Go into **Jobs**, in the sidebar. The screen has two views — **Queue** and **Calendar** — over the same set of tasks, with the same filters at the top: by agent, by who it's assigned to, by status, by type, by link (internal, lead, conversation or contact), and by who created it.  ## 2.1 A task's statuses | Status | Meaning | | --- | --- | | Suggestion | The assistant proposed the task on its own (for example, it committed to something with a customer) and is waiting for you to confirm or reject it. | | Scheduled | It has a date, recurrence, event or condition set; its turn hasn't come up yet. | | Active | Its turn has come up and it's running right now. | | Waiting for approval | It needs someone on the team to approve it before continuing. | | Completed | It finished successfully. | | Failed | It finished with a problem. | | Overdue | It had a deadline and it passed without being resolved. | | Cancelled | Someone cancelled it by hand. | A task's **type** tells you where it came from: **Manual assignment** (someone on the team created it from this same form), **Assistant's promise** (the assistant committed to something in a conversation), **Recurring** (a fixed schedule), or **Flow** (it's a step of a flow — section 3). ## 2.2 Creating a task by hand **Requirements:** Admin or Editor role. 1. Tap **"Create task"**, at the top right. 2. Type the **title** and the **instruction** — in your own words, as if you were explaining it to someone on the team. Next to the field, the **"Dictate by voice"** button (microphone icon) lets you say it instead of typing it — see the note below. 3. Choose the **agent** that's going to run it. 4. Choose the **trigger**: once (date and time), recurring (a frequency from an already-ready list: every day, every Monday…), by event (when a new contact comes in or moves stage), or condition. 5. Choose **who runs it**: the digital agent, or a specific operator on your team. 6. Choose the **link**: internal (no associated contact), lead, or contact. 7. Tap **"Create task"**. > **Note — Dictating the instruction instead of typing it** > > Press the microphone next to the "Instruction" field and speak — up to 60 seconds. While recording you > see a timer ("Recording… N s") and can stop it whenever you want; once you release it, a "Transcribing…" > notice precedes the text, which appears LITERALLY in the field — it never saves on its own. You read it, > edit it if needed, and tap "Create task" yourself, just like with any hand-typed instruction. If your > browser has no microphone or doesn't allow it, the button disappears ("Voice dictation not available in > this browser") without blocking the rest of the form; if the dictation service doesn't respond, an > honest notice tells you so instead of simulating a transcription. > **Note — A condition can be checked in two ways** > > "When an event occurs" is free: it fires the exact instant something happens (a lead is created, it > changes stage), without spending AI budget. "When the assistant evaluates it" has a cost: the assistant > reviews the condition every so often using AI, until it's met or the deadline passes — reserve it for > what a machine can't decide on its own ("check whether the customer seems interested"). ## 2.3 A task's record Tap any row in the queue (or a task on the calendar) to open its record. You'll find: - **Instruction**: what was asked of the agent, in plain language. - **Next steps and timing**: when the next step will fire (or fired). - **Assignment**: who it's currently assigned to, with the option to reassign it — to the digital agent or to an operator, in either direction. - **Step-by-step history**: every event in its life (created, scheduled, fired, run, approved, completed…) with the exact date and time. ## 2.4 The calendar Tap the **"Calendar"** tab to see the same queue in a weekly layout. Navigate week by week with the arrows or jump back to "Today". Your Calendar's appointments show up here too, read-only, so you have everything scheduled at a glance — the assistant's jobs and human appointments together. ## 2.5 Approving a task A step can be marked "requires approval" — in its own **Approval** field, or by picking the **Approval → "Request approval"** piece in a flow's visual editor (section 3). When the trigger reaches that step, instead of running its action directly, the task moves to **"Waiting for approval"** status and shows the **Approve** and **Reject** buttons, both on the queue row itself and on its own record. Any operator with access approves or rejects it, from the row or from the record — it doesn't have to be whoever created it or whoever it's assigned to. If approved, the action actually runs and the task moves to **"Completed"**, with its full step-by-step history: created, scheduled, triggered, approval requested, approval resolved (with who approved it), action executed and completed. > **Note — A step with no action attached never reaches "Waiting for approval"** > > Approval gates a specific curated action (a message, a spreadsheet row, a tool from an MCP server…). A > step with no action attached has nothing to approve: it stays "Active", waiting for someone to complete > it by hand with closing notes (section 2.3). ## 2.6 Approving item by item (flows with repetition) When the task requiring approval is a step inside a flow with **"Repeat for each item"** (section 3.7) — for example, reviewing a low-confidence invoice inside the "Invoices by email" template, Manual 61 — the **Jobs** screen groups them differently: instead of a table of loose rows, you'll see an **approval queue grouped by run**, with one header per flow ("‹flow name› · N pending") and, underneath, one row per item with its own identifier, the specific step, and the reason. The reason for each row is always one of these two, never free text: | Reason | What it means | | --- | --- | | Low confidence | An earlier step for THIS item extracted or validated a piece of data with little certainty (for example, a hard-to-read invoice). The row shows the exact confidence figure and the threshold it needed to clear — never just "low confidence" with no numbers. | | Approval required by the flow | The step is marked "requires approval" unconditionally: it always stops for review, whatever the confidence. | A **"View run"** link in the group header takes you to that run's own record (section 3.8) to see the rest of its items at a glance. Approving or rejecting one row **never affects the others** in the same group — each item has its own task, its own history and its own decision, exactly like any two loose tasks. > **Note — It may ask you to approve more than once for the same item** > > When the step that writes performs several actions in a row (for example, "Log to sheet + Upload to folder + Mark as processed", the "Invoices by email" template in Manual 61), it's normal to be asked to approve once per remaining action for that specific item — keep approving until the item reaches "Done", it's not an error or a duplicate approval.