Loom AI: Screen Recording, Workflows, and Agent Prompts

Loom is moving beyond asynchronous video messaging. On July 22, 2026, Atlassian introduced Loom Video Prompts, an open-beta recording mode that turns a narrated screen walkthrough into structured instructions for an AI agent. Instead of writing a long prompt, collecting screenshots, and manually listing every task, a user can demonstrate the work on screen, explain the intended result, and let Loom assemble an Action Plan for review.

This is a meaningful product shift. Loom has long helped people explain work to other people. Its newest workflow is designed to help people explain work to software. The recording remains the source context, but the useful output is no longer only a video link, transcript, summary, or chapter list. It can become a structured plan that an agent can interpret or a set of Jira work items that a team can approve and assign.

### How Video Prompts work

Atlassian's [launch announcement](https://www.atlassian.com/blog/loom/introducing-loom-video-prompts) describes a simple interaction: record the screen, talk through what should be done, and receive a structured action plan with richer instructions than a normal text prompt. The [official support guide](https://support.atlassian.com/loom/docs/record-video-prompts-for-ai-agents/) provides the important operating details.

Video Prompt mode currently lives in the Loom Chrome extension. A user opens the Generate tab, chooses Video Prompt, shares the relevant screen, and narrates the desired work while navigating the interface. When recording stops, Loom creates an Action Plan on the share page. The creator can review and copy the plan into another agent, or eligible Atlassian-managed users can turn selected actions into Jira work items.

The recording is limited to five minutes. Loom automatically disables the camera so the agent's context stays focused on the shared screen. This is not simply audio transcription. The mode can capture the spoken transcript, visual keyframes, interaction events such as clicks or hovered areas, and the links visited during the walkthrough. The resulting actions may include a concise title, bulleted description, supporting transcript quote, visual reference, and timestamp.

That combination can be much richer than a prompt written from memory. A product manager can move between a design reference and a live product while explaining a specific interface change. An engineer can point at the failing state, identify the relevant setting, and clarify what must remain unchanged. A marketer can demonstrate how content should be updated across a workflow while naming brand constraints. The agent receives evidence and sequence, not just an abstract instruction.

### Availability is narrower than the headline

Video Prompts are not a general free feature at launch. Atlassian says the beta is available to Loom Business + AI and Enterprise users. It currently requires the Chrome extension. The direct flow for creating Jira work items is restricted to Atlassian-managed accounts with an active Jira subscription and the necessary permissions. Other users can copy the reviewed plan into a third-party agent.

The open-beta label matters. Teams should expect the interface, outputs, eligibility, limits, and integrations to evolve. A beta is suitable for controlled testing and approved workflows, but organizations should avoid making it the only way a critical process is documented. Keep the original requirement, recording, reviewed Action Plan, and final Jira or agent result together until the workflow proves dependable.

The five-minute limit is also a design signal. Video Prompts work best when the task is focused. A long recording that mixes a website redesign, database change, support workflow, and marketing request may produce actions that look organized while hiding conflicting priorities. Split complex work into separate prompts with one outcome, one relevant set of screens, and explicit boundaries.

### The agent still needs a human-approved brief

The Action Plan should be treated as a draft, not an authorization. Loom can capture what was said and shown, but it cannot automatically know whether every instruction is correct, whether the speaker has authority, whether an opened tab exposes confidential information, or whether an agent should be allowed to perform a destructive step.

Before sharing a plan, verify every file name, URL, account, visual reference, acceptance criterion, owner, deadline, and prohibited change. Remove credentials, private customer data, internal messages, unrelated browser history, hidden tokens, and any information the agent does not need. Define whether the agent may only analyse, propose, edit a draft, create a branch, open a Jira item, or make an external change. High-impact actions should remain behind explicit approval.

The screen itself creates a new privacy surface. Atlassian's support page says Video Prompts can capture keyframes, interaction events, and visited links. That is useful context, but it can also reveal more than the spoken transcript. Teams should prepare a clean browser profile or test environment, disable notifications, close unrelated tabs, use non-production records where possible, and inspect the generated plan and attachments before they leave the workspace.

### Loom's broader AI workflow is now more practical

Video Prompts build on a wider Loom AI product that has become more than automatic summaries. The current [Loom AI page](https://www.loom.com/ai) lists automatic titles, summaries, chapters, tasks, calls to action, filler-word removal, silence removal, meeting notes, recap emails, and video-to-text workflows. A recording can become an SOP, step-by-step guide, pull-request description, QA process, code document, message, Jira issue, or Linear issue.

This is most valuable when a person already knows the process but struggles to write it down. Demonstrating a task while explaining exceptions can be faster and more natural than starting with a blank document. Loom then gives the team a structured draft to refine. The important word is draft. An SOP created from one clean demonstration may still omit permissions, alternative paths, failure recovery, legal controls, accessibility, data handling, escalation, or changes that occurred after recording. The process owner should test the document before it becomes a source of truth.

