Quick answer: Autopilot sends without a person; copilot drafts and waits. Read at source on 24 September 2026, five of six AI SDR vendors let a person hold the send step. AiSDR does not: you approve sample messages before launch. Only 11x and Artisan document a booking-specific hold, and only Artisan states its send setting can change after a campaign is live.
Most comparisons of autopilot and copilot argue about whether autonomy is a good idea. A buyer is not choosing a philosophy but a product whose defaults may not change, so the useful question is narrower: at which step of the outbound cycle can I insert a person, and can I turn that gate on and off myself?
This page is an audit of what six vendors publish about that — Salesforge (Agent Frank), 11x (Alice), Artisan (Ava), Amplemarket (Duo Copilot and Duo Inbox), AiSDR, and Outreach (Revenue Agent) — read on 24 September 2026 from each vendor’s product pages, its documentation and its support or help centre, with the exact wording quoted. For Salesforge, Artisan, Amplemarket, AiSDR and Outreach, a setting this audit turns on is described in the help centre. Where a vendor does not document something, this page says so and names the pages that were opened.
The five gates in an outbound cycle, and the three states each one can be in
Autonomy is not a dial with two settings. An outbound cycle has five points where a human can be inserted, and a product can be open at one and closed at another. We call it the Five-Gate Autonomy Audit, and it is the thing to take into a vendor call.
| Gate | The decision it holds | What it looks like when it is open |
|---|---|---|
| Gate 1 — List | Who gets contacted at all | A person approves the target list or any change to the targeting rules before sourcing spends anything |
| Gate 2 — Draft | What the message says | A person reads the generated copy and can edit or regenerate it, not just the template it came from |
| Gate 3 — Send | Whether it goes | Nothing leaves until someone clicks. This is the gate every vendor means when they say approval mode |
| Gate 4 — Reply | What the agent says back | Inbound replies route to a person, or to the agent, by category |
| Gate 5 — Book | What lands in a diary | A meeting, or the link that creates one, waits for a person rather than going out on the agent’s own decision |
Each gate sits in one of three states:
- Open — a person must act for the cycle to continue.
- Switchable — you can open or close it yourself, and the documentation says at what scope (per campaign, per segment, per reply category, per thread, per workspace).
- Closed — the product performs the step and there is no documented way to put a person in front of it.
Copilot usually means Gate 3 is permanently open; autopilot, that it is closed or switchable. Neither label, on its own, tells you whether Gates 1, 4 and 5 are covered: Salesforge’s Agent Frank Co-pilot reviews replies as well as first touches, while Artisan switches sending and replies separately.
Autopilot vs copilot: which mode each vendor actually ships
Six vendors, read at source. The mode column quotes the vendor. The gate column is our reading of that wording against the five gates above, and the last column says who holds the switch and whether it can move once live.
| Vendor and agent | The mode, in the vendor’s own words | Gates a person can hold | Who sets it, and can it change once live? |
|---|---|---|---|
| Salesforge — Agent Frank | Help centre: “It operates in two modes: Auto-pilot, where it’s fully autonomous, and Co-pilot, where you can monitor and approve its actions.” Product page (Internet Archive capture of 9 August 2026; the live page refuses automated requests): “In Auto-Pilot he won’t require any supervision and in Co-Pilot he will ask for your review before sending messages.” | Gates 2 and 3 together, as one of two named modes: in Co-pilot “you will receive regular updates from Agent Frank requiring you to accept or edit his work”. Gate 1 before launch: the Targeting step lets you review leads and remove any you do not want Frank to target. Co-pilot reviews every message Frank sends, replies included, but no separate reply or booking setting is documented for Agent Frank. Salesforge’s own Multichannel sequences do separate the reply step: its help centre documents an AI reply mode per sequence, Reply as human, Co-pilot (“The AI drafts a reply for each incoming message and waits for your approval before it is sent”) or Auto-pilot | Chosen when the agent is set up (“Choose Agent Frank’s level of autonomy”). Neither the Agent Frank help article nor the product page says whether the mode can be changed once the agent is running. On Multichannel sequences, the reply mode can: “Can I switch modes mid-sequence? Yes.” |
