The Weekend Playbook: Become an AI Pilled VP of Sales in 48 hours
You don’t need reps who write better emails. You need to hit the number.
You don’t need reps who write better emails. You need to hit the number.
That’s the whole test as a VP of Sales.
It's not just about whether “the deck looked sharper,” or “the rep sounded smarter on the call,” or if “we shipped more content.” If it doesn’t show up in metrics like close rate, ASP, cycle time, or bookings, it isn’t really making a difference. Many sales leaders who are currently using AI are focusing on the wrong areas of this test. They’re using Claude to write follow-up emails and summarize call notes, which is okay. But, honestly, that’s just scratching the surface of what the tool can do.
Here’s the bigger one. The most valuable thing in your sales org isn’t your CRM data. It’s your judgment.
It’s the way you listen to a discovery call and know, in the first four minutes, that the rep skipped the agenda and is about to lose control of the deal. It’s the way you look at a forecast category and know which “commit” deals are actually a coin flip. It’s the way you run a 1:1 so the rep leaves with one thing to fix, not five.
That judgment is why you have the title. It’s also completely trapped in your head, available to exactly one person, for exactly as many hours as you’re awake.
Claude Skills is the mechanism to get it out of your head and into your org, without writing a line of code.
What a Skill actually is
A Skill is a folder with instructions, at least one file: SKILL.md. This file has two parts: a short frontmatter with name and description (max 200 characters) that explains what the skill does and when to use it, and plain markdown with detailed instructions. For complex skills, you can add reference files, examples, or scripts. Most of this work is writing clear, concise instructions, not software.
You turn Skills on in Customize > Skills (this requires code execution to be enabled on your plan). From there you can either upload a skill you built as a folder, or use the built-in skill-creator, which interviews you and drafts the SKILL.md for you. Once a skill exists and is toggled on, Claude decides when to use it. You don’t have to remember a command.
You ask for the thing, and if your description is good, Claude reaches for your skill instead of guessing from scratch.
The part that matters most: a skill built this way works the same whether you’re in a normal Claude chat, in Cowork, in Claude Code, or via the API.
Build it once, over the weekend, and it’s available everywhere you work from that point forward.
Step 1: Map the job before you touch a keyboard
Before you build anything, write down every discrete thing you actually do. Not “sales leadership.” The actual tasks. Use Wispr Flow or just talk into your phone and get it into a doc, then dump the whole thing into a Claude project and ask it to turn the ramble into a clean, numbered list. You’ll end up with something like this:
Reviewing overall pipeline
Reviewing a specific deal against your methodology (MEDDPICC, Command of the Message, whatever you run)
Running a 1:1
Coaching a rep after a call, based on what actually happened on the call
Auditing and improving sales assets (decks, one-pagers, email sequences)
Interviewing a sales candidate
Running an enablement session
Building an enablement deck
Forecasting: pipeline movement week over week, stage integrity, forecast category discipline, rep-by-rep behavior patterns
Rolling that into a spreadsheet and mapping close rates by stage and by time in stage
Applying your judgment to the raw number, then loading your call into whatever forecasting tool your org uses
Running the forecast call itself: which deals get pressure-tested, which ones you take on faith
Sizing the biggest deals in the funnel to put a plus-or-minus on your number
Rolling your team’s forecast call up into your own leadership forecast call
Tracking pipe generated, by rep, by week
Analyzing where pipeline is actually coming from, so you know what to double down on
Drafting follow-up emails on stuck deals
Helping a rep get unstuck: bringing in an exec, reframing the story, finding a new way into a stakeholder
That’s roughly seventeen distinct jobs. Each one is a candidate for a skill. Don’t try to build all seventeen Saturday morning. Pick the three or four that eat the most of your week, and start there.
Step 2: Build the skill by teaching, not by prompting
This is where most people get it wrong. They open a chat and type “write me a rubric for coaching sales calls.” That gets you a generic rubric. It doesn’t get you your rubric.
Here’s the actual process, using post-call coaching as the example since it’s the highest-leverage one for most sales leaders.
Open a new chat and ask Claude to use the skill-creator to help you build a skill for coaching reps after a call. It will ask you clarifying questions before it writes anything: what methodology you run, what a good call looks like to you, what “bad” actually sounds like. Answer honestly and specifically. Vague answers produce a vague skill.
Then go get a real call. Pull one from Gong, Granalo, or Fireflies. If your org has an MCP connector set up for either tool, pull it in that way. If not, just download the transcript and upload it directly to the chat; that works fine too.
Now do the part that actually builds the skill: coach the call yourself, out loud, in the chat. Not a summary. Your real reaction, sentence by sentence. “I’d flag this open, no agenda was set.” “This is the moment they should have asked a second-level question and didn’t, here’s the follow-up I’d have used.” “We never got a real next step, we got a maybe.” “This close is too soft, given where they are in MEDDPICC.”
Use Wispr Flow if talking is faster than typing. The point is volume of your actual judgment, not polish.
