The engine owns engineering. You own scope.
CAIA replaces the engineering decision layer between business intent and deployed software with an autonomous engine, and exposes a navigation UI to a non-coder operator. Every architecture, stack, integration, and infrastructure choice is engine-resolved. The human owns scope.
The killer demo, in one paragraph.
A Figma file becomes a deployed UI in the customer’s existing codebase, conforming to the customer’s own conventions, in under 10 minutes.
The buyer is US retail and e-commerce mid-market ($50M–$500M revenue). The moat triplet is read-only legacy adoption, exportable repositories, and an MCP-first marketplace shipped at MVP. We are raising $2.5–4M pre-seed/seed to convert verbal interest into 25 paying customers and $250K–$1M ARR by Q4 2027.
Non-coders cannot ship enterprise-grade software.
The engineering talent shortage is structural. Low-code platforms own the runtime. AI coding tools require a developer in the seat. The mid-market is stuck.
- • IDC forecasts a 1.4M developer shortage in the US alone by 2027.
- • Gartner: 70%+ of low-code projects fail to scale past their initial use case.
- • Mid-market CIOs report 200+ internal app requests, capacity for ~40 of them.
An engine that owns engineering decisions, and a UI that exposes scope.
CAIA is built on Claude subagents via the Claude Code SDK. The engine resolves architecture, stack, integration, and infrastructure choices. The operator never sees code — they see the decisions being made and approves the ones that need approval.
- • Interview agent elicits scope in business language.
- • PPO decomposition breaks intent into engineering decisions.
- • Hero loop: Figma → deployed UI in <10 minutes.
- • Greenfield + adoption flows: works against existing codebases.
A four-layer architecture.
- 1. Interview agent — captures scope as a structured spec.
- 2. Decision engine — resolves architecture and stack choices against the customer’s conventions.
- 3. MCP integration layer — talks to the customer’s systems via Model Context Protocol servers.
- 4. Repo emitter — produces a standard repository in the customer’s preferred VCS.
Watch the ten-minute Figma loop.
Recorded on a customer’s production environment, with consent. No edits.
Three moats. We ship all three at MVP.
Read-only-first legacy adoption
CAIA can run against systems that cannot be modified. We read your repo, your schema, your conventions — without asking permission you cannot get.
Exportable repositories
Output is a standard repo on your own GitHub, GitLab, or Bitbucket. Deployed through your existing CI/CD. No proprietary runtime, no hosted lock-in.
MCP-first marketplace
Every integration is a Model Context Protocol server. We ship the marketplace at MVP. The long tail of integrations becomes a community surface.
Five forces, one window.
Claude Code adoption in enterprise dev orgs since 2024
Active MCP servers across enterprise infrastructure
Forecast TAM for AI-native engineering platforms by 2030
Of mid-market roadmaps now have an AI-tooling line item
Window before the major clouds bundle an alternative
TAM. SAM. SOM. Cited.
$40B addressable, $2.5–3.5B serviceable, $70M obtainable over five years.
AI-native engineering platforms — 2026 base. $40–100B by 2030 across mid-market and enterprise segments.
US retail & e-commerce mid-market ($50M–$500M revenue) — the addressable wedge we serve at MVP. ~7,500 companies.
Subscription ARR base case at Year 5. ~1.5% of 2031 SAM. Conservative — does not include on-prem deals or services revenue.
| Year | Customers | ACV | Subscription ARR |
|---|---|---|---|
| 2027 | 5 design partners | — | Letters of intent |
| 2028 | 25 | $40K | $1.0M |
| 2029 | 80 | $70K | $5.6M |
| 2030 | 180 | $95K | $17.1M |
| 2031 | 350 | $135K | $47.3M |
A four-cell matrix. We occupy the empty one.
Two axes: who holds the keyboard (developer vs non-coder) and what gets shipped (proprietary runtime vs exportable repo). The lower-right cell — non-coder + exportable repo on real infrastructure — is structurally empty.
Code-level alternatives
- Cursor — $2B ARR / $29.3B
- Aider
- Cody
If the enterprise decides the developer keeps the keyboard.
Developer + exportable repo
- GitHub Copilot — 4.7M seats
- GitHub Spark — React/TS/Azure
- Builder.io Visual Copilot
Developer-facing tools that ship exportable code; none serve a non-coder operator.
Non-coder + proprietary runtime
- Lovable — $200M ARR / $6.6B
- Bolt.new
- v0 (Vercel)
- Replit Agent — $9B / $150M ARR
- Retool · OutSystems · Mendix
- Power Platform · ServiceNow
Consumer/SMB prosumer OR enterprise low-code with proprietary lock-in.
Non-coder + exportable repo · real infra
- CAIA
Read-only adoption + non-coder UI + exportable repos + MCP-first marketplace + on-prem option. Structurally empty cell.
Anthropic’s Cowork is a non-developer desktop tool for general computing tasks. CAIA is a non-coder enterprise platform for shipping production software. Different surface, complementary substrate. We see Cowork as adjacent — Anthropic owns the consumer agent, CAIA owns the enterprise outcome.
