Automations
Describe a process in chat and Lightfield's agent programs an automation that runs it on a schedule or trigger.
A Lightfield automation is a sequence of AI steps that the agent builds with you in chat. There is no node editor and no conditionals to write. You describe the outcome, for example “when a meeting signals a real deal, open an opportunity on the right account,” and the agent generates an automation with a name, a description, a trigger, and an ordered set of steps.
Each step is carried out by the agent with access to your meetings, emails, accounts, and opportunities, plus your team’s Skills and knowledge, so a step can work from more than the fields in a trigger payload. In the example above, the step reads the meeting transcript, decides whether there’s buying intent, matches the account, checks whether an opportunity already exists, and creates one with an amount and stage.
Drafts and activation
Section titled “Drafts and activation”Automations are first created as a draft by the agent in chat. You can test run and edit a draft iteratively until it does what you want. Once it’s drafted and tested, you can activate it, and later pause or update it with further changes.
Triggers
Section titled “Triggers”Every automation runs on a set of triggers it listens for.
- Scheduled triggers fire on a cadence: every morning at 8am, weekly, monthly, or a custom schedule, all timezone-aware.
- Event triggers fire when something happens in your CRM: an account is created, an opportunity is updated, a watched field value changes.
- Webhook triggers fire on an inbound HTTP event: a customer is created in Stripe, a message is received in LinkedIn, an internal tool posts a payload.
When triggered, an automation starts a new run. In each run the agent follows the step-by-step instructions in a series of turns until completion. Steps execute in a shared “thread,” so later steps can see what earlier steps produced without any manual wiring.
The agent watches each step for completion and marks one failed if it detects an error. You and the agent can both look back over any run (from the run details page or from chat) to troubleshoot and improve it together.
Runs execute to completion without interruption. You cannot follow up or approve actions mid-run. Human-in-the-loop review happens after a run, through For review.
Automations can use the same tools as chat, plus the Lightfield API. Across its steps, an automation can:
- Create and update records, Skills, and knowledge;
- Generate objects like tasks, notes, and files;
- Use connectors, Skills, and knowledge;
- Make network requests and use workspace secrets for authentication;
- Draft emails for user review.
Permissions
Section titled “Permissions”Before testing or activating an automation, the agent asks you to grant permissions for the tools each step will use. Permissions are defined per step: an explicit list of exactly which tools, connectors, APIs, and integrations that step is allowed to use, each with a reason. Anything outside that list is blocked. You see the full set of permissions for all steps and approve it before the automation can run on its triggers.
Run as
Section titled “Run as”Automations can run as a generic workspace member (giving the run only workspace-shared context from records, objects, and activity) or run as a specific user with their unique context.
The specific user can be predetermined or variable based on the trigger (for example, the opportunity owner when triggering on an opportunity-created event). When an automation runs as a specific user, it uses their permissions and connected accounts, and the run becomes private to them and to workspace admins.
For review
Section titled “For review”When automations run, they can automatically create or update records and objects in the CRM, or they can draft them for review.
For review is where people approve or dismiss what automations draft. Drafts produced by an automation run are assigned a reviewer by the agent when they’re generated and show up in that reviewer’s For review queue. A member’s most recent For review items appear at the top of Up next, alongside their meetings and tasks, and the full list is available from the For review page in the global sidebar, a dedicated queue of pending work produced by automations, where items can be approved or dismissed.
For now, email drafts are the only available review item type: an automation can draft an email and route it to a reviewer to approve and send. Record creation, record updates, and more are coming to For review.
For how drafts are routed to reviewers and what approving or dismissing one does, see How For review works.
Where to go next
Section titled “Where to go next”- Building automations: triggers, steps, permissions, run as, and the draft-test-activate lifecycle.
- How automations work: execution model, reliability, and operational guarantees.
- How For review works: the human-in-the-loop surface for approving the drafts automations produce.