# Connecting tools (email, calendar, CRM, browser) ### Calendar, CRM, Browser) The agent is much more useful once it can read your email, see your calendar, and touch your CRM. Connecting tools is also where most beginners hurt themselves. This chapter is the right way. ### The general principle: read-only first When you connect a tool, default to read-only access. Let the agent see, not change. Once you trust the loop, expand to write access. For email: read all inboxes, draft mode only — never auto-send. You review and click send. For calendar: read everything, create events with confirmation. For your CRM: read everything, write to dedicated fields you have approved. For your bank or finance tools: read-only forever. The agent does not need write access to your money. This is the "hard guardrail" move — restricting what the connection allows in the first place, so the agent can't do the dangerous thing even if it tries. See Ch 16 for the full framing on hard vs soft guardrails and when each one fits. ### How to connect a tool You don't run an integration script. You don't paste API keys into chat. You just tell the agent which tool you want connected, and let it handle the rest. The exact phrase to use: Connect [Tool X] for me. Do what you need to do so you never forget you have access to this tool. That's it. The agent will figure out how to authenticate the tool, store whatever it needs to store, read the docs, build itself a skill for using it, and report back when it's ready. Every future session will know the tool exists. If the tool needs you to do something on your end — log in once in a browser, grab something from a settings page — the agent will tell you exactly what to click. You don't have to know any of this in advance. The user move is one sentence per tool. The agent does the technical work. ### When a tool has no clean way to connect Some tools don't have an official way for the agent to talk to them, or hide it behind an enterprise plan. Three options, in order of preference: 1. Ask the agent if a community wrapper exists. A lot of tools have third-party connectors built by people online. The agent can check, install one, and use it. Works most of the time. Note the risk: it can break when the tool updates. 2. Browser-drive it. Give the agent the login. Tell it: "log into this tool, explore every page and button, then save a skill that drives it like a user would." Slower than an API but real. 3. Skip the tool. Some tools just aren't connectable yet. Better to skip than to wire something that breaks weekly. Before a connection session, log into every tool you want connected. Open all the tabs. The agent will need to read each one. With prep, you can connect ten or fifteen tools in 90 minutes. Without it, maybe five. ### Give the agent its own workspace Two ways to connect to your existing accounts: Option A: connect the agent to YOUR account. Give it scoped permissions on your real Gmail, Calendar, Drive. Wrap the sensitive actions in a rule ("never send an email without my permission"). Option B: give the agent its own account. A burner account it owns. You share your real docs and calendar into it. The agent can act freely there without touching your primary account. Option B is the lower-anxiety default for new users. The burner is the agent's own desk. Calendar invites, drive folders, sent mail — all isolated. Your real account stays untouched. ### A real stack A typical stack for someone running their business through the agent: notes and SOPs in a docs tool, email and calendar through a workspace product, a meeting recorder on every sales call, analytics for every site and channel, a CRM, and the messaging app you already use. The agent scans all of it on an interval. Cheap-model sub-agents in parallel bring the data back. The main agent drafts your follow-ups in your voice. You skim and say "send." Yours will be different. The pattern carries: read everything, draft everything, you press send. ### Browser automation: skip when you can The agent can drive a real browser — log in, click, fill forms, scrape pages. Use it sparingly. APIs are stable; browsers break when sites change, captchas appear, or sessions expire. If a tool has a clean way to connect, use that. Browser automation is the fallback. When you do use it, the same "read first" rule applies — have the agent observe the page, take notes on what it sees, and only then build the automation. ### A near-free CFO A worked example that pays for itself the first time you run it. Most people have no real idea what they're spending. Software bloat, duplicate tools, free trials that quietly turned into $40 a month, vendors that bumped prices and hoped nobody noticed. A real human CFO is $5k+ a month. A read-only bank feed plus your agent is a couple bucks a month. Connect a read-only bank feed — there are services that give your agent visibility into your bank and credit card transactions for around a dollar fifty a month, with no payment power and no write access. Then ask the agent to: Audit every recurring charge. Suggest free or open-source replacements for paid tools. Categorize spend by function. Catch anomalies. Build you a plain-English monthly P&L. Forecast your cash position. Prep you for negotiations. Two-step setup: connect the feed, then ask the agent to make a monthly summary skill. Run the skill on the first of every month. Done. Most people find three forgotten subscriptions the first run, and pay back the setup in under a minute. ### Other high-leverage integrations A short list of integrations that pay back fast: A meeting recorder. Every sales call, every coaching call, every internal meeting. The agent now has perfect recall of everything you have ever said in a meeting. A unified messaging view. Email, your messaging app, your sales tool — all readable by the agent. The follow-up sweep gets ten times more useful. A site-analytics tool. The agent scans traffic and conversion data weekly and tells you where to focus. A heatmap tool on your site. A free heatmap product can show what users actually do on your site — the agent can read those over time and suggest conversion-rate fixes. A document store. Whatever you use for notes (Notion, a plain folder, anything). The agent writes its own SOPs and references back to them. You do not need all of this on day one. Add one tool a week. Each one compounds. How to plug the agent into the tools you already use without any of the technical work.