| 11x — Alice | “Three levels, in increasing autonomy. Choose per segment and per motion, not once for the whole workspace.” The three are Review every message (“Nothing sends unseen”), Approval workflow (“A named person signs off”) and Autopilot (“No per-message approval”) | Gate 3 at three levels, with Gate 2 inside the review level, where “Every message can be reviewed, edited, or overridden before it sends”. Gate 4 per reply category: four categories (Information requests, Product inquiries, Meeting requests, Objections), each set to Respond automatically, Skip or Escalate. Gate 5 through the Meeting requests category. Gate 1 is rules rather than approval: ICP filters and exclusions | Configured by workspace admins (the Set up Alice guide is addressed to them), per segment and per motion; reply handling per category. Its FAQs tell you to “move the category back to escalate while you validate” and to “move to an approval workflow until you have confirmed the boundary holds in practice” |
| Artisan — Ava | “Fully autonomous by default… If you want more control while you build trust, you can require approvals at any stage.” And: “Run a campaign in approval mode or let her send. Both are settings, not stages.” Support: with Require approval before sending on, “every message Ava generates waits for your approval before it goes out” | Gates 3, 1, 4 and 5. Gate 3 is Require approval before sending, with Gate 2 inside it: in review mode you approve a message or regenerate it with optional instructions. Gate 1 is “Approve targeting changes”, one of ten controls on the product page. Gate 4 is Autopilot replies: off, Ava waits “for you to send each reply”. Gate 5 sits inside Autopilot replies (see the next section) | You, per campaign (“Chosen per campaign”). The support site says “Settings can be changed at any time, even after a campaign is live”, and a change applies to leads already in progress. The default is autonomous, so each gate is something you add rather than something you remove |
| Amplemarket — Duo Copilot and Duo Inbox | Duo: “Reps stay in control at every step. Duo suggests, learns, and improves, but never sends or changes things without visibility.” Duo Inbox: “Nothing is sent automatically. You review the draft, make edits if you want and send only when you are ready.” | Gates 1 to 3 open per Duo Copilot recommendation: the rep decides “if you want to hit send or dismiss” and can “Edit, regenerate, or send the sequence with one click”. Duo Voice lets you “automatically send LinkedIn voice notes as part of your sequences”. Gate 4 open for Duo Inbox reply drafts, which the rep sends, plus a rep-held switch per thread: the Gmail banner “will also allow you to automatically send follow-up emails that are only sent if the lead doesn’t reply back to you” | The rep, per recommendation and per thread. Users can “control when replies are generated”: for all conversations, only conversations with Amplemarket contacts, or excluding cases such as out-of-office or low-signal replies. Automatic follow-ups need Gmail and the Amplemarket Chrome extension. Admins alone change the signal nudge and competitor battlecards |
| AiSDR | “No manual approval is needed for each outgoing message, which keeps your outreach efficient and scalable.” The approval step is a pre-launch Preview Sequence over sample messages. On replies: “AiSDR does not support a ‘review before sending’ mode” | Gate 4 only, as one switch per campaign, AI Smart Replies: off, “AI will stop engaging as soon as the lead replies” and you take over; on, “the AI will send a link to schedule a call” once the lead shows interest. Gate 3 is closed for email and LinkedIn: “While emails and LinkedIn messages go out automatically, calls are handled by humans”. Gate 2 is approved as samples before launch, not per message | Per campaign: “When you creating a campaign, you’ll have the option to enable AI Smart Replies or leave them disabled” (AiSDR’s wording). Off by default on the pre-launch checklist, and listed among the sequence settings that stay editable “Even in active campaigns” |
| Outreach — Revenue Agent | Personalization Agent: “Reps can review and send the message themselves, or the agent can be configured to send autonomously without manual review.” Plays “can be fully autonomous with agents sending e-mails on behalf of the seller.” Support: “Agents can support both fully automated and semi-automated workflows” | Gates 3, 2 and 1, switchable at the sequence. Support recommends a first step “set as a manual email so your reps can double-check the AI personalized content”, marks AI-drafted steps with a “Personalized” label and purple highlighting, and documents a throttle setting that chooses whether the Agent enrols prospects automatically “or whether the rep needs to activate them first”. The AI agents hub lists Reply Agent, which would handle replies, as “Coming soon” | Whoever configures the Agent: it is “Set up by admins or authorized users”, and both the manual first step and the throttle are sequence settings. The rep then holds the send on that step. On top sits AI Control Hub, an org-wide Off or Allowed setting plus per-feature switches; “only Org Admins can view or change AI feature settings”, and a disabled feature halts in-progress processes before their next step |
Read down the last column, not the mode column. Four of the six put the send decision within reach of whoever runs the campaign or configures the agent (11x, Artisan, Salesforge and Outreach). Amplemarket leaves it with the rep on every Duo Copilot sequence and Duo Inbox reply, plus a per-thread switch for automatic follow-ups. AiSDR has no per-message send approval, and its one switch sits on replies, per campaign.
