Reliable Workflows
Workflows give an important employee job one saved place from your request to its recorded result. Choose a job from the Work menu or ask in chat, then see its progress, decisions, and evidence without having to reconstruct the work later.
A workflow gives a supported job a durable home. Start with the outcome you want, such as managing your inbox, preparing for a meeting, building a prospect list, or creating and publishing social content. Your employee saves the brief, completes the relevant stages, and records what happened as the work moves forward. You can return to the same job from the conversation instead of explaining the context again.
Workflows keep flexible thinking and dependable progress together. Your employee can research, write, and adapt to the work in front of it, while the workflow keeps track of the stage, results, waiting reason, and the next decision. A job can pause when it needs a connection, a detail from you, or approval for a protected action, then continue from the saved state when that decision arrives.
Playbooks and workflows have different jobs. A playbook is reusable guidance for a kind of work. A workflow is the saved progress for one outcome. The same playbook can help many workflows without copying instructions into each one, and a workflow can use more than one playbook when the outcome needs several kinds of work. That separation keeps the system easier to improve while preserving a clear record for every job.
You stay in control of connected actions. Reading from a connected inbox or calendar can supply the facts an employee needs. When an action could affect someone else, such as sending an email, publishing content, or changing a calendar event, the workflow follows your tool approval setting. Ask first keeps the exact proposed action in front of you. You can approve it, request changes, reject it, or choose a more automatic setting later for tools you decide to trust.
A one time request stays one time. If you want work to recur, you set up a routine with its purpose, cadence, scope, end date, and run limit. The routine can start the supported workflow at the agreed time, while the workflow still enforces its connection and approval requirements. This lets an employee become useful between conversations without turning a single request into an open ended background task.
Choose the Outcome, Not the Process
The Work menu is organized around results you can ask an employee to deliver. Each job describes what it will produce, so you can select a useful outcome without learning internal steps or configuring an automation first. You can also ask for the same work in chat when you already know what you need.
Once the work begins, the employee gathers the brief it needs and saves it with the job. You can give a preference, edit the direction, or leave a decision for later. The saved job keeps those choices attached to the result instead of mixing them into unrelated conversations.
See What Is Done and What Needs You
A workflow makes its state visible while it runs. You can see completed work, the current stage, and why it is waiting when the employee needs a decision, a connection, or a missing detail. Results and source evidence remain with the job so you can review what the employee used to reach its recommendation.
This is especially useful for work that crosses several moments. An inbox review can sort messages, save reply proposals, and wait for your decisions. Meeting preparation can check the selected calendar, preserve the meeting context, and reuse permitted research before it prepares the next brief. Sales and marketing work can keep their own outcomes separate while following the same reliable model.
Keep Sensitive Actions Under Your Control
Connecting a tool gives the employee only the access you choose during the secure sign in. It does not change your approval policy. A workflow uses the same central tool settings you see in Approvals, so the rule is consistent whether work starts from chat, the Work menu, or a routine.
For an action that needs your approval, you see the actual proposed email, post, or calendar change before it happens. You can approve the exact action, request edits, reject it, or ignore an email that needs no reply. A rejection never gives the employee permission to send a replacement on its own.
How It Works
A supported outcome keeps one record from the initial brief through its recorded result.
Choose a job from an employee's Work menu or describe the outcome in chat. The workflow collects the brief and starts its first saved stage.
The employee uses the relevant playbooks, skills, and connected tools to do the work. Each stage records its result, evidence, or waiting reason before the job moves forward.
When work needs your decision, the workflow pauses at that exact point. Your choice continues the same saved job, so the employee does not need to start again.
Use Cases
Review an inbox without losing reply decisions
Ask your Personal Assistant to manage your inbox. It can sort a bounded review, highlight what needs attention, and save each reply proposal for your approval, change request, rejection, or ignore decision.
Keep a sales outcome moving through its next step
Ask Sales to build a prospect list, research a target account, send approved outreach, or prepare for a meeting. The selected outcome keeps its own saved progress while the employee works with the information you have approved.
Create social content with the right review point
Ask Marketing to create a post or a batch of content. The workflow can preserve the brief, drafts, and selected channels while keeping publication behind your chosen approval setting.
Comparison
| Before | After |
| Handling reliable workflows through separate tools, notes, and follow-ups makes the process difficult to repeat or inspect. | Reliable Workflows gives your AI workforce one defined workflow: Workflows give an important employee job one saved place from your request to its recorded result. Choose a job from the Work menu or ask in chat, then see its progress, decisions, and evidence without having to reconstruct the work later. |
FAQ
Do I need to configure every workflow before I use it?
