The argument in one line.
Delegating cold email copy to an AI agent only works when the agent receives a detailed opinionated brief that bans templates enforces a word count and defines exactly what a soft CTA sounds like.
Read if. Skip if.
- You run outbound sales or a cold email agency and want to collapse the research-to-send workflow into a single agentic session.
- You use Claude Code or Cursor and want to see how to chain an external CLI tool into an AI coding agent.
- You send B2B cold email and suspect your copy is the bottleneck but are not sure how to diagnose it.
- You manage multiple client outreach campaigns and want a project-level npm dependency pattern for CLI tools.
- You do not send cold email and have no plans to.
- You are looking for a general Claude Code tutorial -- this is almost entirely Instantly-specific.
The full version, fast.
The video argues that AI-written cold email fails not because of the model but because the brief is too vague. The presenter dictates a rigid ruleset to Claude Code -- no templates max 100 words mobile-readable hook in 6-12 words follow-up adds insight not a bump -- and Claude turns that into skill markdown files governing every draft. The agent then uses the Instantly CLI to segment a scraped CSV by Starlink product fit write fully personalized emails per prospect and push three-step campaigns directly into Instantly without leaving the IDE.
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.

01 · Claude Code orientation plus setup
Overview of Claude Code access methods navigating to topoffunnel.com for the Instantly CLI creating an API key.

02 · Voice-dictated context brief
Brandon dictates a detailed project brief via Spokenly covering the offer ICP cold email philosophy copy rules and CTA framework. Claude scaffolds SKILL.md files.

03 · CSV segmentation with Polars
Claude sets up a Python venv installs Polars explores the CSV schema and distributions and segments the list into cohorts by Starlink product fit. Instantly CLI installed at project level.

04 · Calibration round plus copy refinement
Claude generates 5 sample emails per Tier 1A sub-segment. Brandon iterates on weak hooks. Second session after an API error. Final copy approved.

05 · Campaign creation via Instantly CLI
Claude creates 3-step campaigns for all sub-segments. Brandon verifies leads uploaded per-lead personalization confirmed sequence threading correct spam score zero.

06 · Outro CTA
Talking head outro recommending next video on starting an agency in 2026.
Lines worth screenshotting.
- AI-written cold email fails when the brief is vague not when the model is weak.
- A 100-word cap forces the model to cut filler that would have killed reply rates anyway.
- Follow-up email two should add a new insight not say bump -- most people do not remember the first email.
- Storing cold email rules in a SKILL.md file means every future Claude session inherits the playbook without re-prompting.
- Polars is faster and uses less RAM than Pandas for CSV exploration on a local machine.
- Installing a CLI at the project level keeps client work isolated and prevents API key bleed across projects.
- B2B buyers act on three motivators only: make money save money or save time.
- Voice-dictating a context brief to an AI agent is faster than typing and produces more opinionated constraints.
- A zero spam score is achievable with personalized plain-text email.
- The CTA goal is not a meeting -- it is a one-word yes to a hypothetical which gets far more replies.
Nine rules that make AI-written cold email actually work.
The model is not the bottleneck in AI cold email -- the brief is and a nine-rule document stored as a SKILL.md file changes every output that follows.
- Banning templates and merge tags forces the agent to research each prospect individually rather than slot names into shared copy.
- A 100-word hard cap eliminates the filler sentences that kill reply rates before the reader reaches the ask.
- The hook sentence should reference something specific to this company that could not appear in any other email on the list.
- Follow-up email two is not a bump -- it adds a new piece of information the prospect did not have after the first email.
- B2B buyers transact for three reasons: make money save money or save time -- assigning one to each sequence step gives each email a genuinely different angle.
- Storing rules in a SKILL.md file means a new Claude Code session inherits the entire playbook without re-prompting.
- Installing a CLI at the project level rather than globally keeps API keys and dependencies isolated per client or campaign.
- A zero spam score is achievable with plain-text personalized email -- the formatting choices that make email look polished are the same ones that trigger spam filters.
Terms worth knowing.
- Instantly CLI
- An npm package that exposes Instantly.ai campaign management as command-line commands letting an AI coding agent create campaigns add leads and configure sequences programmatically.
- SKILL.md
- A markdown file placed in a Claude Code project that defines domain knowledge or behavioral rules the agent reads at the start of every session acting as a persistent system prompt for that project.
- Polars
- A Python DataFrame library written in Rust significantly faster and more memory-efficient than Pandas for large CSV exploration tasks.
- ICP
- Ideal Customer Profile -- the specific firmographic and behavioral description of the buyer most likely to convert used to filter and segment a lead list.
- Tier 1A sub-segment
- The highest-priority cohort within the ICP list identified by the AI after analyzing the CSV -- companies with the most immediate and obvious fit for the offer.
Things they pointed at.
Lines you could clip.
“There is absolutely zero trust. So it is critical that the copy that we send is extremely relevant that every single sentence and every single word is carefully crafted.”
“Everybody loves to be reminded on why we did not reply in the first place. So therefore it is actually a waste of a perfectly good opportunity.”
“If you want to make sales you still have to reach out you have to follow up and you have to have a repeatable system that turns leads into clients.”
Word for word.
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
The bait, then the rug-pull.
Two billion cold emails sent. That is the number on screen before anyone speaks -- a social proof graphic Instantly uses to establish authority before the tutorial begins. What follows is not a concept pitch but a live working session: one practitioner one AI agent and a raw CSV of prospects turned into live multi-step campaigns in under half an hour.
Named ideas worth stealing.
The 9-Rule Cold Email Brief
- No templates no merge tags -- every email is one-to-one
- Max 100 words
- 1-2 sentences per paragraph
- Mobile-readable in 5 seconds
- No filler openers no em-dashes
- Hook 6-12 words ultra-specific to this prospect
- Middle: credible peers-did-this proof without pitching
- CTA: one-word-reply ask (would that be worth a look?)
- Follow-up 2: fresh insight not a bump; Follow-up 3: reframe via make/save money or save time
The complete ruleset Brandon dictated to Claude Code as a SKILL.md file that governs every cold email draft the agent writes.
B2B Buyer Motivation Triad
- Make money
- Save money
- Save time
The three reasons B2B buyers transact. Used to assign each email in the sequence a different motivator.
How they asked for the click.
“Click this video right here where it breaks down exactly what we would do to start a brand new agency in 2026 from scratch.”
Standard end-screen CTA. No email list pitch no product upsell.










































































