How Lobby's Prompt-Free Email Workflow Agent Works

Lobby's email agent works by learning a behavioral model of your judgment from the tools you already use, then predicting your email workflows end to end so you approve actions instead of writing prompts. According to the site, this requires connecting your tools first; after that, the model handles coordination and you act as the approver. The company states 92% prediction accuracy, a 5-task prediction depth, and a 7-day horizon to converge, with 80+ sub-agents involved.

The core mechanism: model first, prompt never

Lobby frames its product as an "extension of human cognition" rather than a chat-style assistant. The distinction matters for how you use it:

• Traditional agents wait for a prompt, then act. You carry the burden of describing context every time.
• Lobby's approach builds a behavioral model that "learns, predicts, and amplifies how you think and act," per the site. The agent then acts with what Lobby calls "full fidelity to who you are" — without you prompting it.

The site describes the artifact as a behavioral model that is user-owned and portable across systems. In time, Lobby says these models will coordinate with each other to amplify collective intelligence. For day-to-day use, the practical takeaway is that the value depends on the model having enough signal about your workflows — which is why connection comes first.

What you actually do: connect, then approve

The workflow has two user actions and one system action:

1. Connect your tools. Lobby says it coordinates email workflows "across the tools you already use." The site lists email and calendars as covered, with a 50+ tools addon. This is the input stage — the model needs visibility into your existing email and calendar activity to learn your patterns.
2. The model predicts your workflows end to end. No prompt is required. Lobby describes the output as predicted email workflows that span multiple steps, not single replies.
3. You approve. The site's phrasing is direct: "No prompt, you just approve." Your role shifts from instructing the agent to reviewing its predicted actions.

What "prediction depth" means here

Lobby states a 5-task prediction depth and a 7-day window to converge. Read together, this suggests the model plans several linked tasks ahead rather than one action at a time, and that it takes about a week of signal to reach its stated operating accuracy. The site also cites 80+ sub-agents, which implies the work is decomposed across many specialized components coordinated by the behavioral model rather than handled by a single agent.

The numbers Lobby publishes — and how to read them

These are vendor-stated figures. The site does not publish the evaluation method, sample size, or what counts as a "correct" prediction, so treat 92% as a claim to validate against your own workflow during the convergence week rather than a guaranteed outcome.

Who this fits — and the conditions

Lobby's own framing targets people whose email load exceeds what prompting-based agents can absorb: the site argues that "99% of the world cannot sustain this load" of tracking thousands of relationships, decisions, and contexts, and that agents requiring prompts and system maintenance don't free bandwidth.

This approach is a reasonable fit if:

• Your email work is repetitive and pattern-based — coordination, scheduling, follow-ups — so a model has something learnable to predict.
• You are comfortable approving rather than composing — the workflow assumes you'll review predicted actions, not write instructions.
• You can connect your email and calendar to the product, since the model has no signal without that step.
• You want a portable, user-owned model — Lobby states the behavioral model is owned by the user and portable across systems.

It is a weaker fit if your email is highly novel each time (little pattern to model), if you need to keep tools disconnected, or if you require published accuracy methodology before adopting.

Pricing and access

The site lists pricing from $42/month, including unlimited emails and calendars, with a 50+ tools addon. It also shows Login and Try Lobby entry points and iOS/MacOS availability. The page does not state whether a free tier or trial exists, so confirm access terms directly before committing.

The bottom line

Lobby replaces prompt-writing with approval: connect your email and calendar tools, let the behavioral model predict multi-step workflows, and review what it proposes. The published figures — 92% accuracy, 5-task depth, 7-day convergence, 80+ sub-agents — describe the intended experience, and the first week is when you can judge whether the model's predictions match your actual judgment.