Pick the routine. I'll make it run itself.
The automations I build most often, in plain language: what each one does, who it helps, and roughly what it saves. Every one ships with a human in the loop and an off-switch you own.
the 6h is an estimate, not an audited client number
Lead intake & triage
Every enquiry checked, summarised, scored and routed in seconds. A person only writes the reply – to a brief, not a raw form.
Authored estimate of the manual checking it replaces; the watcher itself is real and has run since June 2026.
Market & filings watcher
A cloud watcher reads prices, news and regulatory filings on a schedule, runs them through per-entity rules, and sends you only what clears the bar – with a plain-language summary. It is two-way: you answer it, command it and query it from the same chat. Mine has run since June 2026: its first autonomous run sent 17 alerts with 0 errors.
Authored estimate – roughly the after-meeting typing and chasing it replaces per week.
Meetings to decisions & actions
Drop in a transcript or recording doc. Out comes a structured debrief: decisions made, actions with owner and date, blockers, and drafted follow-ups. Every write is confirmed by you before it lands anywhere.
Authored: before automating, four weeks of this report were reconstructed by hand from 583 commits.
Weekly report, written from the work itself
A collector reads what actually happened in your systems that week; the AI drafts the report section in your house style; you read the preview and it publishes only on your OK – append-only, without touching anyone else's content.
Authored estimate of the research-plus-writing hours per batch of cards it replaces.
Content pipeline where every fact has a source
Photos and files go in; publish-ready catalogue pages come out. The rule is absolute and lives in the data layer: a fact is sourced or it does not exist. Unsourced cards physically cannot render. Mine publishes 50 catalogue cards in production, every one of them built by this pipeline.
Authored: manual browser-testing after AI builds was my own #1 recurring time sink before this existed.
AI ships the code – the pipeline proves it works
After an AI (or anyone) builds a feature, an agent drives the real product in a real browser through the affected flows, checks the database actually changed, and reports with evidence. The strongest version closes the loop: finding, failing test, fix, gates, report – then stops for a human to merge. Mine gates this very site: as of August 2026 a 1,107-test end-to-end battery, 45 visual baselines and a 1.00 Lighthouse score.
Authored estimate – the daily repo-walking it replaced across ~24 projects.
Morning digest of everything that moved
At 08:00 a zero-AI collector walks every repository and diffs against the last acknowledged state. You get one message – only if there is something to read. The AI helps triage when you sit down, not before.
Before selling automations, I built an operating system around myself. Everything on this page is an excerpt from it.
Read the case behind themYour routine isn't on the list? It probably still qualifies – the catalogue is what's most common, not what's possible. Tell me what your team repeats.
Ask about your processNot sure which one pays off first?
That's what the consultation is for – from €290, you leave with a ranked plan.