The same material, one cell per gate. The rest of the page is derived from this table.
| Vendor | Gate 1 — List | Gate 2 — Draft | Gate 3 — Send | Gate 4 — Reply | Gate 5 — Book |
|---|---|---|---|---|---|
| Salesforge | Open before launch (review and remove leads) | Switchable, with Gate 3 (Co-pilot: accept or edit) | Switchable (Auto-pilot or Co-pilot) | Switchable, with Gate 3 (Co-pilot covers replies); no separate Agent Frank setting. Multichannel sequences: per-sequence reply mode | No booking-specific hold documented |
| 11x | Rules only (ICP filters, exclusions) | Switchable (review, edit, override) | Switchable, three levels, per segment and motion | Switchable per reply category | Switchable (Meeting requests: Skip or Escalate) |
| Artisan | Switchable (Approve targeting changes) | Switchable, with Gate 3 (approve or regenerate) | Switchable per campaign | Switchable per campaign (Autopilot replies) | Switchable (three reply-handling choices, qualification, escalation) |
| Amplemarket (Duo) | Open per recommendation (send or dismiss) | Open (edit, regenerate) | Open (rep sends or dismisses each Duo sequence and reply draft); Duo Voice notes can go automatically as sequence steps | Open for reply drafts; rep-held per-thread switch for automatic follow-ups | No booking-specific hold documented |
| AiSDR | Rules only (lead search, suppression lists; a daily auto top-up toggle) | Samples before launch only | Closed for email and LinkedIn (calls are human-led) | Switchable per campaign (AI Smart Replies), no review of automatic replies | No booking-specific hold documented |
| Outreach | Switchable (rep activates prospects, or automatic) | Switchable (manual email step) | Switchable (rep sends, or autonomous) | Reply Agent listed as “Coming soon” | Not documented |
Gate 5: only 11x and Artisan document a booking-specific hold
A booking-specific hold keeps a person in front of the booking step while the rest of the cycle runs on its own. 11x does it through the Meeting requests reply category. The default is “Auto-reply to book, always notify the rep”, and Alice books onto the rep’s calendar. Set the category to Skip (“No automated action. Use for categories you want handled entirely by humans”) or Escalate (“Route to a person, with the conversation context attached”) and a person holds the booking while other categories keep running.
Artisan does it through Autopilot replies. With it on, “You choose how she handles the replies: she can book a meeting with the lead, send a link you’ve set up, or just answer their questions without booking or sending anything.” You can have Ava qualify a lead against up to five conditions before she books or sends a link, and the custom instructions field can tell her “when to escalate a conversation to a person instead of handling it herself”. Artisan’s product pages describe both booking shapes: Ava “schedules meetings directly on your team’s calendar”, or “shares your booking link so the prospect can pick a time”.
For the other four, a booking-specific hold is not documented on the pages read as at 24 September 2026:
- Salesforge (help article Agent Frank): Frank books “either by sending them your calendar link, asking them for their calendar link or directing them to a specific link”. In Co-pilot the message carrying the link is reviewed like any other; in Auto-pilot it is not.
- Amplemarket (Duo Inbox page and help-centre guide): no calendar write is documented. Calendar links go out in drafts the rep sends, or in automatic follow-ups “with contextual messages and calendar links to book more meetings” that the rep has switched on for that thread.
- AiSDR (help articles on AI responses and campaign goals): the goal is Custom, Get a positive response or Book a demo, and a booking goal needs a calendar link. With AI Smart Replies on, “the AI will send a link to schedule a call”; off, a positive reply is handed to you. The only hold is turning automated replies off for the whole campaign.
