AI Reply
Lets the assigned or channel-routed AI agent answer the latest customer message when AI switches, conversation holds and message freshness allow it.
Quick start
- Go to Automate → Workflows and open a workflow. Click + and pick AI Reply.
- Optional: write an instruction for this one step only, for example "keep it to two sentences".
- Right after it, add a Condition on AI sent a reply is equal to Yes.
- On the Condition's branch for when the AI did not answer, add Escalate or Assign To (human).
- Set Status to Published in the top toolbar, Save, confirm Publish. Agents: ask an owner or admin.
Watch out
- Most failures here notify nobody at all. Put Escalate or Assign To (human) on the path where the AI did not answer.



Details
Getting here
Automate → Workflows, open a workflow, click + on the canvas, then pick AI Reply under Add Steps.
Before you start
- An owner, admin or agent can add this step and save the workflow. Only an owner or admin can publish it, and publishing is what makes it run on live conversations.
- Building and running workflows depends on your plan; see pricing.
- An AI agent set up for this workspace. The step answers as whichever agent is currently assigned to the conversation, or an agent matched by the channel’s routing rules. If no agent is bound to the channel, the step fails without generating a reply. Add an assignment or fix routing, and keep a human fallback.
Set up
- Open the workflow and click + where the AI should answer.
- Pick AI Reply.
- Optional: write an instruction — extra direction for this one step, added on top of the agent's own persona, for example "keep it to two sentences".
- Add a Condition on AI sent a reply after it, and put an Escalate or an Assign To (human) on the path where the AI did not answer.
- Save the step. To publish: set Status to Published in the toolbar at the top of the editor, click Save, then confirm with Publish. The dropdown on its own changes nothing until you save.
Fields
- AI instruction
- Extra direction for this step only, appended after the agent's own persona — it does not replace the agent's setup, knowledge base or guardrails.
- Default: empty
How it works
- Before generating anything, the step checks the AI kill switches in order: platform-wide, then this workspace's own AI switch, then the channel's — the same order and the same switches a normal auto-reply checks.
- The platform one is internal; you do not control it.
- The workspace switch is Stop all AI / Resume AI under Settings → AI; the channel switch is Stop AI / Resume AI under Settings → Channels.
- Any one of them being on fails the step with no message sent and no model call made.
- It then checks conversation-level holds: a Pause AI / Unassign step earlier in this or another workflow, or a person already holding the conversation — a teammate took it over, or the customer asked for one.
- Both fail the step the same way: the AI does not speak on either, no matter what a workflow's own steps are otherwise allowed to send there.
- It answers as whichever AI agent is currently assigned to this conversation — by Assign To (AI), or automatic routing — using an agent matched by the channel’s routing rules when none is assigned. If no agent is bound to the channel, the step fails without generating a reply.
- A conversation that already belongs to a specific agent keeps that agent's voice and knowledge base rather than switching to the default mid-conversation.
- It reuses the exact process and limits a normal auto-reply uses — the same knowledge-base lookup, the same check that the answer is grounded in it — rather than a separate model call.
- Cost and guardrails are identical either way.
- The AI's usage is counted even if the reply is not sent — the spend already happened once the model answered.
- What the model decided is written to three fields a Condition step after this one can read.
- They are AI decision (
ai.action: replied, suggested a reply, or handed to a human), AI confidence (ai.confidence) and AI sent a reply (ai.replied). - Only a decision to answer automatically, with non-empty text, is sent. Everything else — a suggestion, an escalation, or an empty reply — is treated as the AI not answering (see When it fails), and "actually replied" is left false.
- If the agent currently answering has its own Auto-reply toggle off, the decision is a suggestion whenever it would otherwise have qualified to send.
- But a message the model reads as sensitive or an explicit request for a human still decides to escalate regardless of the toggle, and so does a confidence below the platform's hard escalate floor. Either way this step never sends on that agent.