Ask for a V1. Now you have something concrete to react to: does this sound like the coaching you’d have given? Where’s it off? Too soft, too clinical, wrong tone, missing a step you always take? Correct it directly, the same way you’d redline a rep’s email.
Feed it more calls. Once V1 feels close, run it against ten or fifteen more transcripts and see where it breaks. A bigger sample surfaces the edge cases: the call that went well despite a shaky open, the rep who’s naturally quieter but still in control. Fold those corrections back in and you’re at V2, then V3.
You’re done when the output would pass as something you wrote. That’s the actual bar. Not “this is helpful.” “This is what I would have said.”
From here forward, any call becomes: drop the link or upload the transcript, ask for the coaching breakdown, and it comes back in your voice, using your framework, calibrated to your standards. You didn’t just save time on one call. You built a piece of infrastructure that outlives any single week.
Step 3: Put it on a schedule instead of running it by hand
Once a skill is solid, the next move is to stop triggering it manually. Claude Cowork has scheduled tasks: you describe a job once, set a cadence, and it runs on its own from there, with access to the same skills, connectors, and tools you’d have in a live session. So instead of remembering to pull each rep’s calls and run the coaching skill yourself, you set up a recurring task: pull yesterday’s calls, run the coaching skill against each one, and save the output somewhere your reps can get to it. Worth knowing going in: scheduled tasks on Desktop only fire while your machine is on and the app is open, so if you’re picturing something running silently overnight on a laptop that’s closed, check the current setup before you build around that assumption, since this is an area Anthropic has been actively shipping into.
The same pattern applies to a forecast rollup. Build a skill that takes your team’s raw pipeline export and produces the stage-integrity and coverage view you actually use in a forecast call. Schedule it weekly. You walk into your own forecast call with the analysis already done, and you spend your time on judgment, which deals are real, instead of building the spreadsheet that gets you there.
The skill stack: what a VP builds this weekend
Here’s the list from Step 1, reorganized into an actual build order. Start at the top.
Pipeline and forecast
Weekly pipeline movement and stage-integrity report, built from a raw CRM export
Coverage and velocity analysis by segment or territory
Close-rate-by-stage-and-time model, for calibrating your forecast number
Forecast call prep: flags the largest and riskiest deals for pressure-testing
Deal and rep judgment 5. Post-call coaching, in your voice, against your methodology 6. Deal review against your qualification framework, done before a deal hits forecast 7. Stuck-deal unblocking: drafts an exec outreach angle or a new stakeholder path
People and enablement 8. Candidate interview scorecard and follow-up questions, built from your actual bar 9. 1:1 prep, pulling the rep’s pipeline, activity, and recent call trends into one page 10. Sales asset critique: reviews a deck or one-pager against what actually converts 11. Enablement deck builder, seeded with your talk track and objection handling
Pipeline generation 12. Pipeline source analysis: where bookings are actually coming from, by channel and rep 13. Follow-up email drafts on stalled deals, in your tone, referencing deal-specific context
Thirteen skills is a real weekend’s work, not six hours. Build two or three well before you build thirteen badly. A sharp coaching skill and a sharp forecast-integrity skill will do more for your number than eleven mediocre ones.
If you actually want to do this in one sitting
Block six hours. Here’s roughly how to spend them.
Hour 1: Map the job. Dump every task from Step 1 into a doc, however messy. Don’t edit yet, just get the full list out of your head.
Hour 2: Pick your first skill and build it with skill-creator. Answer its questions specifically. This is the hour that determines whether the whole exercise works, so don’t rush it to get to the “fun” part.
Hours 3-4: Teach it on real deals. Pull two or three real calls or two or three real forecast exports. Coach them, correct them, react to the V1, get to something that sounds like you. This is the slow, unglamorous part, and it’s also the part nobody skips if they want a skill that’s actually worth using Monday morning.
Hour 5: Stress-test it. Run the skill against five to ten more examples you didn’t use while building it. Watch for where it breaks, where it’s too generic, where it misses something you’d have caught. Fix the instructions, not just the output.
Hour 6: Build your second skill, faster this time. The first one always takes longest because you’re learning the process along with building the thing. The second one goes quicker because you already know what “specific enough” looks like.
You won’t leave the weekend with thirteen skills. You’ll leave it with two or three that actually work, which is worth more than thirteen that don’t.
The part that’s easy to miss
None of this replaces your reps’ job, which is selling, or your job, which is judgment at scale. It’s the opposite of replacing it. A skill built the way described above is worthless without your judgment in it first. The AI isn’t the asset. The instructions are the asset, and the instructions are just your expertise, written down once so it can run without you standing there every time.
That’s the actual shift. Most sales orgs using AI right now are using it to make individual people faster at individual tasks. That’s a productivity tweak. Turning your judgment into a stack of skills that runs on a schedule, coaches every call, flags every risky deal, and preps every forecast before you open the spreadsheet- that’s not a tweak. That’s the org running on your standard, not your bandwidth.
Six hours over a long weekend. Two or three skills, built properly, fed real deals and real calls until they sound like you. That’s the whole exercise. Print this out, block the time, and go build it.