AI automations and MVPs.
Shipped, not pitched.

ArwenDigital is a small engineering shop. We automate the busywork and build the first version of your product. You review it. We ship it.

Start a project or email directly
$ mail arnold@arwen.digital

Send a paragraph. Get back a scope.

Ask Arwen to build something. Try one:
arwen — invoice-intake example session
you

Invoices come in by email. Get them into accounting without anyone retyping them.

●

Plan · 4 steps

1 watch the inbox for PDF invoices

2 extract vendor, amount, due date

3 match each one against open POs

4 draft the bill, a person approves

›

Write flows/invoice-intake.ts

✓

Sample invoices parsed, totals match

◆

1 invoice has no PO. Approve or reject?

●

Live. Checks the inbox every 5 minutes.

you

Booking app for my clinic. Patients pick a slot, staff confirm it.

●

Scope · version one

+

booking page, staff inbox, SMS reminder

−

payments, mobile app, reports → later

›

Create app/ · api/ · db/schema.sql

✓

Booking flow tested end to end

›

Deploy staging

◆

Try it on your phone. Ship it?

●

Shipped. Real patients are booking.

you

Our contractor left. The app half works. Nobody can deploy it.

›

Read repo · 6 years of history

●

Found: no tests, a broken deploy, dead features

−

3 features nobody uses

+

tests on the checkout path

+

one-command deploy + runbook

✓

Deploy green on the first try

◆

Keep or drop the old admin panel?

●

Stable. Your team can run it now.

you

Help our support team answer tickets faster. Nothing goes out unchecked.

●

Plan · draft, never auto-send

read ticket + account + help docs

draft a reply with sources attached

›

Write api/draft-reply.ts

✓

Checked against past tickets, drafts on topic

◆

Agent reviews draft → clicks Send

●

Live in the support inbox.

❯

Most agencies sell design. We sell software that runs. Automations that clear the inbox. MVPs real users log into. Code your next hire can read.

Two jobs.
Done properly.

AI does the grunt work: drafting, coding, testing. A senior engineer decides what ships. That's the whole model.

Automate the busywork

arwen automate --inbox --docs --crm

AI agents for intake, triage, data entry and reporting. Every step logged, anything risky waits for a person.

AI automation →

Build the first version

arwen build --mvp --scope v1

The smallest product that proves the idea. Real login, real data, real users, working software you can click.

MVP development →

Fix what's stuck

arwen rescue --repo ./legacy

Half-built app or inherited code. We read it, cut it down and get the important part deployed.

## Scope before code

Every project starts with one page: who it's for, the one job it does, and what we're not building yet.

v1 · build now
  • Sign up, log in
  • The one core workflow
  • Admin view
  • Email notifications
later · on purpose
  • Payments
  • Mobile app
  • Analytics dashboard
  • Integrations #2–#5

## Build in the open

You get a working link early, then a new increment every week. No big reveal at the end.

❯ arwen log --project your-app
wk 1scope signeddone
wk 2clickable prototype on stagingdone
wk 3core workflow, real datadone
wk 4your feedback → fixesworking
wk 5launchnext

Example timeline. Real timelines get agreed per project.

## Humans sign off

AI writes drafts and code fast. People approve anything that touches money, customers or production.

Review queue 3
needs you

Refund over limit · support-drafts

EditApprove
needs you

Merge: checkout tests · clinic-booking

DiffApprove
auto

Invoice matched PO · invoice-intake

Logged

## Hand it over

Readable code, a README and a runbook. Your team takes over, or we stay on for maintenance.

your-app/├── README.md        # how to run it
├── RUNBOOK.md       # what to do when it breaks
├── DECISIONS.md     # what we cut and why
├── .env.example
├── src/
└── tests/

## Shipped

Products we've built. Client work is in the case studies.

## Questions

What does it cost?

Priced per scoped phase. You see the number before we start building.

What stack do you use?

Whatever fits and your team can maintain. Usually TypeScript, Python, Postgres, and the major LLM APIs. WordPress when it’s the right call.

Is AI writing all the code?

AI speeds up drafting, coding and testing. An engineer reviews every change and owns every decision.

Who owns the code?

Ownership and licensing are agreed in writing per engagement, before work starts.

I’m not technical. Is that okay?

Yes. That’s most of our clients. A paragraph describing the problem is enough to start.

Got a paragraph?
That's enough to start.

Describe the task or product. You get a straight answer on what to automate or build first.

arnold@arwen.digital