- A sent reply still goes through the ordinary send path, so it needs the 24-hour messaging window to be open on a windowed channel, the same as Send a Message.
- It is recorded under the agent's own name, not the workflow's.
- If the customer sends another message while the step is generating or about to send, the step stops rather than answering a question that is no longer the latest one. The run halts on this step instead of continuing with a stale answer.
When it fails
- Before generation ever starts, several checks can fail the step outright, and none of them attempt to notify a human.
- Those checks are: a kill switch on (platform, workspace or channel), the conversation paused by Pause AI / Unassign or held by a person, no customer message to answer, or the agent could not be loaded.
- "No customer message to answer" covers a booking, an order, or a scheduled trigger that started the run with nothing for the AI to reply to. Each of these fails with its own reason and nothing is generated.
- Generation itself fails — a model or infrastructure error, or a timeout. A thrown error fails the step with
AI generation failed; a failed model call is logged like a decline, with confidence 0.00, and tries to notify a human like one. The step does not hand the conversation over itself: the customer's message is left for Inchat's automatic hand-off, which passes it to a person shortly afterwards. If an earlier step of the same run already messaged the customer, that hand-off no longer covers the message, so the step notifies a human straight away instead, once. - The model chose to suggest or escalate — a genuine decline, after generation completed: the step fails, "actually replied" stays false, with a reason such as
AI did not answer (escalate, confidence 0.20). - This case tries to notify a human, and so does an empty reply (below). The push notification is sent every time this happens, with no deduplication.
- The in-app notification is deduplicated whenever an earlier Ask a Question step already set its own "asked at" marker on the run — not only when this step is itself retrying that question's own failure branch.
- The model returned a decision to answer automatically, but with empty text — the generation pipeline produced a decision but no usable words.
- The reason names the empty reply specifically, with the confidence score alongside it, so this is never confused with a deliberate decline.
- The same notify attempt runs as for a decline.
- The human-notification attempt itself can under-deliver, but not simply because the push failed: the in-app row landing is enough to count as notified, even if the push reached nobody.
- The reason only gains an extra
— human page failednote on a decline, never on an empty reply, and only when the push reached nobody and no in-app row landed. - No in-app row lands when there is no assignee or workspace admin to write one for, or the row write failed. Either the push or the row landing counts as notified.
- The customer's message changed — checked both right before generation starts and again right after it finishes: either way, the step fails, with no notify attempt, and the run stops entirely at this step. Later steps in the same run are not attempted on top of a reply that would have answered the wrong question.
- A decline, where the model answered but chose not to send it, lets the flow continue, and so does an empty auto-reply. After a failed generation, the flow's later customer messages are skipped and a later Wait ends the run; bookkeeping and Escalate still run. A kill switch or workflow AI pause also suppresses later customer messages and waits. A changed customer message stops the run.
- Place a Condition on AI sent a reply right after it, with Escalate or Assign on the path where the AI did not reply, to catch that case. A Send a Message on that path is skipped after a failed generation.
- The reason for any of the above is in the workflow's Activity tab, under Execution log, on this step's line.
Limits
- It answers as the agent currently on the conversation, not necessarily the workspace's named default — an earlier Assign To (AI) step changes who answers here.
- The instruction field adds to the agent's persona for this one step; it cannot override the agent's knowledge base, guardrails or the confidence gate itself.
- A low-confidence or declined answer does not retry inside this step — it fails once, tries to notify a human, and moves on. A retry needs another AI Reply step reached some other way, for example after a customer rephrases.
- The human-notification push has no deduplication of its own — every declined run of this step sends one, even if the last one was seconds ago.
- The in-app notification row is deduplicated, keyed off any earlier Ask a Question step's "asked at" marker in the same run, not only one this exact step is retrying on behalf of.
- An agent whose Auto-reply toggle is off never sends through this step. The decision is usually a suggestion.