- Outreach (AI agents hub, Personalization Agent page, Revenue Agent configuration overview): no agent booking step is documented, and Reply Agent is “Coming soon”.
Gate 5 is where a bad qualification becomes a diary entry for a person who did not choose it. With those four, the control you actually have is the gate before it: who sends the message that carries the link. Notification is not a gate: by the time the Slack message arrives, the invitation exists.
Can I approve the sequence messages before they get sent? What AiSDR documents
That is the verbatim title of an AiSDR knowledge base article, and the answer is yes, though not in the sense the question implies. What you approve is a Preview Sequence: sample emails and LinkedIn messages generated from your settings, persona and selected leads or models, before launch. The article is explicit about the limit:
“Live emails will be similar but not identical – AiSDR personalizes each message individually per lead at send time.” … “No manual approval is needed for each outgoing message, which keeps your outreach efficient and scalable.”
That is a Gate 2 control exercised once, not a Gate 3 control exercised per message. After launch you can read what went out, under the Outgoing filter in a campaign’s Latest Events, but not stop it first.
Replies follow the same logic. AiSDR offers three ways to handle them: reply manually from the Dashboard, reply manually from the Inbox, or turn on AI Smart Replies. Its FAQ is direct: “Can AI-generated replies be reviewed before they are sent? At this stage, no,” because a review step “would delay responses, which removes the main advantage of AI”. The article is not consistent on one point. Its Inbox option offers an AI-suggested reply you can “edit it, rewrite it completely, or send it as is”, yet the same FAQ answer goes on to say there is no review mode, “nor is there a way to use AI drafts when replying manually”. Ask AiSDR which is current. What both passages agree on is the narrower point: a reply the AI sends cannot be held. It is a legitimate design, documented plainly.
On Amplemarket the audit turns on scope. Across the Duo, Duo Inbox, Duo Copywriter and Duo Voice product pages, the API documentation index, the academy hub and three Duo help-centre articles, the words autonomous, autonomy and autopilot do not appear once as at 24 September 2026. That is not the same as nothing happening on its own: the Duo Inbox guide describes rep-enabled automatic follow-ups, Duo Voice sends voice notes as sequence steps, and outside Duo, amplemarket.com/ai-sales-automation carries the headline “Sales on autopilot. AI handles the busy work while you close more deals”. The product page calls the follow-ups automatic; that a rep switches them on per thread, and that they only go if the lead does not reply, is in the help centre. That is the general point: a vendor marketing page and the same vendor’s documentation are not equal evidence, and the documentation is the one that describes the software.
How many human minutes an approval gate costs per hundred drafts
An open Gate 3 is not free, and it is the most under-modelled line in an AI SDR business case. Here is the arithmetic, with the inputs labelled so you can substitute your own.
Assumed review rates (these are stated rates, not measured ones — replace them with a stopwatch on your own reviewer):
- 40 seconds per draft to skim, check one claim and approve.
- 90 seconds per draft to open the research citation, verify it and edit a line.
One hundred drafts at 40 seconds each is 100 × 40 = 4,000 seconds, which is 4,000 ÷ 60 = 66.7 minutes, or 1 hour 7 minutes. The same hundred at 90 seconds each is 100 × 90 = 9,000 seconds, which is 9,000 ÷ 60 = 150 minutes, or 2 hours 30 minutes.
Scaled against a 37.5-hour working week, which is 2,250 minutes:
| Drafts needing a decision each week | Review minutes at an assumed 40 seconds each | Share of one 37.5-hour week | What that means for Gate 3 |
|---|---|---|---|
| 100 | 67 min | 3% | Hold Gate 3 open on every draft. At this volume there is no argument for closing it |
| 500 | 333 min (5 h 33 m) | 15% | Still affordable for one reviewer. This is also where reviewers quietly stop reading and start clicking |
| 1,000 | 667 min (11 h 7 m) | 30% | Open on new segments and named accounts; sample the rest |
| 2,000 | 1,333 min (22 h 13 m) | 59% | One reviewer cannot hold this gate. Add a second reviewer or move the control back to Gate 2 |
| 5,000 | 3,333 min (55 h 33 m) | 148% | Gate 3 is theatre. Whoever holds it is rubber-stamping, and you have bought risk without buying control |
Two thresholds fall out of that table. One reviewer tops out at roughly 1,000 drafts a week before the gate becomes a queue rather than a control. And on deal value: an open Gate 3 over a hundred drafts costs 67 minutes, so the gate pays for itself whenever a one-in-a-hundred chance of losing an account is worth more than 67 minutes of that reviewer’s loaded cost. For a high-ACV motion that is almost always true, and it is the case 11x makes for its middle tier, describing an approval workflow as “Appropriate for regulated industries, named strategic accounts, or a first rollout where internal trust matters.”
