How an insurance agent gets started with Claude
- Install the Claude desktop app. Go to claude.ai/download and install the desktop app for your computer. Cowork runs inside this app, not in a browser tab, so the desktop version is required.
- Start on a paid plan. Create an account and start with the $20 a month month-to-month option. You can upgrade to the $100 or $200 plan later. In Ben's experience the $20 plan is enough for many agents to begin with.
- Open Cowork, not regular chat. When the app opens it starts in the regular chat tab. Click the Cowork button to switch into Cowork. This is the feature that plans and does real work, so make sure you are in it before you start.
- Create and select a Claude folder. Make a new folder on your computer named Claude. Cowork saves the Word docs, spreadsheets, and other files it creates into this folder. Add subfolders for individual clients or projects so output stays organized.
- Connect the tools you already use. Click Customize on the left side and enable the connectors you want, such as Gmail or Google Drive. For Gmail, click enable, sign in with your Google or Workspace account, and click connect. It is that simple.
- Start a task and let it work. Click Start a new task and paste in a lead or client website, then ask for what you need, such as the description of operations, locations, and a Word document summary. Cowork builds its own checklist and does the work for you.
What Claude Cowork actually is
Claude is Anthropic's AI assistant, and Cowork is a specific feature inside the Claude desktop app. If you have used Claude before but never clicked the Cowork button, you have only seen the chat side of it. Cowork is the part that does work for you rather than just answering.
The simplest way to picture the difference is a line cook versus a chef. A chatbot like ChatGPT or Gemini is a line cook. You hand it the ingredients, tell it the recipe, and it does that one task and hands back a file or some text to copy and paste. Claude Cowork is the chef. It plans what needs to happen, gathers the ingredients itself by researching and reading your files, does the work, and figures out problems along the way by reasoning back and forth with itself.
Inside Cowork you will see three parts. On the left is the new task button and your history of past tasks. In the middle is the chat, which looks familiar if you have used any AI tool. On the right is the progress list, where Claude writes a checklist for itself and crosses items off as it completes each one. That is what makes it feel less like a search box and more like a coworker who tells you what they are doing.
How Cowork is different from ChatGPT and other chatbots
Most chat tools live in a browser tab. Some have desktop apps, but they do not reach into your files the way Cowork does. Claude runs directly on your computer, which changes what it can do. You can run several agents at the same time, as many as your machine can handle, so one can be researching leads while another drafts emails.
Cowork also connects to your files in real time. If you have ever pulled a Google Doc or Sheet into ChatGPT, you may have noticed it takes a static snapshot, so changing the file later means pasting the new version back in. Claude works with the live, updated file instantly.
Cowork also talks to itself and keeps improving as it works. When something does not go as planned, it reasons through the problem and adjusts, the way a chef handles a setback during dinner service. That self-correction is where much of the real power comes from for day-to-day agency work.
The categories of agency work it can take over
The most useful starting point is lead research. Hand Cowork a website for a lead you already have and it will research the company, analyze the risk exposures, recommend the lines to quote, and generate a structured lead-prep document on your computer. In Ben's demo it researched a commercial photographer's website, LinkedIn, blogs, and Instagram, then produced a Word doc with the email address, a representative's phone number, business hours, industry, locations, the top three lines to quote, a description of operations, and risk exposures marked as required, recommended, or essential. At the bottom it even wrote first-call questions, like whether they carry business insurance and whether their clients specify minimum limits on certificates of insurance.
From there Cowork can draft emails. Ask it to create a Gmail draft with your thoughts on which lines a prospect should carry, and it writes a real draft addressed to the right person, already signed with your name and website, with no placeholder text to fill in. You only review and hit send.
It also handles proposals and policy-comparison work because it can read files on your computer, in Google Drive, or in Microsoft 365, SharePoint, and OneDrive. Point it at a folder, ask it to build a proposal, and it works from the real, current files. It also lets your staff do more without as much hiring or training, since you build a skill once and let the team run these digital employees themselves.
Working in bulk with parallel agents
Researching one lead is helpful, but the feature that tends to surprise agents most is running many at once. After Cowork has researched a single prospect, you can ask it to find ten more similar businesses and draft an email to each, and you add the phrase run parallel agents.
In Ben's demo he asked for ten parallel agents to find other commercial photographers and draft an email to each in Gmail. Cowork did one big search to assemble a list of ten, then launched ten sub-agents that researched each photographer, found contact details, and wrote a tailored draft. Within minutes there were ten drafts in Gmail, each genuinely different and referencing that photographer's actual clients, not a copy-pasted template.
This is prospecting at scale. Point it at your existing book of business to find more leads that look like your best clients, then have it prepare the outreach while you do something else.
Connectors: linking Claude to the tools you already use
Connectors are what let Cowork reach into the software you already run. The list includes Gmail, Google Calendar, Google Drive, HubSpot, Slack, Figma, Monday.com, Gamma for presentations, and on a Mac even iMessage. There is also a connector that controls Chrome, so Claude can navigate browser tabs and pull information from sites or databases that sit behind a login.
A good example is scheduling. Ask when you are free to meet a prospect next week for two hours in person, and Claude checks your calendars for an open slot. Tell it you also need thirty minutes of drive time in each direction and that you care about traffic, and it combines your calendar with live research to suggest the best time, then drafts the email or text to send.
To see exactly what a connector can do, search for it in the connectors list and read its available tools, because connectors have limits. HubSpot, for instance, currently lets Cowork read and search your CRM but not write data back in, so it is mostly for chatting with your data, pulling lists like everyone you have not contacted in ninety days and turning that into a spreadsheet or Gmail drafts. Gmail can create and read drafts but does not yet send on its own. New connectors roll out roughly every couple of weeks.
Skills: turning a task into a repeatable recipe
Claude calls them skills. Ben calls them recipes, because that is what they are: a saved set of instructions so a task runs the same way every time, like grandma's recipe written down so anyone can follow it. The lead-prep document in the demo came from a skill Ben built, which is why a single prompt with a company website triggered the whole research-and-document process automatically.
Building one is genuinely easy and requires no coding. You do the task manually once in Cowork, then type make this into a skill. Claude runs its own skill-creator, walks through the steps with you, bundles everything up, and gives you a button to save it to your skills. From then on you can rinse and repeat.
This is the payoff for a non-technical agency. Once a good lead-prep or email-drafting skill exists, anyone on the team can run it and get consistent output. You are not retraining people every time, you are handing them a reliable digital employee.
What it costs and how to start small
The cost is modest for what it replaces. Claude offers $20, $100, and $200 per month plans. Ben recommends non-technical agents start on the $20 month-to-month plan and upgrade only if they need to. He runs heavily on the $100 plan and has not maxed it out, and many of his clients find the $20 plan is plenty to begin. Counting Google Workspace, his all-in cost is roughly $120 a month, with no large tech stack to assemble.
The smartest way to start is with one repeatable task. Install the app, open Cowork, point it at a Claude folder, connect Gmail, and run a lead-prep request on a single real prospect. Once that works, save it as a skill and add connectors as you go. WhereToBegin.ai is an independent course and is not affiliated with or endorsed by Anthropic, the maker of Claude. The goal is simple: get one digital employee working for your agency, then grow from there.