- But a message the model reads as sensitive or a direct request for a human, or confidence below the platform's hard floor, still decides to escalate regardless of the toggle.
- Most failures on this step — every kill switch, both conversation holds, no customer message, an agent that could not load, and a stale customer message — notify nobody at all: no push, no in-app row.
- Only a decline, an empty auto-reply, a failed model call, or a thrown error after an earlier step already messaged the customer tries to notify here, and "tried" is not "succeeded". Otherwise a thrown error is left to the automatic hand-off.
- The push alone reaching nobody does not on its own mark the attempt as failed — the in-app row landing is enough.
- It does not pause the run to wait for anything — the step already generated and, if it qualified, sent its one reply by the time the run continues.
- It does not stop the AI from answering on other, unrelated workflows or on plain inbound messages to this conversation.
- Only the specific kill switches and holds listed in How it works silence it, and only for the duration they are set.
Best practices
- Always follow it with a Condition on AI sent a reply. Put a human fallback — Escalate or Assign To (human) — on the path where the AI did not answer.
- Do not rely on the step's own human-notification alone.
- Most of this step's failures never attempt to notify anyone at all: every kill switch, both holds, no customer message, an unloadable agent, a stale message. It is a best-effort alert on the decline path only, not a guarantee someone was told about every way this step can fail.
- Use a Condition on AI sent a reply rather than assuming success when you need the flow to branch differently depending on whether the AI actually spoke.
- Keep the instruction short and specific to this one turn — it is not the place to redefine the agent's whole personality for a single step.
Use cases
- Let the AI answer a routine question mid-flow, with a Condition on AI sent a reply and Escalate on the path where it declined.
- Give the AI a one-off instruction for this step only — such as "summarise in one sentence" — without changing its permanent setup.
- Branch on AI sent a reply to decide whether to close the conversation automatically or hand it to a teammate.
FAQ and troubleshooting
- Does AI Reply check the same kill switches as a normal inbound reply?
- Yes — platform, workspace, then channel, in that order, plus the conversation-level holds (Pause AI / Unassign, or a person holding the conversation). Any of them silences this step exactly the way it silences a normal auto-reply.
- What happens if the AI isn't confident enough to answer?
- The step fails, its "actually replied" fact stays false, and it tries to notify a human that the AI did not answer. Nothing is sent to the customer. The push side of that notification is not deduplicated — a workflow that keeps re-running this step alerts fresh every time. The in-app row is deduplicated whenever an earlier Ask a Question step already set its own timestamp on the run. Put a Condition on AI sent a reply is equal to Yes after this step, with Escalate or Assign on its None branch, so someone follows up regardless.
- The agent I assigned has Auto-reply switched off. Does AI Reply still work?
- It generates and evaluates the reply, but this step never sends on that agent. The decision is usually a suggestion. A message the model reads as sensitive or a direct request for a human — or confidence below the platform's hard escalate floor — still decides to escalate even with the toggle off. Either way, a human is who ends up notified, not the customer.
- Which AI agent answers if I've pinned this conversation to a specific one?
- Whichever agent is currently assigned to the conversation — by an earlier Assign To (AI) step or automatic routing — or an agent matched by the channel’s routing rules if none is assigned. If no agent is bound to the channel, the step fails without generating a reply. Add an assignment or fix routing, and keep a human fallback.
- Does the run continue if the AI didn't answer?
- After a decline or an empty auto-reply, yes. After a failed generation (a model or infrastructure error, or a timeout), the flow's later customer messages are skipped and a later Wait ends the run. Bookkeeping and Escalate still run. Inchat's automatic hand-off passes the customer's message to a person shortly afterwards. A failed model call also notifies a human at once, and so does any failed generation after an earlier step already messaged the customer. A kill switch or workflow AI pause also suppresses later customer messages and waits. A changed customer message stops the run. Keep an Escalate or Assign To (human) fallback.
Reading this with an AI assistant? Plain Markdown version.