Give it a goal.
Get a funnel.
A job is a persistent goal your agent pursues for days or weeks — not a task it does once. Outreach, launches, playbook builds, analysis, or hybrids. It researches the targets, runs the outreach, answers the replies, books the meetings, and shows up to them. You approve the plan once and watch the funnel move.
The right plan shape for the goal
Every job gets a kind — inferred from your goal, or pinned by you. Outreach stays a single brief; bigger goals become a graph of workstreams that can run in parallel.
Outreach
Book demos and jobs with an ICP. Single brief: research → outreach → reply → book → meet.
Campaign
Launch or announce. Graph: research → content → outreach — coordinated workstreams.
Implement
Automate a playbook or build a pipeline. Graph: research → design → build → wire → verify.
Analyze
Performance, audit, diagnose. Graph: research → analyze → report — judgment you can act on.
Hybrid
Build and book in parallel — implement the system while outreach fills the calendar.
From sentence to booked meetings
The entire interface is a goal you'd say out loud and a brief (or graph) you approve. Everything else is the agent's job.
State the goal
In plain English on the marketing site, in Ask, or in Slack: “Book 10 demos with HVAC companies in Texas by end of month, budget $60.” Or paste a playbook URL — that’s the whole input.
Approve the brief (or graph)
The agent drafts a job brief — or a multi-workstream graph for launches, builds, and analysis. Targets, channels, sequence, budget, autonomy. You approve once — or edit first.
The funnel runs
Research finds and verifies targets. Outreach goes out on the lanes you approved. Replies get answered, follow-ups get scheduled, meetings get booked onto your calendar.
It reports and adapts
You see the live funnel — researched, contacted, replied, booked — and the agent re-plans as results come in. Every brief revision is versioned; every touch is logged.
Book is a first-class step
Jobs don't stop at “interested.” The Scheduling Service lists real open slots and books them on your Waveguide book page or Calendly — respecting SOUL windows, length, and buffer. Prefer your own URL? External-link mode shares it only. Template SDR: voice → book → Meet runs the full path: outbound call, book from the transcript, join the Meet.
- One service — same tools whether the prospect uses your page or Calendly.
- Ledger sync — booked and canceled invitees become leads and updates.
- You can take the Meet — or let the agent attend and post the transcript.
30-minute estimate · buffer respected · times from your open calendar.
Slot from SOUL windows · Meet link on calendar · confirmation text queued · lead in ledger.
- ✓ Waveguide booker — your slug, your hours
- ✓ Calendly — same agent tools, their UI
- ✓ SDR path — voice → book → Meet, one job
You choose the autonomy
Four levels, per job. Start cautious, loosen as trust builds. Whatever you pick, financial and irreversible actions always require a human — that's enforced by architecture, not policy.
Per-message
You approve every single outbound message before it goes. Maximum control, useful for the first week while you calibrate the agent's voice.
Batch
The agent prepares a batch of messages and you approve them together — one review a day instead of twenty interruptions.
Brief
You approve the job brief and standing rules; the agent sends within them without asking per message. The default for most jobs.
Autopilot
The agent runs the job end to end within its caps. Even here, tier-3 actions — anything financial or irreversible — still require a human. That rule has no override.
Every lane a real business uses
A job isn't an email blast. It's coordinated, paced outreach across the channels your prospects actually answer on — with the channel-specific compliance handled for you.
Hosted Waveguide sending or your Gmail. Warming, SPF/DKIM/DMARC, reply detection, and automatic suppression. Replies route back to the agent — and to you.
Social
Posts and DMs through your connected social accounts, with per-platform pacing. The agent reads comments and replies too, so a conversation started in public continues properly.
Voice
Outbound AI phone calls with consent gates and AI disclosure built in — the agent identifies itself as an AI, jurisdiction rules are enforced before dialing, and transcripts land in the job log.
Book + Google Meet
Scheduling Service books into your Waveguide page or Calendly (external-link mode shares a URL only). When a meeting lands, the agent can attend the Meet, hold the conversation, and post the transcript back.
Lanes activate per job and per plan; connect the accounts and tools you want the agent to use on the integrations page.
It doesn't stop at the inbox
When a prospect books, the job keeps going. The agent puts the meeting on your calendar with a Meet link, attends it as a voice participant if you want it to, and posts the transcript and outcomes back — so the follow-up email that goes out an hour later actually reflects the conversation.
- Always disclosed — the agent identifies itself as an AI on calls and in meetings.
- Always logged — transcripts and outcomes land in the job's audit trail.
- Always yours — take any meeting yourself; the agent preps you with context instead.
- Tue 2:14 PM
Outbound call to Rivera Mechanical — consent verified, AI disclosed. Outcome: interested, asked for Thursday. Slot booked via Scheduling Service. - Thu 10:00 AM
Joined Meet with Rivera Mechanical (28 min). Discussed maintenance contracts; they want the premium tier pricing. Transcript attached. Follow-up drafted. - Thu 11:02 AM
Follow-up email queued for your approval — references the premium tier discussion. Awaiting you.
Outreach you won't have to apologize for
Autonomous outreach is only useful if it can't embarrass you. The rails are built in, on every job, on every lane.
Budgeted per job
Every job has its own budget. At 100% it soft-lands — the agent economizes and warns you. At 120% it stops entirely and waits for you. Your worst-case month is bounded and you chose it.
Suppression and consent
A tenant-wide suppression list is enforced on every lane. Voice calls check consent and jurisdiction rules before dialing. Unsubscribes and opt-outs apply everywhere, instantly.
Frequency governance
A hard cross-job cap means two concurrent jobs can never over-contact the same person — at most a handful of touches per contact per week, across all channels, no matter what.
Dry-run and practice modes
Run a job in dry-run to see exactly what it would send, to whom, and when — without a single message leaving. Practice mode lets you rehearse the agent's voice before anything is real.
Waveguide
Your next campaign is one sentence long
State the goal tonight. Review the job brief with your coffee tomorrow.
$5 free AI credits · no card required · upgrade anytime