Work a list — email, message, follow up.
Try“Email the trial users who never activated, then follow up in 3 days”
MISSIONS
Say it once. It finishes.
Say “chase the invoices more than 30 days overdue” and it becomes a workflow that runs on a schedule or the moment something happens — working a list person by person, waiting for replies, and stopping to ask when a judgement is yours to make. The follow-up nobody has time for is the work it does best.
Scheduled or event-triggeredRemembers every contactWaits for repliesSafety caps on by defaultRehearse before it goes live
WHY AUTOMATIONS BREAK
Work isn’t a zap.
Automation tools move data from A to B until the data looks different, then fail silently. Chat assistants do one task and forget. Neither can email a list on Monday, wait for replies, call the people who didn’t answer on Wednesday, and tell you honestly on Friday who said yes.
SIX WAYS TO START
Six shapes, your words.
You are not filling in a workflow builder. Pick the shape of the job, describe it the way you would to a new starter, and the scoping happens in the conversation.
Keep an eye on something; stay silent until it moves.
Try“Tell me when a deal over $10k has had no activity for a week”A digest of what changed, on a schedule.
Try“Every morning, email me any task in my tracker that is overdue”Fill the gaps, dedupe, tag what’s untagged.
Try“Find company size and industry for every row in my leads sheet”Keep records flowing from one tool to another.
Try“Add every new form response to my CRM as a lead”A site, an app, a landing page — hands off to Build.
Try“An internal dashboard showing this week’s signups”SCOPE → CHECK → ARM → RUN → REPORT
It scopes before it runs.
A competent new employee asks what they are allowed to do before they start. This does the same: it reads your workspace, shows you samples, and will not arm itself until every step is bound to something real and you have granted the rights one by one.
01
Scope in a conversation
Describe the goal. The scoping agent looks inside your workspace first — which apps are connected, which Slack channels, sheets and calendars actually exist — and asks real questions with real options to click.
02
See samples first
Before asking for any access, it shows what the mission will produce — a couple of draft messages or images — while the only thing at stake is a re-word.
03
Watch the workflow take shape
The workflow draws itself as boxes: gather → transform → decide → act → deliver. It stays a draft until the agent is confident it can be built end to end.
04
Go green
Every box turns green only when its tools are connected and every resource is bound to a real sheet, channel or list. A mission can’t be armed until all of them are.
05
Arm with an itemized consent slip
You grant specific rights — “Send emails”, “Post to Slack”, “Change ad campaigns” — each with its risk spelled out. An ungranted channel cannot fire.
06
Run and report
It runs on its schedule or trigger, records every step, and delivers a report checked against what actually happened.
THE CONTACT LEDGER
One row per person.
Most outreach tools treat a list as a blob. Missions keep a ledger: who each person is, what happened when they were contacted, and what’s next. That’s what makes “day 7, then day 30” possible, stops the same person being contacted twice, and lets the report say who replied instead of “the campaign performed well”.
- One row per person, not per campaign
- What happened when they were contacted
- What is scheduled next
- Counted across every mission in the workspace
- No person touched more than 3 times in 7 days
LONG-RUNNING BY DESIGN
Send. Wait. Act.
Most automation ends the moment it has sent something. A mission is written to survive the gap between sending and hearing back, which is where the actual work of following up lives.
- It waits for the replysend an email or a message, then pause until someone answers before deciding what happens next
- An approval is the same kind of waitwhen a step needs a person, the mission stops and holds rather than guessing and carrying on
- Days, not secondsa sequence with a three-day follow-up is one mission, not three you have to remember to run
- It knows what it already didthe contact ledger travels with it, so a reply lands against the right person and the right history
ON A CLOCK, OR ON AN EVENT
Schedules that mean it.
“Every morning at 10” is the kind of instruction most schedulers quietly reinterpret. This one runs at 10 where you are, on the days you meant, and does not replay a week of missed runs the moment it comes back.
