Agents that
clock in every night.
Report bots, chief-of-staff agents, outreach crews and SEO writers. Built for real companies, running on plain Linux servers, talking over Telegram, WhatsApp and Slack.
// n agents · one orchestrator · zero dashboards
Same shape,
every time.
The client gets a small VPS. On it runs an agent runtime, OpenClaw or Hermes, wired to whatever the work touches: a mailbox, a CRM, an ad account, a task tracker. You talk to it from the chat app your team already uses.
Their data goes through their account, billed to them, visible to them. I never sit in the middle of it.
Twelve stores,
one email.
Furtados runs 12 stores across Mumbai. Each one emails a daily accounting report to a central mailbox, and somebody had to open all 12 every evening, cross-check the figures by eye and decide what needed attention before close of day.
A bot on their own server pulls the day's reports over IMAP, parses whatever format each store sent (Excel, CSV, PDF, the formats drift), merges everything into one master sheet and checks every line against rules the team can edit in a plain config file.
Cash in the drawer doesn't match the POS total.
A store hasn't submitted by cutoff. An early alert fires so someone can chase them before the nightly run.
Revenue far below that store's own 7-day average, no holiday on record.
Refund ratio creeping up, an expense line jumping, totals that don't add up.
$ run daily_intel --date today ✓ mailbox check .......... 12/12 reports in ✓ parse + normalise ...... xlsx ×9 · csv ×2 · pdf ×1 ✓ consolidate ............ master sheet built ⚠ flags .................. 2 WARNING · 1 CRITICAL › store 07 · cash vs POS mismatch CRITICAL › store 03 · refund ratio above limit WARNING › store 11 · petty cash over cap WARNING ✓ report sent ............ 1 email · run took 41s
// illustrative run. real store data stays on the client's server.
One email in the manager's inbox. Flag count in the subject line, and a recap of whether yesterday's criticals actually got resolved.
A chief of staff
that lives in Slack.
Pocket Fund is a micro private equity firm I work with. Their setup has grown into a small crew of agents, each with exactly one job.
Meetings → tasks
Transcripts get dropped into Slack. The agents pull out decisions and action items and suggest tasks. Nothing is filed until a human taps approve.
transcript → classify → suggest → approve → trackerSearch-led writing
Research with Google Trends and keyword data, topics people actually search for, drafts pushed to Beehiiv. Nobody publishes what a human hasn't read.
trends → topics → draft → review → publishCampaigns from chat
Send the creative over Telegram, a video or a few images, tell the bot the budget and audience, and the campaign gets built.
creative → brief → campaign → live- follow up with the Canada prospect (analyst)
- fix the DD checklist template (finance ops)
- move the Reddit report to Fridays (research)
An org chart
made of agents.
Kautilya runs an ambassador programme for chartered accountants. That means outreach to thousands of CAs across India. Too many for a person, too nuanced for a mail-merge. So the org chart is made of agents, run by an orchestrator named Chanakya. Chank, to the team.
A management hierarchy, except the middle managers cost a few rupees a day and never forget to follow up.
Personal bots
& odd jobs.
Not everything is a company system. Jade is a research-and-outreach agent in the works for store events: it shortlists artists and authors worth contacting, then works LinkedIn and email. There are also personal assistant bots running calendars and notes over Telegram, and SEO bots turning keyword movements into content briefs.
Got a chore worth killing?
Tell me what eats your team's evenings. There's probably a bot for that, and if there isn't, I'll build it.
Start a conversation ↗← back to all work