI platform — multi-model chat, image and video, personas, team workspaces, billing. We ran product from the first paying customers and shipped every week for two years.
Problem: Magai had its first paying customers and $10k a month in revenue. The product had to cover a lot at once — multi-model chat, image and video, personas, team workspaces and billing — and keep growing without slowing down.
What we did: Ran product from the first paying customers onward. Shipped every week, for two years. Architected the billing and org system that team workspaces run on. Fixed billing and the product itself before adding support agents — so support load fell per subscriber instead of rising.
Result: Monthly recurring revenue went from $10k to $100k over two years of weekly releases. More recently, support conversations per subscriber fell 70% in four months — after we fixed billing and the product, not by adding agents.
Numbers: 10× (MRR, $10k → $100k, over two years); −70% (support conversations per subscriber, in four months); 2 yrs (of weekly releases)
Founder: “I started Magai because AI felt out of reach for most people. My whole mission has been to change that.” — Dustin W. Stout, Founder, Magai · one of MSN's Top 10 AI Leaders to Follow in 2026
### Senna Research (Research e-commerce) — https://prizmstack.io/work/senna-research
Client work.
Summary: A research-materials store that shipped as sixty-six identical vials in a single grid. We rebuilt the storefront into five research categories with a featured set and a documented-lot brand — and we run the ads too.
Problem: The store shipped as sixty-six identical vials in a single grid. Nothing told a buyer where to start, what was different, or why to trust the product.
What we did: Rebuilt the catalog into five research categories. Added a featured set so buyers have a clear place to start. Built a documented-lot brand that puts the proof up front. Run the ads that bring buyers to it.
Result: One undifferentiated grid became a storefront a researcher can navigate — and we keep feeding it traffic.
Numbers: 66 → 5 (identical vials, rebuilt into five categories); 1 (featured set up front); Ads (run by us, ongoing)
### FijiApp (Brokerage valuation) — https://prizmstack.io/work/fijiapp
Client work.
Summary: Tells brokerage owners what their business is worth and connects them with buyers, anonymously. Valuation engine, marketplace, FIJI Insights — and now the marketing.
Problem: Brokerage owners want to know what their business is worth, and to meet buyers, without announcing to the market that they're thinking of selling.
What we did: Built the valuation engine that tells owners what their brokerage is worth. Built the marketplace that connects them with buyers, anonymously. Built FIJI Insights. Now run the marketing too.
Result: Subscription revenue grows 10–15% every month. In 2026, FIJI took 2nd place out of 16 at the T360 tech battle.
Numbers: +10–15% (subscription revenue, monthly); 2nd of 16 (at the T360 tech battle, 2026); 3 (products: valuation, marketplace, Insights)
Client review: “Prizmstack is the best software development company we've ever worked with — and we've worked with many.” — Brad Clayton, President, FIJI · verified review, Mar 2026
### Dosify (Mobile health) — https://prizmstack.io/work/dosify
Client work.
Summary: Supplement and protocol tracking with safety checks, on iOS and Android from one codebase. We designed and built the app, the brand, and the launch — email, TikTok, community.
Problem: Supplement and protocol tracking needs safety checks you can trust, and it needs to live on the phone people already carry — iOS and Android both.
What we did: Designed and built the app on Expo, so iOS and Android ship from one codebase. Built safety checks into protocol tracking. Made the brand. Ran the launch across email, TikTok and community.
Result: One team, one codebase, two app stores, and a launch that went out with the product instead of after it.
Numbers: iOS + Android (from one Expo codebase); 1 PR (ships both platforms); 3 (launch channels: email, TikTok, community)
### AI estimator for a demolition contractor (Construction · AI) — https://prizmstack.io/work/construction-estimator
Client work.
Summary: Quoting a job meant a senior estimator reading blueprints for most of a day. We built an AI takeoff pipeline from plan PDFs to a bid workbook, with a review screen for the estimator.
Problem: Quoting a demolition job meant a senior estimator reading plan PDFs for most of a day. Every bid waited on that person's time.
What we did: Built an AI takeoff pipeline that reads plan PDFs and pulls out the quantities an estimate needs. Writes the result into a bid workbook. Gave the estimator a review interface to check and correct every line before a bid goes out.
Result: Quoting a job went from 8 hours to 10 minutes — and the estimator still reviews every line before a bid goes out.
Numbers: 8h → 10m (to quote a job); PDF → bid (plan PDFs in, bid workbook out)
### AR billing and customer portal for Hoskins Equipment (Equipment rental) — https://prizmstack.io/work/equipment-rental-portal
Client work.
Summary: Invoicing lived on a green-screen ERP and a spreadsheet. We built the customer portal, the billing app that replaced the ERP screens, and the bridge into QuickBooks — a modern layer that works with the legacy system instead of replacing it.
Problem: Invoicing lived on a green-screen ERP and a spreadsheet. Replacing the legacy system outright wasn't on the table — the business runs on it.
What we did: Built a customer portal where customers see and pay their invoices. Built the billing app that replaced the ERP screens for the team. Built the bridge into QuickBooks. Designed the whole thing as a modern layer that works with the legacy system instead of replacing it.
Result: More than $4M collected through the portal in its first fifteen months.
Numbers: $4M+ (collected in the first fifteen months); 0 (legacy systems ripped out); 3 (pieces: portal, billing app, QuickBooks bridge)
Client review: “Prizmstack's speed and communication are impressive. As far as agencies go, they're pretty cutting-edge in terms of AI tool use.” — IT Manager, Hoskins Equipment · verified review, Apr 2026
### Moneo (SOP software · AI) — https://prizmstack.io/work/moneo
Our own product.
