Not a workflow.
An operator.
A Scribe runs the recurring work behind your company. Campaigns, code, exceptions, reports, research.
Not a workflow.
Runs daily
on its own schedule
Own inbox
own VM, own trail
By the hour
never tokens or seats
Last 28 days, Actual Production Scribes
0
tasks shipped
Spanning ads, code, reports, research, and admin. Delivered to queues, not chat windows.
0
hours active work
Billed only while the Scribe was actually working. No seats, no idle time, no token math.
A Scribe's Week
A queue that makes moves.
Active Task List
Scribe Benedict
Owns paid-growth execution, engineering triage, and weekly reporting end to end.
1 active
Building a cross-channel campaign
Waiting
Creative variants pending your review
Done
Completed work leaves a trail
Open
2
Research top waitlist prospects
Pull company signals, recent hires, funding, and product changes. Write a personalized first-touch draft per account.
Analyze pricing-tier impact on LTV
Query the product database, segment cohorts by tier, plot retention curves, and write a recommendation memo.
In Progress
1
Build out Google Search and Meta campaign
Creative, audiences, budgets, pixel events. Pushed live across both networks with safezone QA before launch.
Working now
Waiting
1
3 Instagram story ad variations, review before push
Scribe generated the creative, ran the safezone box, fixed headline position, and attached transparent-logo PNGs. Paused for approval.
Approval needed
Done
3
Fix error-monitor sampling and broken frontend config
Diagnosed via CLI, patched backend sampling, corrected the public-layout config, opened a PR, merged after checks passed.
Completed
Daily dependency upgrade audit
Reviewed 6 open bump PRs, merged 4 after test runs, escalated 1 with a breaking change for your review.
Completed
Weekly ads deep-analysis report
Pulled 7 days of spend data across Meta, Google, and Reddit, joined to product activation, summarized by campaign, emailed the report.
Completed
08:02
Pulled main branches for both repos, checked overnight deploys, logged two warnings to review later.
08:14
Opened ad accounts across three networks, captured yesterday's spend and conversions, drafted the daily status email.
08:32
Generated a safezone preview for the new creative variants, flagged one for manual review, attached it to the approval task.
08:47
Triaged overnight exceptions, auto-resolved three, escalated one with enough context to decide in under a minute.
Surface
Speaks the systems you run on.
Structured plugins where an API earns its keep. A browser for everything else.
Standing Orders
Shows up for work on its own.
Real briefs Scribes are running today.
Every morning, 5AM
Check active campaigns on Meta, Google, and Reddit. Show me yesterday's spend vs. the day before, flag anything weird, and pause anything CAC has blown out on.
When an email lands from Sentry
Verify it's a real issue in our code. Archive the noise. If it's ours, open a PR with a fix. Email me either way.
Every Monday, 8AM
Pull last week's revenue, churn, and pipeline from Stripe and HubSpot. Write me three paragraphs. Call out the top three blockers.
Every Friday, 2PM
Walk the competitor pricing pages on the list. If anything moved, write a one-pager. If nothing moved, say so and stop.
Every morning, 6AM
Sweep open Dependabot PRs. Read the changelogs, run the tests, merge the safe ones. Email me a list of anything with a major bump or that touches auth.
Every weekday, 4PM
Sweep the support inbox. Draft replies to anything routine and queue them for my approval. Flag the ones that need me with context and a suggested next step.
Priced Like Talent
By the hour, not the token.
Active work time only. Reasoning, browsing, writing, calling APIs. No seats, no token math, no idle fees. A transparent invoice every month, with a trail of what your Scribe actually did.
The Obelisk Model
Start narrow, keep humans on approvals, and expand only after the queue proves itself.
01
Start with one queue.
02
Keep humans on approvals.
03
Expand after trust is earned.
Your Scribe's Inbox
Every Scribe has its own inbox.
Email it from your phone, a meeting, the back of a cab. It picks up the work, does it, and writes back with the result — not a status update, the actual thing you asked for.
Growth
Paid-growth execution
Push campaigns, adjust bids, generate creative with safezone QA, report outcomes mapped to real activation, not just clicks.
Engineering
Exception triage and small fixes
Read the monitor, cluster the noise, auto-resolve what's trivial, escalate what's real, ship small fixes via code-agent CLIs on the VM.
Operations
The recurring admin queue
Inboxes, reconciliations, vendor requests, follow-ups that go cold. The work that sits between teams and never gets owned.
Research
Analysis that answers a real question
Query the product database, pull from the ad platforms, synthesize a memo. Not a chatbot answer, a written artifact you can forward.
Latest
Busy work begone.
