3 · Automate with approval

Routine work, taken off their plate. Only when they say yes.

We do not sell you another AI agent. We build the automation on the AI you already pay for, with the person’s own permissions.

Their permissions, nothing moreAsks before every changeEvery action logged

Inside Claude Code, before a change is saved
┌ 20xwork ─────────────────────────────┐
│ Before this is saved                 │
│ missing   a date                     │
│ try       "… by Tue Oct 6"           │
│ [keep]  [adapt]  [dismiss]           │
└──────────────────────────────────────┘
Want this done for you every week?

Every Monday, draft the follow-up for each deal that moved last week, in your words, for you to review.

Uses your own Salesforce access and your Claude seatAsks you before every changeKeeps a log of what it did
ApproveNot now

How it works

In three steps.

Offer

Want this done for you?

The suggestion offers to take the task on as a weekly automation. Nothing is built unless the person says yes.

Build

On the AI you already own

Claude, Copilot or Agentforce, whichever you run. It gets the person’s access and nothing more.

Run

Asks before every change

It drafts, the person approves. Every action is logged against the suggestion that started it.

Guardrails

It cannot do anything the person cannot.

Same permissions, never wider. A human approval before every change. A full log of what it did.

  • Same permissions as the person, never more
  • Approval before every change
  • A log entry for every action
Automation · guardrails
Runs asthe person, with their Salesforce role
Can changeonly what they can
Before each changeasks, and waits
Loggedthe suggestion, the decision, the result
Built onClaude, Copilot or Agentforce
One morning, as it happened (example)
09:04SalesforceOpens the deal. Stage moved to Negotiation.
09:11ClaudeDrafts the follow-up.
09:18SalesforceSaves next step: “Send revised quote” saved · incomplete
09:1820xworkSuggests adding a date.
09:21SalesforceSaves: “… by Tue Oct 6” saved · complete
09:40OutlookQuote sent. Step closed.
The same task across the team23 people · 61% complete on the first try · 3h 10m a week lost to redoing it

It keeps learning

Every result makes the next suggestion better.

Used, changed or ignored, each result goes back into the picture, so the next suggestion already knows what this person does.

  • Used, changed or ignored
  • Added to the person’s own history
  • Counted in the measured return

Privacy and security

We look at the work, not the person.

Access your admin grants and can take away. No screen recording, no keystrokes, no private messages.

  • Each employee sees their own results first
  • We keep summaries, never documents or messages
  • For coaching, never the sole basis for discipline
Read the Trust Centre →
What 20xwork can see (admin view)
What we see
  • Work saved in the tools your admin connects
  • What people asked the AI, if your organisation shares it
  • Which tasks happened, in what order, and when
  • How the work is supposed to be done
Never
  • ✕Screens or screenshots
  • ✕Keystrokes, clipboard, webcam
  • ✕Personal accounts or devices
  • ✕Private messages or the contents of documents
Read-only accessFull audit logSSOGDPRSOC 2 Type II in progress

Getting started

A data export on day one. A number by week four.

Day 1

A data export

No integration needed. We show you what we can already see.

Week 1

One team connected

Read-only access for one team. The first suggestions appear in their tools.

Week 4

The number

Hours back per employee, with a before, an after and a control group.

Bring one team. We will show you what AI changed for each of them.

One team, one data export, a number in four weeks.

Book a working session