Gate 4 is worse, because replies arrive on the prospect’s clock. Decide which replies a person must see by category rather than by volume — the same logic behind setting confidence thresholds so an agent escalates rather than guesses. 11x documents this shape explicitly across its four reply categories: information requests “Auto-reply from the knowledge base”, product inquiries “Auto-reply if covered; escalate if not”, meeting requests “Auto-reply to book, always notify the rep”, and objections “Escalate early on; automate once drafts prove reliable”.
Who each mode is wrong for
- Autopilot by default (Artisan’s shape) is wrong for a team whose product knowledge is not yet written down, unless someone turns on approvals before launch. 11x warns against its own autopilot tier that teams skipping review typically blame the technology when the real problem was a thin knowledge base.
- One mode switch per agent (Agent Frank’s shape) is wrong for a team that wants to review replies but not first-touch messages, or the reverse, because the Agent Frank pages describe a single Auto-pilot or Co-pilot choice over the agent’s actions rather than a setting per gate. Salesforge’s Multichannel sequences are the exception inside the same product: there the reply mode is set per sequence, separately from the messages you wrote.
- Rep-sent reply drafts (Duo Inbox’s shape) are wrong for a reactivation motion over tens of thousands of dormant records and any team that needs a reply answered at 11pm. They suit the opposite: few, valuable, signal-triggered conversations a rep would want to own anyway.
- Pre-launch sample approval (AiSDR’s shape) is wrong for anyone whose compliance obligation attaches to the individual message rather than the campaign, anyone who needs to hold back one named account while the rest of the campaign runs, and anyone who needs to read an automatic reply before it goes.
- Step-level switchable under admin governance (Outreach’s shape) is wrong for a team with nobody owning sequence design: if no one sets the first step to a manual email there is no review, and the admin’s org-wide AI switch is not a substitute. It suits an enterprise that wants reps holding the send while an admin decides which AI features exist.
- Per-gate switchable (11x’s shape) is wrong for teams without time to configure it, because it is the most configuration available to get wrong: typically, leaving objections on automatic, where a confident wrong answer about pricing costs more than the reply saved.
If you are designing the gates rather than shopping for them, the companion to this page is our guide to building human-in-the-loop controls into an AI sales agent, which covers what to gate permanently and how to graduate a message type out of review. For deployment model, channels and packaging, see our capability matrix of ten AI sales agent platforms.
Where Zian sits, and what we do not publish
Zian AI publishes a rollout model rather than a gate matrix. Agents start supervised and take on autonomy as performance is demonstrated, across the Discover, Deploy and Scale phases, with pacing and claim boundaries configured up front. SmartReach AI™ orchestrates message, channel and timing across phone, SMS, email and WhatsApp, and PrecisionPitch AI™ split-tests scripts continuously, inside the boundaries set at configuration rather than around them.
By the standard of this page that is an incomplete answer, and it would be dishonest to score our own row from material we have not published the way the six vendors published theirs. Zian is in partnership-application beta, and gate configuration is set with the team during deployment rather than exposed as a self-serve switch. If you are running the Five-Gate Autonomy Audit against vendors, run it against us too, and ask for the answer per gate.
Frequently asked questions
What is the difference between autopilot and copilot mode in an AI SDR?
Copilot means a person must act before a message leaves: the agent researches and drafts, and a human accepts or edits. Autopilot means the agent sends without that step. Salesforge documents both as named modes for Agent Frank: Auto-pilot, where it is fully autonomous, and Co-pilot, where you can monitor and approve its actions. The labels usually describe the send step only, not who handles replies, approves the list or decides a meeting gets booked.
Can I switch an AI SDR from autopilot back to copilot once it is running?
