asked on Dear SaaStr
What should I know before rolling out our first AI SDR?
That it is not set-and-forget. Read every message it writes for the first month, keep a human approving anything that leaves, and expect to spend real time managing it. Agents left alone have invented figures, made promises nobody approved and emailed people about events that had already happened.
By The Marketinque TeamReviewed September 2026
What did people in the thread say?
- SaaStr’s own answer: read every message for the first 30 days; set-and-forget is false. source
- In one week, SaaStr’s marketing agents ran an A/B test nobody authorised that gave away free tickets and invited people to an event that had already happened; Jason Lemkin says agents take about as much managing as people. source
- A founder who ran an agent as an autonomous employee found it made up revenue figures and got product details wrong in its marketing copy. source
- On approval fatigue: you read every request at first, never see one you would deny, then stop reading and just click approve. source
What would we do?
- Start every action on approval, and read the claims and the numbers, not the adjectives.
- Grant autonomy per action, not all at once, and only after a streak of approvals you actually read.
- Only let it act alone where a mistake can be undone or fades; keep spending and cold outreach on approval for good.
- Watch bounce and complaint rates, and take autonomy back the day they spike.
Where does marketinque help, and where not?
marketinque works this way: every action starts on Ask me, an action can go out after a veto window you can cancel, and runs on its own only after five clean approvals in a row, only where marketinque sends it itself. A bounce or complaint spike sends email back to asking.
Where it does not: Approval fatigue is real and marketinque does not solve it yet: nothing notices when you approve without reading.
We build marketinque, so weigh this section accordingly. It is free while projects open by waitlist.