Skip to content
The Receipts

What Downshift spends so the founders we pick don't have to.

The Public Build campaign claims Downshift already runs heavy AI engineering infrastructure. This page is the proof. Real numbers. Itemized. Updated monthly. Last updated .

Section 01

Monthly AI infrastructure spend

Itemized by service. Real invoices for April 2026.

Total this month
$3,847.62
Service April 2026
Anthropic (Claude API)
AI inference
Primary engineering model. Sonnet 4.6 + Opus 4.7 across all client builds.
$1,847.32
Cursor
Coding environment
8 seats at $20/mo. Tab completions, Composer, agent mode.
$160.00
FairStack
Generative media (image, voice, video, music)
Marketing assets, demo voiceovers, social cuts for client launches.
$412.18
OpenRouter
Cross-model evals
Used for adversarial reviews and model-as-judge evaluation runs.
$98.44
Vercel
Hosting (web)
Pro team plan. Hosts marketing sites, Next.js client builds, preview environments.
$240.00
Railway
Hosting (apps + databases)
Production app workloads, Postgres, Redis, worker services for active client products.
$612.84
Sentry
Error tracking
Team plan across the active client portfolio.
$104.00
Resend
Transactional email
Pro plan. Auth, receipts, app notifications across products.
$60.00
Cloudflare
DNS + CDN + R2 storage
DNS for every property, R2 for asset storage, Pages for an internal portal.
$72.50
PostHog
Product analytics
Self-hosted-on-cloud product analytics for every active build.
$112.00
GitHub
Code hosting + CI
Team plan. Private repos for every client build. Actions minutes for CI.
$84.00
Linear
Project tracking
Engineering and product tracking for active client work.
$44.34
Section 02

Active public builds in flight

Founders with permission to publish. Week-by-week status. What shipped this week.

No active public builds yet

Applications are open. The first cohort starts as soon as the right founders apply. Until then, the receipts above are the proof we already run the infrastructure.

Apply to be built publicly
Section 03

Hours of AI-assisted engineering shipped this month

412

Counted from agent session telemetry across the active client portfolio. Measures hours where an engineer was actively pairing with an AI agent on shipped code, not idle or planning time.

Section 04

Lines of code shipped this month

38,418

Net additions across all client repos, measured by git log on main branches in April 2026. Excludes lockfiles, generated assets, and reverts.

Section 05

Last updated

Numbers come from invoices, agent telemetry, and git log. The data file is at src/_data/transparency.json and is updated monthly.