Venture · United States

AI engineering for US companies, from demo to production.

DVNC Dev is the US-facing engineering venture. It does one thing well: take AI that works in a demo and make it survive real users, real data, and the security review. Claude Code infrastructure, production agents, and the control layer that keeps an AI stack visible and under budget.

AI EngineeringAgentsInfrastructure
Cover image
Overview

An engineering property for the part nobody finishes.

Most teams can get an AI prototype working. Far fewer can get it deployed, instrumented, and trusted by the people who have to run it. DVNC Dev is built around that gap.

The site is a publication and a service surface at once: it ships engineering writing on Claude Code, agents, and production patterns, and routes serious work into fixed-scope builds.

At a glance
Role
Founder, AI engineer
Market
United States
Year
2026
Status
Building
The challenge

Demos are easy. Production is the work.

The hard part of AI is never the model. It is the permissions, the logging, the evals, the cost control, and the review the system has to pass before it goes live. DVNC Dev exists to own exactly that distance, for teams that have a prototype and need a system.

Where it hurt
Prototypes that fall apart on real dataAgents with no logs, no permissions, no known blast radiusAI cost that runs away the moment usage scalesStalls at the security and compliance review
The approach

What I built, and how it came together.

The build

A publication and a build practice on one Cloudflare-backed stack.

Next.js on Vercel for the reader site, a Cloudflare Worker for the AI and data layer, and an editorial pipeline that publishes without a human in the deploy loop.

What it includes
Reader site with the full design system and article rendererCloudflare Worker: D1, R2, Vectorize, AI GatewayMarkdown-canonical publishing pipelineCost control and logging on every paid AI callService routing into fixed-scope engineering builds
How it was built

A small number of moves, each one shipped.

Each stage shipped its own deliverables, so the work stayed visible and correctable before the next one started.

01

Stand up the stack

Frontend on Vercel, Worker on Cloudflare, strict HTTP boundary between them.

Next.js siteCloudflare WorkerBoundary lock
02

Wire the AI layer

Embeddings, retrieval, and cover generation run through one metered chokepoint.

AI GatewayVectorizeCost cap
03

Automate publishing

A claude.ai routine researches, writes, and posts to the publish endpoint.

Publish APIEditorial routineRevalidation
04

Route the work

The site sends production-grade engineering work into scoped builds.

Service pagesLead captureHand-off
The work

A closer look at what shipped.

Homepage
Homepage
An engineering article
An engineering article
The publishing pipeline
The publishing pipeline
Service routing
Service routing
Mobile reading view
Mobile reading view
The control layer
The control layer
The result

Built on the same patterns it sells.

DVNC Dev runs the exact architecture it recommends to clients: a thin Vercel frontend, a Cloudflare Worker for everything stateful, and a cost-capped AI layer. The venture is its own reference implementation.

By the numbers

2

deploy targets, one repo

By the numbers

$5

daily AI cost ceiling, enforced

By the numbers

Owned

infrastructure, end to end

The stack

Built with

Next.jsReactTailwind CSSCloudflare WorkersD1R2VectorizeClaude
Have something to build?

Start with a short call. Straight answer either way.

Tell me what you are trying to ship. We scope it, price it, and decide the right way in.

Newsletter

One letter, every Sunday. Working systems, not hot takes.

Build logs, working systems, and field notes from running a portfolio of AI ventures. Sent weekly, never more.

Weekly. No spam. Unsubscribe anytime.