It depends on the vendor, and only one of the six says so for the send step. Artisan states that settings can be changed at any time, even after a campaign is live, and that a change applies to leads already in progress. 11x tells you to move to an approval workflow until a boundary holds, and to move a badly handled reply category back to escalate. AiSDR keeps AI Smart Replies editable in active campaigns, but has no send approval to switch back to. For Agent Frank, Salesforge documents choosing the mode and Outreach documents configuring it, not switching it mid-run, on the pages read on 24 September 2026. The Salesforge help centre does say the AI reply mode on its Multichannel sequences can be switched mid-sequence. Amplemarket documents no autonomous send mode on its Duo pages; Duo Inbox follow-ups are switched on by the rep, per thread.
Which AI SDR vendors let a human approve every individual message?
Of the six read on 24 September 2026, five: 11x, whose first level is review every message; Artisan, when Require approval before sending is on for the campaign; Salesforge in Co-pilot mode; Outreach, whose Personalization Agent page says reps can review and send the message themselves, or the agent can be configured to send autonomously without manual review; and Amplemarket, where the rep sends each Duo Copilot sequence and Duo Inbox reply, though Duo Voice notes can go automatically as sequence steps and a rep can switch on automatic follow-ups. AiSDR does not, and says so plainly in its knowledge base: no manual approval is needed for each outgoing message, because you approve samples before launch, and its automatic replies cannot be reviewed first.
How many minutes does it take to review 100 AI-generated drafts?
At 40 seconds per draft, 100 drafts is 4,000 seconds, or 1 hour 7 minutes. At 90 seconds per draft, where the reviewer opens the research citation and edits a line, it is 9,000 seconds, or 2 hours 30 minutes. Those rates are assumptions, so time your own reviewer on twenty drafts. The threshold that matters: one reviewer holding an open send gate tops out at roughly 1,000 drafts a week, which is 11 hours 7 minutes, or 30% of a 37.5-hour week.
Does an approval gate on every message slow outbound down?
Only at the send step, in proportion to volume. At 500 drafts a week the review cost is 15% of one working week, which most teams absorb. At 2,000 drafts a week it is 59% of a week for one reviewer, and the gate stops being a control and becomes a queue. The usual answer is not to remove the gate but to narrow it: hold it open on new segments and named accounts, sample the rest, and set reply handling by category.
Which step should stay behind a human gate permanently?
Our answer is Gate 3 on named strategic accounts and on any message that touches price, contract terms or a compliance claim, regardless of how well the agent performs. Of the six vendors, 11x is the one whose documentation names these specifically on the pages read. Its brand protection page calls an approval workflow appropriate for regulated industries, named strategic accounts, or a first rollout where internal trust matters, and lists six unconditional handoffs to a human regardless of confidence: pricing beyond your published posture; contract, legal or procurement terms; security and compliance specifics; roadmap or delivery commitments; anything from a named strategic account or an existing customer; and any explicit request to speak to a person. Artisan lets you write your own escalation instructions.
Where every figure on this page comes from
| Figure | Who published it | Link | Date read |
|---|---|---|---|
| Agent Frank “operates in two modes: Auto-pilot, where it’s fully autonomous, and Co-pilot, where you can monitor and approve its actions”; Co-pilot sends “regular updates from Agent Frank requiring you to accept or edit his work”; the three booking routes, “sending them your calendar link, asking them for their calendar link or directing them to a specific link” | Salesforge (help centre) | help.salesforge.ai: Agent Frank | 24 September 2026 |
| AI replies on the Multichannel “Human Path” in three modes, Reply as human, Co-pilot (“The AI drafts a reply for each incoming message and waits for your approval before it is sent”) and Auto-pilot; “Can I switch modes mid-sequence? Yes.” | Salesforge (help centre) | AI Replies for Sequences | 24 September 2026 |
| The Targeting step lets you review leads and “Remove leads that you don’t want to target” | Salesforge (help centre) | Using Leadsforge-Powered ICP Setup in Agent Frank | 24 September 2026 |
