Lindy
A personal AI teammate that triages your email, takes meeting notes, and books itself onto your calendar.
Your inbox has 180 unread messages. Forty need responses; the rest are informational or spam. Before reading the first one, you already know that you'll spend two hours doing this and finish unsatisfied. What if an agent could read all 180, draft replies to the forty, and summarize the rest?
Why this tool matters
Lindy is a platform for building personal AI agents that take real actions — read email, join meetings, search the web, update CRMs, book calendar time — on your behalf. Unlike chatbots that respond when you ask, Lindy agents sit running in the background, triggered by events (a new email arrives, a meeting starts, a Slack message is sent) and take their assigned actions automatically.
The design insight: most knowledge work is pattern-following. The same kinds of emails show up every day and want the same kinds of responses. The same meeting-prep motion runs before every customer call. The same CRM-update ritual happens after every conversation. Lindy lets you describe the pattern in plain English, connect the tools it needs, and deploy an agent that handles the pattern while you do work that requires actual judgment.
For a researcher, a Lindy agent can monitor new papers in specific journals and email you summaries weekly. For a consultant, an agent can draft responses to prospect inquiries overnight, save drafts to a review queue, and only escalate the genuinely complex. For a solo business owner, an agent can triage the whole inbox every morning and present a 3-line briefing with the messages that actually need you. This is the frontier of personal AI — where the tool starts to feel less like software and more like staff.
Setup
Account: lindy.ai free tier gives you 400 tasks/month and access to the full agent builder — enough to deploy one real agent and see it work. Pro ($49.99/mo) scales to 5,000 tasks; Business ($299/mo) adds team features and phone integration.
Prereq: the tools you want the agent to work with (Gmail, Google Calendar, Slack, HubSpot, Airtable, etc.) need to be integrated via OAuth. Have those accounts ready and be willing to grant Lindy the permissions its agents require — or pick tasks that stay inside Lindy's sandbox.
Walkthrough
Step 1: Pick a template for your first agent
At lindy.ai, open the Template Gallery. Templates cover the common patterns: email triage, meeting notes, lead qualification, scheduling, research monitoring, inbox auto-draft. Pick one that maps to a real pain you have this week. Don't try to design a custom agent from scratch on your first try.
Step 2: Connect the tools it needs
Lindy asks which accounts to link (Gmail, Calendar, etc.) before it can run. Grant only what the template requires. Read the permissions scope carefully — an agent that “drafts replies” is different from one that “sends replies without review.” Start with drafts-only.
Step 3: Customize the agent's instructions
Every Lindy agent has a plain-English instruction block. Edit it to match your voice and rules: “For emails from existing clients, draft a warm reply that references our last conversation. For prospecting emails, draft a polite decline. For everything else, flag for my review.” Be specific. Specificity is the difference between a useful agent and a noisy one.
Step 4: Run in sandbox mode first
Lindy offers a test mode where the agent processes real inputs but doesn't take real actions — drafts stay as drafts, messages get flagged but not sent. Let it run for a full day in sandbox. Review everything the agent would have done. Correct the instructions based on what you see.
Step 5: Enable live mode — but keep review
When the sandbox results are 80%+ right, switch to live. For high-stakes actions (sending emails, creating calendar events, updating CRM records), keep a human-in-the-loop for the first month: the agent drafts, you approve. Only graduate to autonomous action after you've audited a few hundred examples.
Step 6: Stack agents for bigger workflows
Lindy agents can hand off to each other: an email-triage agent flags a meeting request, a scheduling agent proposes times, a CRM agent logs the booking. Each agent stays narrow and explicable; the combination handles workflows no single agent could. This is where power users live.
Your turn
Basic: An inbox triage agent
Set up an email-triage agent using a Lindy template. Customize the instructions to reflect your real rules (what gets drafted, what gets flagged, what gets archived). Run in sandbox for three days and review every proposed action.
After three days, decide: are the proposals 80% right? If yes, graduate to live-with-review. If not, refine the instructions and sandbox another three days. The instructions are the product — invest here.
Advanced: Build a 3-agent workflow
Design a real business workflow that requires multiple agents coordinating. Example: (1) an email-triage agent classifies incoming messages, (2) a meeting-prep agent produces a briefing whenever a calendar event is about to start, (3) a follow-up agent drafts a thank-you email within 30 minutes of a meeting ending, referencing action items from Otter (Day 12).
Build each agent. Test in sandbox. Wire them together so output of one triggers the next. Run the full chain on a real day of work.
Write a 250-word retrospective answering: (a) how many hours did the chain save, (b) what failed and why, (c) where is judgment still required, (d) what would you automate next given this proves out?
This exercise is the gateway to operating with a personal AI workforce. Most people never cross it — the ones who do have hours per day back.
Pitfalls and pro tips
Autonomous action at scale is dangerous. An agent that sends emails on your behalf will eventually send something wrong to someone important. Keep a human review step for any outbound communication until the agent has proven itself on hundreds of examples — and even then, random-sample audits.
Permissions creep. Agents accumulate access over time: Gmail, Calendar, Slack, HubSpot, Notion, Drive. Each new permission is a new attack surface. Quarterly, review which agents have which permissions and revoke anything the agent no longer needs.
Instructions drift from reality. Your business changes; your agent's instructions often don't keep up. Every month, spend 15 minutes re-reading each agent's instructions and updating them to reflect how you actually work now. Agents without maintenance become noisy, then wrong, then harmful.
How it compares
Lindy's competitors include Zapier AI (Course 1; lighter-weight, trigger-action focused, less agent-like), n8n (self-hostable, developer-facing), Make.com (similar to Zapier), and Relevance AI (more technical, better for building custom multi-step agents). Lindy's edge is the focus on personal productivity agents — agents that sit between you and your work, not between systems. For technical users building complex automations, Relevance AI is stronger. For business users delegating email and meeting work, Lindy is unusually well-designed.
When to use — and when not to
Use Lindy when you notice yourself doing the same motion repeatedly every day or week — inbox triage, meeting prep, CRM updates, lead qualification, research monitoring. The pattern is the signal that an agent can help.
Do not use Lindy when the task requires substantial domain judgment (it will produce confident wrong answers), when compliance forbids third-party access to sensitive data, or when the volume doesn't justify the setup cost (one email a week is not worth an agent).