Five tiers, plus on-prem.
Free → Team → Business (the wedge) → Enterprise → On-Prem. Land on Business; expand into Enterprise once SOC 2 ships.
Free
Solo operators and tinkerers — non-production use, single seat.
Team
Small teams up to 5 operators. Shared workspace, basic MCP catalog.
Business
Mid-market. Unlimited editors, SSO/SAML, audit logs, dedicated tenancy.
Enterprise
SOC 2 II, custom MCP servers, dedicated success engineer, white-glove onboarding.
On-Prem
Air-gapped deployment. Engine + MCP gateway + success engineer. Source-available.
Year-5 base case: ~$47M subscription ARR + ~$25M on-prem + ~$12M services ≈ $84.5M consolidated revenue. Gross margins trajectory: 62% → 74% as the MCP marketplace and vertical templates reduce per-customer engineering cost.
Six milestones, twelve months, then a ghost future.
MVP — Figma → deployed UI loop on a real customer codebase. First 5 design partners signed.
Marketplace v1. 50 MCP servers. First three paying customers on the Business tier.
Engine v2. Multi-environment promotion (staging → prod). Hires 1 + 2.
SOC 2 Type I in flight. 25 paying customers. $250K–$1M ARR. Series A conversation.
On-prem GA. First Fortune-500 deployment. Marketplace doubles to 100+ servers.
Multi-region. EU data residency. Vertical pre-packaged operator templates.
Vertical operator templates · public cloud marketplace listings · Series A.
One founder, twenty-five years of context, three AI cofounders.
Prakash Tiwari — 25 years of enterprise architecture across financial services, retail, and pharma. Principal Enterprise Architect at Envestnet. CAIA is built by AI agents under his direction. First human hires planned for February 2027.
enterprise architecture · Fortune 50 retail
production code written by agents under direction
CVS Health · AutoZone · Dollar General
Claude Code SDK · MCP-first substrate
Why Anthropic is the right substrate.
CAIA is built on Claude. The Claude Code SDK is our agentic core; Claude is our reasoning substrate. MCP is the integration layer we bet on a year before it had a name.
We are model-portable but model-opinionated. Anthropic owns the consumer agent (Cowork). CAIA owns the enterprise outcome (engine + exportable repo). Complementary.
Anthropic’s focus on safety, interpretability, and tool-use rigor is the substrate the buyer needs — not the substrate the hyperscalers ship. Architecture aligned to values.
$2.5–4M pre-seed. 18 months. Four milestones unlocked.
Use of funds breaks into three categories. Each one moves the same set of milestones forward. The capital is the throttle; the milestones are the proof.
Two senior engineers (Q1 2027) + one design-partner success engineer. Anchored on F50-retail context and Claude Code SDK fluency.
SOC 2 Type I audit, Vanta tooling, penetration test, DPA library. Pre-requisite for the Enterprise tier and any regulated buyer.
Design-partner onboarding for the first 25 customers, marketplace v1 curation, conference presence (QCon, KubeCon, Money 20/20).
paying customers by Q4 2027
ARR by Q4 2027
completed by Q3 2027
in conversation by Q4 2027
Frequently asked.
Is CAIA a low-code platform?
No. Low-code platforms own the runtime. CAIA emits a standard codebase you own and can run anywhere. You can fire us tomorrow and the code keeps running.
How is this different from AI coding assistants like Cursor or Copilot?
Cursor and Copilot help engineers write code faster. CAIA replaces the engineering decision layer so a non-coder operator can deploy production software. Different user. Different job. Different bet.
What is the killer demo?
A Figma file becomes a deployed UI in the customer’s existing codebase, conforming to the customer’s own conventions, in under 10 minutes. We have run this in front of CTOs. The reaction is unmistakable.
Why won’t the big clouds just do this?
They will, eventually. We have an 18-month window during which the right product, the right wedge, and the right community make us hard to displace. The moat triplet is the answer.
How much are you raising?
$2.5–4M pre-seed. Use of funds: two engineering hires, design-partner onboarding, compliance, and the marketplace v1.
Who is the buyer?
US retail and e-commerce mid-market — companies with $50M–$500M revenue, three to twenty engineers, and an obvious backlog of internal apps they cannot ship.
How does this relate to Claude Cowork?
Cowork is a non-developer desktop tool for general computing tasks. CAIA is a non-coder enterprise platform for shipping production software. Different surface, complementary substrate.
What happens if Anthropic deprecates Claude Code SDK?
CAIA is model-portable. The engine consumes any reasoning substrate that supports tool use and long context. Anthropic is the default, not the dependency.
Three ways in.
Email, calendar, LinkedIn. Or the diligence form — gated by an NDA acknowledgement — for investors who want the deck and the data room together.
[email protected]
Investors, design partners, technical leads.
Book 30 minutes
Two-week look-ahead. No pre-read required.
in/prakashtiwari
Background, recommendations, mutual contacts.
Request the data room.
Deck, financial model, design-partner letters, technical architecture, and the v2 spec. Sent within 48 hours of NDA acknowledgement. We do not call without your prior consent.