| “Choose Agent Frank’s level of autonomy. In Auto-Pilot he won’t require any supervision and in Co-Pilot he will ask for your review before sending messages.” | Salesforge (product page) | salesforge.ai/agent/frank (the live page returns HTTP 403 to automated requests, so this wording is read from the Internet Archive capture of 9 August 2026; the help-centre article above is the current source) | 24 September 2026 |
| Three approval levels, chosen “per segment and per motion”: Review every message (“Nothing sends unseen”), Approval workflow (“A named person signs off. Appropriate for regulated industries, named strategic accounts, or a first rollout where internal trust matters.”) and Autopilot (“No per-message approval”); the six unconditional escalations; the FAQ answer “move to an approval workflow until you have confirmed the boundary holds in practice” | 11x | docs.11x.ai/security/brand-protection | 24 September 2026 |
| “Every message can be reviewed, edited, or overridden before it sends”; the warning that teams that skip the review phase “typically conclude the technology doesn’t work when the real problem was a thin knowledge base” | 11x | docs.11x.ai/alice/ai-personalization | 24 September 2026 |
| Four reply categories and their defaults (Information requests, Product inquiries, Meeting requests “Auto-reply to book, always notify the rep”, Objections); three handling options, Respond automatically, Skip (“No automated action. Use for categories you want handled entirely by humans”) and Escalate (“Route to a person, with the conversation context attached”); the FAQ answer “move the category back to escalate while you validate” | 11x | docs.11x.ai/alice/smart-replies | 24 September 2026 |
| Alice “Books the meeting” “onto your calendar” | 11x (help centre) | Getting started as a User | 24 September 2026 |
| Alice’s ICP as “firmographic filters, role filters, and exclusions” | 11x | docs.11x.ai/alice/icp-and-targeting | 24 September 2026 |
| The Set up Alice guide applies to “workspace Admins configuring Alice” | 11x | docs.11x.ai/alice/setup | 24 September 2026 |
| “Fully autonomous by default”; approvals can be required “at any stage”; “Full control or full autonomy”, “Chosen per campaign”; “Both are settings, not stages”; four of the ten controls listed (the ten are Approve before sending; Lock tone, CTA, and length; Set budget caps per campaign; Control reply sensitivity; Choose active channels; Define lead routing rules; Coach Ava with instructions; Approve targeting changes; Set auto-reply threshold; Pause campaigns instantly); Ava “schedules meetings directly on your team’s calendar” | Artisan | artisan.co/ai-sales-agent | 24 September 2026 |
| Ava “shares your booking link so the prospect can pick a time” | Artisan | artisan.co/meeting-scheduling-software | 24 September 2026 |
| Require approval before sending (“every message Ava generates waits for your approval before it goes out”); Autopilot replies and its three choices; up to five qualification conditions; custom instructions for “when to escalate a conversation to a person”; Route booking meetings; “Settings can be changed at any time, even after a campaign is live” | Artisan (support site) | support.artisan.co/configure-campaign-settings | 24 September 2026 |
| Review mode (approve, or Regenerate with optional instructions) and auto send mode | Artisan (support site) | What happens after you launch a campaign? | 24 September 2026 |
| “Reps stay in control at every step. Duo suggests, learns, and improves, but never sends or changes things without visibility.”; “decide if you want to hit send or dismiss”; “Edit, regenerate, or send the sequence with one click” | Amplemarket | amplemarket.com/duo | 24 September 2026 |
| “Nothing is sent automatically. You review the draft, make edits if you want and send only when you are ready.”; “Automatically follow up on prospect replies with contextual messages and calendar links to book more meetings”; “control when replies are generated”, with the three draft-generation options | Amplemarket | amplemarket.com/duo-inbox | 24 September 2026 |
| The Gmail banner “will also allow you to automatically send follow-up emails that are only sent if the lead doesn’t reply back to you”; automatic follow-ups need the Chrome extension, which supports Gmail only | Amplemarket (help centre) | Duo Inbox: A detailed guide | 24 September 2026 |
| “you can automatically send LinkedIn voice notes as part of your sequences” | Amplemarket | amplemarket.com/duo-voice | 24 September 2026 |
| The signal nudge and competitor battlecards: “Only account admins can change it” | Amplemarket (help centre) | Duo Copilot: How to adjust Duo’s AI sequences | 24 September 2026 |
| “Sales on autopilot. AI handles the busy work while you close more deals” | Amplemarket | amplemarket.com/ai-sales-automation | 24 September 2026 |