No. Choose a supported outcome from the Work menu or ask in chat. The employee asks only for the details and connections needed for that job.
What is the difference between a workflow and a playbook?
A playbook is reusable guidance for a type of work. A workflow saves the stages, decisions, and results for one supported outcome. The same playbook can help many workflows without duplicating it.
Can a workflow send email or change my calendar without asking?
Only if you explicitly change that connected tool's approval setting. Ask first keeps protected outgoing actions waiting for your decision, and connecting an account does not change that setting.
Can work run on a schedule?
Yes, when a supported workflow has been set up as a routine with an agreed scope, cadence, end date, and run limit. A one time request does not create recurring work.
Where Reliable Workflows fits
Reliable Workflows is part of How they coordinate as a team.
Run your AI workforce the way you run a good team. When a stretch of work deserves a name, your team lead opens a sprint, agrees one goal with you, works it, and closes it with an honest written account of what landed and what did not. Agents delegate tasks to each other through team chat, visible in your activity feed.
Read the guide
More in Sprints Planning
- Give Your Team an AI Leader: Every team you hire gets one employee assigned as its leader, marked with a crown badge. The leader checks the team's task board on a recurring schedule, delegates pending work to the right specialist, reviews what comes back, and unblocks anyone who is stuck. You always have one clear first contact for a team instead of guessing who to message.
- Delegation: When you ask a team leader for something, it splits the work and hands pieces to the right teammates instead of doing everything itself. The content writer drafts copy, the designer builds visuals, the social manager schedules posts, all coordinated through team chat while you talk to one person. Delegation follows your org structure: a leader hands off to its own members, or to another team's leader for cross-functional work. Every agent-to-agent message is logged and visible in the activity feed, so you can watch the handoffs instead of trusting a black box.
- Sprints: Give a team's work a goal and a time window instead of an endless task list. Open a sprint, the team leader delegates and works it on its own schedule, and you (or the leader) close it with a written review of what shipped and what didn't. Sprints are entirely optional: a team with no sprint still runs its task board normally.
- Team OKRs and KPIs: Give a team a real scoreboard: qualitative Objectives for the cycle, and measurable KPIs with a baseline, a target, and a current value that either roll up under an Objective as its Key Result or stand alone as an ongoing health metric. Any team member can set or update these through plain conversation with the team leader owning them by default, and every sprint and every autonomous wake reads the same numbers, so the team is working toward something specific instead of just clearing a task list.
- Team Goals & Guidelines: Give a team a shared Vision statement and a set of Guidelines that every member reads before they act. The Vision is the one-line long-horizon anchor for where the team is heading; Guidelines are the working rules, standards, and coordination norms the team follows day to day. Both live on the team's Charter and stay attached to every sprint and task the team runs.
- Start a Task Now: Every open task has a Start now button. Click it and the assigned employee begins working that task right away, instead of waiting for it to come up in its normal order.
- Routines: Set a task on a rhythm once, daily, weekly, or a custom cycle, and it runs itself from then on. No re-creating the same task every Monday, no remembering to assign it again next week.
- Playbooks: Every employee leads with a short set of proven playbooks, the highest-value jobs they can run start to finish for the role you hired them for.
- Comment on a Sprint: Ask a question or leave a note directly on a sprint, right where the work is happening. The sprint's team leader, the employee who owns and runs it, wakes up, reads your comment with full context, and answers in the same thread.
- The Weekly Sprint Check-In: Once a week your team lead looks at the state of the work and raises the sprint question with you directly: is there a stretch of work worth naming and turning into a sprint, or has the open sprint run past its date and earned a close? You answer in a sentence and the team acts on it. When you agree to wrap up, the lead writes the closing review in their own words, drawn from the Task Board and the Work Journal, and unfinished work stays exactly where it is until you decide what carries forward. Nothing opens, closes, or expires on a timer; a sprint only ever moves because you or your team lead moved it.
- Work Toward Your OKRs: Give your AI team objectives and key results in plain language, the same way you'd brief a human manager, and the team leader plans every sprint against them. The task board shows work tied back to a goal, the weekly check-in decides when a sprint wraps, and closing a sprint produces a written account of what moved and what slipped. Ask about any goal at any time and get an answer grounded in the team's actual work journal, not a status meeting.
Explore