Summary: Screen recording in, step-by-step SOP out. Moneo reads the pixels, not the page, so it documents green screens and legacy ERPs that other SOP tools can't see. We designed it, built it, and run it — billing, site and all.
Problem: SOP tools read the browser's page structure, so they see nothing in the software a lot of back-office work still runs on: green-screen ERPs, AS/400 terminals, old Windows apps. The steps live in the heads of the people who've used those systems longest. We built Moneo to write them down from a screen recording.
What we did: A pipeline that reads pixels, not the page: Whisper transcribes the narration, the frames where the screen changed become steps, and Claude reads each one — screen names, function keys, field labels — into plain language. A three-panel builder — transcript, editable steps, and every frame of the recording — so a step that grabbed the wrong screenshot is one swap away, not a re-record. Recording in the browser that can switch screens mid-take, plus arrows, highlights and blur before anything is published. Publishing to a web viewer, Word, PDF, or Markdown for AI tools, inside team workspaces with folder-level access. Then the business around it: Stripe billing on SOP credits, a free plan, and the site at usemoneo.com.
Result: Five weeks from a single-user MVP to a live, multi-tenant product with billing — and it keeps shipping. Moneo is live at usemoneo.com, free to start.
Numbers: 5 weeks (from a single-user MVP to live, billed SaaS); Any screen (it reads pixels, so green screens, AS/400 terminals and legacy ERPs work too); 92 (pull requests merged, July to October 2026)
### Tengu Studio OS (Studio OS) — https://prizmstack.io/work/tengu
Our own product.
Summary: Tengu Studio OS sits under every engagement, and it's built with Claude Code. Any ticket runs in Claude Code with one click: a session in your repo opens the pull request, Claude QA tests its preview, and a person signs off before it ships. Humans decide. Agents carry.
Problem: Most of what goes wrong between a client and a studio happens in the gaps: a request made on a call that never becomes a ticket, a Slack message that sinks, a founder who has to chase someone to find out where things stand. And once Claude Code started writing our code, there was a new gap: agent work living in terminals nobody else could see. We built Tengu so every request is a ticket, Claude Code works from the same tickets our people do, and every change is checked before it ships.
What we did: Built with Claude Code, end to end. From September 23, when Tengu's current codebase was started, to October 30, 554 of the 582 pull requests merged into it came from Claude Code sessions; most of the rest are Dependabot bumps. A person merges each one, and main deploys straight to tengu.work. Run in Claude Code on any ticket. One click starts a Claude Code session in the list's own repo with the ticket as the brief. It plans, opens a PR with a live preview, and the ticket follows it from plan to live. When Claude needs a decision, it asks on the ticket. Its ticket turns orange, and whoever answers writes straight back into the session. Nobody hunts through terminals for a run that's stuck. QA with Claude. A second session tests the PR's preview against what the ticket asked for and posts what passed, what failed with steps to reproduce, and what to fix. A person passes it or sends it back. We build exclusively with Claude Code, so the same loop runs on every client account. Ask in Slack, on a call or in a two-minute screen recording and it becomes a ticket, the code lands in the client's own repo, and the Agents tab shows every Claude run across every project. Each week rolls up into a demo of what shipped. Humans decide what's next. Agents carry.
Result: Tengu ships itself through Tengu. Claude Code pull requests merged a week went 47, 86, 133, then 194 across October, each one live on tengu.work once merged. The Run in Claude Code button on our own tickets is the same one on every client's, so how we build Tengu is how we build their products.
Numbers: 194 (Claude Code pull requests merged and live on tengu.work in one week, Oct 19–25); 4× (the weekly pace over four weeks: 47, 86, 133, then 194); 554 of 582 (pull requests merged into Tengu came from Claude Code sessions)
## Products we run ourselves
- Moneo: video to SOPs
- MarketClip: AI marketing video
## Stack
- Claude Code (AI): All of our development, exclusively. Every product we ship is built with it, Tengu and this site included. People review every change and make every decision.
- Claude Design (Design): Screens, prototypes and design systems start in Claude Design, then Claude Code builds them.
- Claude Cowork (Operations): Everything around the code: research, docs, reports and the studio's day-to-day work, done with Claude.
- Next.js (Web): Every web product we ship. App Router, server components, edge by default.
- React Native + Expo (Mobile): Dosify runs on iOS and Android from one codebase. Ship both from one PR.
- Supabase (Data): Postgres, auth, realtime and storage. The backend that gets out of the way.
- Vercel (Infra): Previews on every branch, edge everywhere. This page is running on it right now.
- TypeScript (Language): End to end. The same types in the database, the API and the screen.
- Tengu Studio OS (Ours): Our own studio OS. Every request becomes a ticket, and every ticket passes review and QA. The studio runs on it.
## Guides
- Bubble to Next.js — without freezing the product: https://prizmstack.io/bubble-to-nextjs — Moving off Bubble shouldn't mean six months of nothing shipping. We keep your Bubble app running and move the parts that are actually hurting — onto Next.js, Supabase and Vercel — while your customers keep using the product.
- Built it with Lovable, Bolt or Replit? We'll take it from here: https://prizmstack.io/lovable-app-takeover — AI builders are a great way to get a product to its first users — and a hard place to grow it. We take over apps built with Lovable, Bolt, v0 or Replit, keep them running, and turn them into products a team can ship on every week.
- An AI-native product studio: https://prizmstack.io/ai-product-studio — We build AI into products — an estimator that quotes a demolition job in ten minutes instead of eight hours, a multi-model platform that grew ten-fold — and we build exclusively with Claude, on our own studio operating system, so something ships every week.