| Unattended sequence work: “automatically start a connection sequence” when a meeting is booked, “automatically nurture them back with lighter outreach after a waiting period” | Amplemarket | amplemarket.com/workflows | 24 September 2026 |
| Zero occurrences of autonomous, autonomy or autopilot across the Duo, Duo Inbox, Duo Copywriter and Duo Voice pages, the API documentation index, the academy hub and three Duo help-centre articles, each fetch positive-controlled on the string Amplemarket | Amplemarket (pages read and counted for this page) | duo-copywriter, docs.amplemarket.com, academy, Duo Copilot Dismisses Explained, plus the Duo pages and articles above | 24 September 2026 |
| “No manual approval is needed for each outgoing message”; approval is a pre-launch Preview Sequence over samples; live emails “similar but not identical” | AiSDR | Can I Approve the Sequence Messages Before They Get Sent? | 24 September 2026 |
| AI Smart Replies on and off, and what changes in each state; “Once the lead shows interest, the AI will send a link to schedule a call” | AiSDR | AI responses ON and AI responses OFF | 24 September 2026 |
| “When you creating a campaign, you’ll have the option to enable AI Smart Replies or leave them disabled” (AiSDR’s wording); the three reply options; “Can AI-generated replies be reviewed before they are sent? At this stage, no.” and “nor is there a way to use AI drafts when replying manually”; the AI-suggested reply in the Inbox | AiSDR | How to Manage Lead Replies in AiSDR? | 24 September 2026 |
| “AI Smart Replies → OFF (enable later, after first positive replies)” among the defaults on the pre-launch checklist | AiSDR | Time to Launch: Final Checklist Before Activating Your Campaign | 24 September 2026 |
| AI Smart Replies listed among the sequence settings adjustable “Even in active campaigns” | AiSDR | How to Edit a Campaign: What’s Safe, What Has Consequences | 24 September 2026 |
| The three goal options, Custom, Get a positive response and Book a demo; a booking goal needs a calendar link | AiSDR | What Does “Goal” Mean in a Campaign Builder? | 24 September 2026 |
| “While emails and LinkedIn messages go out automatically, calls are handled by humans”; the “Add fresh leads daily” toggle that tops up leads automatically | AiSDR | Campaign Setup: Step-by-Step Guide After Adding Leads | 24 September 2026 |
| Sent emails visible under the Outgoing filter in Latest Events | AiSDR | Reviewing Outgoing Emails in AiSDR | 24 September 2026 |
| Plays “can be fully autonomous with agents sending e-mails on behalf of the seller”; “Admins can control how autonomous agents are across segments”; Reply Agent listed as “Coming soon” | Outreach | outreach.ai/ai-agents | 24 September 2026 |
| “Reps can review and send the message themselves, or the agent can be configured to send autonomously without manual review” | Outreach | outreach.ai/ai-agents/personalization-agent | 24 September 2026 |
| “when they launch an email where the agent has done the heavy lifting, they simply need to review the output, make any tweaks if they want to make adjustments, and hit Send” | Outreach | outreach.ai/ai-agents/revenue-agent | 24 September 2026 |
| “Set up by admins or authorized users”; “Agents can support both fully automated and semi-automated workflows”; the recommended first step “set as a manual email so your reps can double-check the AI personalized content”; the “Personalized” label and purple highlighting; the throttle choice between automatic enrolment and “the rep needs to activate them first” | Outreach (support site) | Revenue Agent configuration overview | 24 September 2026 |
| AI Control Hub as an org-wide Off or Allowed setting with per-feature switches; “only Org Admins can view or change AI feature settings”; a disabled feature halts in-progress processes “before their next step” | Outreach (support site) | How to enable and manage agents in AI Control Hub | 24 September 2026 |
| 40 and 90 seconds per draft review; 67 minutes and 150 minutes per hundred drafts; the weekly share table | Zian AI, stated assumptions with the arithmetic shown on this page. Not a measurement | This page | 24 September 2026 |
Audit your shortlist before you pick a mode
Take the Five-Gate Autonomy Audit into your next three vendor calls and ask for the state of each gate, in writing, with the documentation page that says so — and ask for the help-centre page, where the switches are described. Zian AI is in partnership-application beta and configures gate behaviour with your team during deployment. Apply For Partnership to walk through the five gates against your own motion.