(Ada · Ext. 107 · knowledge base)

Developer.

Ships fixed-scope sprints — sites, integrations, internal tools, automations — with every line human-reviewed before it touches production.

Launch this role

What it does from day one

(no ramp-up, no aspiration)

How the work flows

(brief to logged output)

The stack it plugs into

(real tools, official APIs)
  • GitHub (API + gh CLI + Actions) — the system of record for every deliverable

  • Vercel/Netlify API — deploy previews and production hosting

  • Linear or Jira API — ticket intake and a sprint board the client can watch

  • Slack API — task assignment, progress updates, escalations (the proven enterprise interface for coding agents)

  • n8n (self-hosted) + Zapier/Make — the deliverable format for client automations

  • Supabase/Postgres + Stripe + HubSpot APIs — the recurring backend trio in client asks

  • Sentry + uptime monitor — error and downtime signals that auto-generate maintenance tasks

Its machine

(8GB RAM · 4 vCores, dedicated)

Git, Node/pnpm and Python toolchains, a coding-agent harness, one headless Chromium for visual verification, local dev servers, Docker for self-hosted n8n, and CLI deploy tools. 8GB works because work is serialized — one Next.js build (1-2GB peak) plus one browser fits, but never run a client's production DB or app on this box, and pin the lightweight Chromium build.

Seat economics

heavy token load

The heaviest non-voice role: whole-repo reads and long edit-build-test loops plausibly burn $150-400/mo at cached rates — viable but the tightest margin in the lineup, so fair use meters sprints or active agent-hours per month, not 'unlimited coding.'

Flat seat from $499/agent/mo — one agent, one dedicated machine. See pricing.

Judged by

(the numbers that matter)
  • First-pass acceptance rate — % of deliverables accepted without rework (vendor norm: rework <15% of dev hours)

  • On-time delivery for fixed-scope sprints (agency benchmark ~98%; DORA top-15% lead time <1 day)

  • Change failure rate / post-ship defects — DORA top 15% is <4%; AI code averages 1.7x more major issues than human code, which is exactly why the human-review layer is the pitch

Tasks in this role

(each documented in the task library)

Never alone

(guardrails — how trust works here)

Full policy: our AI disclosure.

Try it

(alpha)

"Watch your developer build it — live": visitor types a one-line brief ('landing page for a Miami dental clinic with booking form') and watches a split view — narrated plan and live diff/terminal on the left, hot-reloading preview on the right — finishing with a shareable URL in 2-3 minutes. Runs in StackBlitz WebContainers so Node executes entirely in the visitor's browser sandbox: no server-side execution risk, instant boot, constrained template. Cold calling ships first — try that demo today.