In This Video
- 0:00Intro
- 0:16Building The Sales Board
- 1:37Setting Up The Sales Stages
- 4:32The Columns The Agent Needs
- 8:27Adding A Test Deal
- 9:43How The Agent Matches A Transcript To The Right Deal
- 10:45Opening Agents In The Claude Console
- 12:20Setting Up The Google Drive Folders
- 15:05Dictating The Agent's Instructions To Claude
- 19:14Turning Your Proposal Into A Skill
- 21:43Generating The Agent
- 22:05Choosing The Model & Effort Level
- 23:54Connecting monday.com & Google Drive
- 26:03Tool Permissions
- 28:17Adding The Skill To The Agent
- 30:36Advisors, Sub-Agents & Workflows
- 33:25Creating The Environment
- 35:01Setting Up The Deployment
- 36:46Triggers: Manual, Schedule Or Webhook
- 38:28Running It On A Test Transcript
- 40:01Results: $2.76 & 6 Minutes
- 41:45What The Agent Flagged For You To Fill In
- 42:57Refining It & The Time It Saves
- 44:17Wrap-Up
How Do You Set Up A Simple Sales Board In monday.com?
Create a board for your deals with a group for each sales stage and a status column whose labels match the groups. The agent uses it to find the deal each sales call belongs to and to store a link to the proposal.
Ours is called Sales, with items called Deals and groups for New Deals, Proposal Sent, Follow Up, In Negotiation, Closed Won and Closed Lost. New Deals holds every deal still waiting on a proposal, Follow Up catches deals that go cold after one is sent, and New is the default status.
Which Columns Does The Proposal Agent Need?
Five: the contact's name and email, a proposal sent date, a proposal link and a transcript ID. If you already have a sales board, these are all you need to add.
- POC Name, a text column, and POC Email, an email column, for the deal's point of contact. The agent checks the transcript against both.
- Proposal Sent Date, a date column the agent fills in.
- Proposal Link, a link column, which works better with an AI agent than a files column. The proposal stays in Google Drive, and the link opens it for review and edits.
- Transcript ID, a text column for the agent rather than the sales team. Zoom and Teams give each meeting's transcript a unique reference, and this column links it to the deal.
Everything else is optional, such as a deal value, automations that move deals between groups as the status changes, or a date closed column for average time to close.
How Does The Agent Match A Transcript To The Right Deal?
By the contact's email first, then their name, then the company name. The agent saves the proposal to Google Drive before it looks for the deal, so it needs reliable reference points.
The name works as long as the contact was invited to the call under it, and the company is likely to come up in the transcript. If nothing matches, which should be rare, we'd add an escalation saying the proposal exists but its deal can't be found, perhaps because it's missing from the board.
For the test, we added one deal for a sample company, with a contact name and an example email.
What Do You Need Before You Build The Agent?
Admin access to the Claude Console, where the agent is built, and three Google Drive folders for it to work in. Create them first, because the instructions must tell Claude where each file lives and where it goes.
We use Transcripts, a Completed Transcripts folder inside it, and Proposals. Each new transcript gets a proposal saved to Proposals, then moves to Completed Transcripts.
The plan is for a Zapier webhook to add each finished Zoom, Teams or Google Meet transcript to Transcripts, though that isn't built in this video. OneDrive works as well as Google Drive.
What Should The Agent's Instructions Say?
What the agent is responsible for, where the files are and which columns to update. The Create agent screen under Agents has no dictate option, so we talked it through in a Claude chat and had Claude write the instructions to paste in.
Ours tell the agent to check Transcripts every 15 minutes and turn any new transcript into a proposal using the skill. It then saves the proposal to Proposals, moves the transcript to Completed Transcripts, finds the deal on the Sales board and fills in the proposal link and date.
We told Claude a skill was coming so the names would match. With monday.com connected to our Claude account through MCP, Claude also checked the board and filled in the column we'd forgotten to mention, Transcript ID.
How Do You Turn Your Proposal Into A Claude Skill?
Upload an example of the proposal you already send to a Claude chat and ask it to create a skill. A skill is a tightly defined set of instructions the agent follows every time, so each proposal comes out the same way.
Ours used made-up data and has an executive summary, what we heard, the proposed solution, deliverables, timeline and investment. The skill sets out how the proposal is built, which details go in and how it's formatted.