Meeting notes and recap emails extend Loom into synchronous work as well. Business + AI and Enterprise currently include automatic meeting notes, action items, and follow-up. That can reduce manual note-taking, but the output should be confirmed by participants. A mistaken owner, date, number, or decision can create more confusion than no notes at all. Recording consent and access controls must also be handled before the meeting begins.

For software teams, Atlassian has been deepening Loom's Jira role. Enhanced bug-reporting mode can capture technical context such as device information, console logs, and network information, while AI-assisted flows can create or update work items from video and meeting context. This is useful because many bugs are difficult to describe without showing the actual behavior. It is also sensitive, because logs can contain user data, request payloads, internal endpoints, or tokens. Reproduce issues in a safe environment and review every captured field.

### Current plans and pricing

Loom's [official pricing page](https://www.loom.com/pricing) currently lists four plans. Starter is free. It allows up to 25 recordings per person, limits ordinary screen recordings to five minutes, supports video up to 720p, includes unlimited meeting recording length, and provides transcription in more than 50 languages. Automatic titles are currently included, but the richer AI suite is not.

Business is listed at $18 per user per month. It removes the normal recording and video-count limits, supports quality up to 4K, adds trim and stitch editing, video uploads and downloads, branding controls, password protection, workspace privacy defaults, search, engagement insights, and broader content management. It suits teams that need reliable video communication but do not require the full AI workflow.

Business + AI is listed at $24 per user per month and includes the features most people mean when they search for Loom AI. It adds automatic summaries and chapters, tasks, advanced editing, transcript-based edits, filler-word and silence removal, video-to-text automation, meeting notes and recaps, video variables, admin insights, and Video Prompts. New eligible workspaces can try the tier for 14 days.

Enterprise uses a custom annual quote. It adds stronger administration and security, including Atlassian Guard Standard with SSO and SCIM, advanced content privacy, custom retention, account transfer and ownership controls, activity reporting, Salesforce integration, a 99.95% uptime service-level agreement, and dedicated support. Loom says annual billing can save up to 17% on eligible self-serve subscriptions, so teams should compare the actual checkout total with the monthly headline before budgeting.

### Data handling deserves a real review

Atlassian's [Loom AI information page](https://www.atlassian.com/software/loom/ai) says OpenAI receives transcript data as text files to generate titles and summaries, while full videos and audio are not sent to OpenAI for those functions. Atlassian says its third-party subprocessors are subject to privacy and security controls and data-processing agreements.

That is useful disclosure, but it is not a universal approval for every organization or every recording. Before enabling Loom AI, review the current AI terms, subprocessor list, data location, retention, deletion, model-training terms, access logs, admin settings, contractual obligations, and rules for customer or employee data. Decide which teams may use AI features, which categories of information are prohibited, and when a recording must be stored in another approved system.

A Loom can contain far more context than its title suggests. A two-minute screen recording might expose email previews, browser history, private chat, customer records, access tokens, financial information, health information, source code, internal dashboards, or unreleased product details. Use least-access sharing, name an owner, set retention, and periodically audit public links and stale workspace content.

### The best way to adopt Video Prompts

Start with one low-risk, reversible workflow. Good pilots include creating a draft Jira issue from a test-environment bug, preparing an internal documentation outline, or generating an agent brief for a small interface change on a non-production branch. Avoid financial actions, permission changes, deletions, customer communication, or production deployment until the organization has tested accuracy, privacy, and approval controls.

Create a standard recording checklist: one outcome, a clean browser, no secrets, test data, clear constraints, explicit acceptance criteria, and a statement of what the agent must not change. After recording, compare the Action Plan with the video, correct every step, and pass it to an agent that has only the permissions required for the test. Review the agent's result against the original screen evidence.

Measure more than speed. Track how often plans need correction, whether agents misunderstand visual context, how many Jira items require rewriting, whether sensitive information is captured, time saved after review, and the quality of the finished work. If the workflow produces polished plans that still cause rework, the apparent automation is not a real gain.

### The WhatAI verdict

Loom remains one of the clearest tools for explaining work visually, and the new Video Prompts beta gives that recording a new audience. The ability to combine narration, screen context, keyframes, interactions, and links into a reviewed agent brief is more useful than asking people to copy a transcript into a generic chatbot and hope it understands the interface.

The strongest use case is not autonomous execution without oversight. It is faster preparation of a richer brief. Loom can help a person show the work once, create a structured plan, and move that context into Jira or an agent workflow. Human review, access control, data hygiene, and acceptance testing are what turn that convenience into dependable work.

Choose Loom if your team already communicates through demonstrations and wants those recordings to become documents, issues, recaps, or agent instructions. Choose a basic plan if recording and sharing are the goal. Pay for Business + AI when the automated outputs save enough reviewed work to justify the per-seat cost. Look elsewhere if you need cinematic production, complex multitrack editing, local-only processing, or an AI system that can safely act without a human-defined permission boundary.