Claim 01
Web Apps, Not Websites.
Lovable, Bolt, v0, Replit, and Vercel ship single-page websites. We ship multi-window, persistent, desktop-grade applications.
Generative coding tools have converged on the same output: a single React page rendered in an iframe. That works for a marketing site or a CRUD form. It does not work for the software that actually runs a business — file managers, IDEs, design tools, ERP screens, debug consoles — the applications that keep multiple states alive and let the user move fluidly between them.
The constraint is not what the AI can write. It is what the runtime can host. A single-page app cannot meaningfully manage parallel windows, persistent sessions, or cross-window communication. We host applications that can.
|
Frames Studio |
Lovable / Bolt / v0 |
| Output shape |
Multi-window web app |
Single-page website |
| Persistent windows |
Across sessions |
None |
| Prompt-to-deployed app |
End-to-end |
Code only |
Claim 02
MDI as Architecture, Not Nostalgia.
The Multiple Document Interface is not a UX preference. It is the only paradigm in which an AI agent and a developer can collaborate at the same speed.
When an agent generates a component, the developer needs to view the generated code, the rendered output, the database schema, and the debug console — simultaneously. A single-page interface forces sequential attention. The browser becomes a desktop OS: virtual z-index management, focus and activation rules, cross-window messaging, persistent layout.
The industry abandoned MDI when laptops and phones forced single-task UIs onto every developer. The era of AI-assisted development restores the case for parallel attention — and the architecture that made it possible.
|
Frames Studio |
Cursor / VS Code / Replit |
| Window model |
Persistent, parallel |
Linear, modal |
| AI agent integration surface |
Database |
Editor + shell |
| Cross-window communication |
Native |
None / hack-only |
Claim 03
Code as Data, Not Files.
An AI agent cannot reliably manage a Git repository. It can read and write database records with perfect fidelity. The substrate determines the ceiling.
The brittleness of agentic coding tools today is not a model problem. It is a substrate problem. Files in a repository require the agent to coordinate across shells, package managers, version control, build tools, CI/CD APIs, and cloud provider SDKs. Each tool has a different interface, different failure modes, different retry semantics, and different ways of being subtly wrong.
When code is rows in a SQL database, the agent has one interface. Reads, writes, history, deployment, and rollback are all SQL. Tests run against the database. Promotion between environments is a single statement. The reliability of the system is bounded by the reliability of the database — which is bounded by decades of production hardening.
|
Frames Studio |
Devin / Cursor / OpenHands |
| Tool surfaces the agent must master |
1 |
6+ (shell, fs, Git, CI, cloud APIs) |
| Deploy operation |
SQL INSERT |
Build + push + pipeline |
| Failure modes during deploy |
1 (transaction) |
Many (build, lint, deploy, infra) |
Claim 04
Removing a Cost Category, Not Cutting a Line Item.
$450K–$700K of operational overhead per 10-developer team is not a sticker on a vendor invoice. It is the sum of the things you do not realize you are paying for.
License fees are the visible part of the cost. The hidden part is everything that surrounds them — the DevOps headcount required to run the pipeline, the cloud minutes burned by builds that fail before they touch business logic, the onboarding weeks lost while a new hire learns a build system, the security audits required to vet the supply chain.
Removing the toolchain does not produce a cheaper version of the same thing. It eliminates a cost category. That is the difference between a 30% margin improvement and a structural change in how the unit economics of an engineering organization work.
| Line item |
Unit cost |
Annual total (10-dev team) |
| IDE & dev tools (per dev) |
$3,000 |
$30,000 |
| Version control & collaboration |
$25 / dev / mo |
$3,000 |
| CI/CD pipeline |
cloud minutes |
$80,000–$150,000 |
| Security & dependency scanning |
vendor stack |
$40,000–$80,000 |
| DevOps headcount |
1.5 FTE |
$250,000–$375,000 |
| Onboarding (per new dev) |
2–4 weeks |
$50,000–$60,000 |
| Total / yr (10-dev team) |
$450,000–$700,000 |