- Once, recurring or on an eventrun now, on a schedule, or instantly when a webhook fires (“a demo no-show”, “an abandoned cart”)
- Your clock, not UTC“every morning at 10” runs at 10 in your timezone, through daylight saving and half-hour zones
- Several times a day“post at 11am and 7pm” is one mission with two firing times
- Specific daysweekdays, Mondays, the 1st, or the last day of the month
- Sensible when things overlapif the last run is still going, the next one waits rather than doubling up
- No flood after downtimemissed windows are skipped by default, not replayed as a burst of duplicate work
THE TRUST LAYER
Rehearse before it’s real.
Nothing has to be armed to be tried. Every mission also carries these caps from its first run, and if a count cannot be read it stops rather than continuing — an unknown number is never treated as zero. They are volume and cost limits, not a compliance layer: quiet hours, do-not-disturb and suppression lists are not built yet, so regulated outreach is worth scoping with us first.
50 outreach per run200 per day$10 per run3 touches per person in 7 days
- Mockrun the whole mission; nothing is sent
- Redirectevery message comes to you instead of the recipient
- One contact firstrun it for real on a single person before the whole list
- Auto or AskAuto acts on its own within the rights you granted; Ask holds every outreach for your review. “Chase my pipeline” starts in Ask
NO “MISSION COMPLETED SUCCESSFULLY”
The report is checked.
Before a report reaches you it is held against the contact ledger. A summary claiming it reached forty leads when the ledger shows twelve is rewritten from the real numbers, and an honest empty result is allowed to stay empty.
- Checked against the ledgerevery number in the report has to match what was actually recorded, person by person
- Nothing pending is called donean action that did not run is reported as not run, never as “queued” or “saved as a draft”
- Zero is a real answera run that found nothing says so, instead of padding the report to look busy
- The evidence comes with itper-contact outcomes, the steps taken and what the run cost
WHAT’S UNDER THE HOOD
Code enforces the rules.
A prompt asking an agent to behave is a hope. Every rule on this page is checked outside the model, where no amount of clever wording can talk its way past it.
- Grounded in your real toolsthe builder looks up each app’s actual actions and inputs before writing a step, and checks its work until it compiles
- A Rule Keeperevery mission spec passes a deterministic rule check, and every defect ever found becomes a new rule
- Rewritten when things changerename a column, swap a connector or edit the goal, and the workflow is regenerated instead of running against stale assumptions
- Judgement where it’s neededa read-only in-run agent handles the unexpected (an unfamiliar sheet, “does this reply count as interest?”) without being able to message anyone
- Locked-down executionmission code runs in a separate process with no secrets, no file system and no network of its own
- Knows your companymissions can search the web, recall your Company Brain, and use your connected apps
CAPABILITIES
Read. Think. Act. Report.
What a mission can reach for, in one list. Everything here runs inside the rights you granted and the caps above.
- Read your apps
- Pull from sheets, CRMs, trackers, inboxes and calendars through your connected accounts.
- Search the web
- Look things up with sources.
- Recall company truth
- Ask Company Brain what your company already knows.
- Write and decide
- Draft personalised messages; make a bounded judgement call on what came back.
- Reach people
- Email and Slack within granted rights and caps — plus WhatsApp and SMS once your workspace has a connected number.
- Wait for replies
- Pause until someone answers, then continue.
- Act in your tools
- Update the CRM, edit a sheet, file a tracker item — through the same approval rules.
- Ask a human
- Hold an action for approval in Monitor and on the mission page.
- Deliver
- Show results in the app, email a report, or run as outreach.
SAFE BY DESIGN
Every run on the record.
- Mission approvals appear in Monitor — the one place your team acts — and on the mission’s own page
- Run history with per-contact outcomes, step-by-step evidence and cost
- An ungranted channel cannot fire, whatever the workflow asks for
- The conversation that created the mission travels with it
- If a count can’t be read, the mission stops rather than guessing
START WITH ONE LIST
Follow-ups, done.
Pick the list nobody has got to. Describe what should happen to it, rehearse the whole thing on yourself, and arm it when it looks right.
Start a mission
Rehearsed on you first ✓
Rights granted ✓
Reported from the ledger ✓
Show us the work you want to delegate.
Book a tailored demo for AI Crew, Voice, Build or Missions. Track is available free for 30 days.