Which Model & Effort Level Should The Agent Use?
The console defaulted to Opus 5.5 on low effort, probably more than this job needs. A more complex agent needs a more capable model, which costs more, so test cheaper ones and refine as you go.
Weigh that against the alternative. A couple of dollars for 90% of a proposal is small next to maybe $20 to $50 of your team's time writing it, time they could spend selling.
We pasted Claude's instructions into the agent's summary and pressed Generate, which wrote the name, description and system prompt. The prompt came out much shorter than Claude's version, probably on purpose, because the agent reads it every run and a longer one gives it more chances to get confused.
Which Tools & Permissions Does The Agent Need?
Google Drive, to read transcripts, save proposals and move files, and monday.com, to update the deal, both as MCP servers. Generate added both, and connecting each means picking a vault for the credentials and authorising it.
Each tool action can be set to Auto, Always allow, Always ask or Always deny. Always ask makes the agent wait for you every time, which defeats the point of an autonomous agent. We recommend Always allow for nearly everything, switching off what it doesn't need, such as creating workspaces or folders.
You can't limit the monday.com MCP to certain boards. It reads everything or nothing, and with nothing the agent can't work.
How Do You Add A Skill To An Agent In The Claude Console?
Upload it to the console, then attach it to the agent. Generate had added a skill that doesn't exist, so remove that first.
- Download the skill from the Claude chat. Skills come as a zip file.
- Open the skills page in a new tab (right click, then Open Link in new tab). The agent isn't saved yet, so leaving the page loses your work.
- Press Create skill, choose the zip file and confirm with Create Skill.
- Back on the agent's tab, press Add Skill and choose the sales proposal skill. If it isn't listed, a trip to the skills tab and back usually refreshes it.
Does A Proposal Agent Need Advisors Or Sub-Agents?
Probably not, though we left an advisor on for this one. Like a line manager, it's a second model, ideally more capable, that the agent asks when it gets stuck on something its instructions don't cover.
Sub-agents and workflows hand parts of the job to other agents, either ones you've saved or ones the agent creates itself. This agent needs neither.
How Do You Deploy The Agent & What Should Trigger It?
Give it an environment to run in, then a deployment that sets how each run starts. Use Claude's cloud unless you have your own servers, and for this agent leave networking unrestricted rather than limited to what you allow.
The deployment picks the environment, adds the vault as a resource and sends an initial message telling the agent to begin work. A budget can cap spend per session, such as $10, but it's optional and we left it off.
A schedule, such as hourly or weekdays at 9am, suits the folder-checking setup and keeps the trigger inside the console. We'd rather have a Zapier webhook start it as soon as a new transcript lands in Google Drive. For the video we ran it manually.
What Does The Proposal Agent Cost To Run?
$2.76 for one proposal, and the run took 6 minutes 12 seconds. Here's how we tested it:
- Upload an example sales call transcript to the Transcripts folder.
- Open the agent's Deployments, click into the deployment and press Run now.
- Press View sessions and click into the session to watch it work in real time.
It did nearly everything we asked without anyone touching it. The transcript moved to Completed Transcripts, the proposal landed in Proposals, and the deal now shows the proposal sent date, transcript ID and link.
The proposal followed our template, with what we heard and the executive summary drawn from the call, but the colours and some formatting differed, which means changes to the skill and instructions, and pricing was left to confirm.
What Does The Agent Leave For A Person To Fill In?
Anything it couldn't get from the call. Its end-of-run summary listed what to fill in before the proposal goes out: prices, a security summary promised on the call, and monday.com certification details and links, which weren't discussed.
It also listed gaps such as the project start date and number of training sessions, and explained its timeline: 11 weeks, inside the 10 to 12 weeks quoted on the call.
A person still checks and finishes every proposal before it goes out. That summary could also be added to the deal in monday.com, so the team can see what's done and what isn't.
How Much Time Does It Save & How Do You Refine It?
Proposal turnaround could drop from two days to a couple of hours. A Zoom transcript arrives about 20 minutes after the call, depending on its length, and the agent takes about six minutes, so a draft can be ready within about 30 minutes for your team to review, refine and send.
Getting there takes testing: use the agent's feedback to change the instructions or skill, redeploy and run it again. Every workflow is full of small things people do without realising, and each one has to be written down before the agent can do it too.