my website-demo agent, v1

👋 What’s up folks!

This is Prompt Punk — the best AI sales newsletter in the universe! 😉

TLDR; inside today’s newsletter

Let’s dig in…

These days everyone’s “doing AI” in outbound sales by smashing ZoomInfo, Apollo, Clay and Listkit with 1,000 “personalized” messages.

The problem? You and 10,000 other reps are doing the same thing. Your “hot lead” got the exact same email… yesterday.

Let me share how you can build a cold email list your competitors can’t buy.

1) Nail the ICP (Ideal Customer Profile)
For example CTOs at Series A startups in Europe funded in the last 24 months. Be specific.

2) Pull the company list
For example buy a Crunchbase subscription (~$99/mo) or an association list and export the exact companies you want.

3) Turn names into domains
Use Jsonify (or any AI web automation) to map company names → website URLs in bulk.

4) Human-powered enrichment
Hire a hardworking, inexpensive freelancer from Upwork (I like the Philippines 🇵🇭 ) to visit each site and capture: First Name, Last Name, Title, Email Address. They can use Email Hunter for support but it should mostly be done manually by visiting sites and searching Google.

I usually work with Christina Lao. Tell her I sent you!

Record a 3-minute Loom showing how to find 3 good emails—zero ambiguity, fewer mistakes.

5) Track and give feedback
They fill a shared sheet; you review every couple days.

Yield & cost: From 1,000 websites, expect ~650 clean emails. Budget $300–$500 for sourcing + ~2 weeks.

Result: a fresh, clean list no one else is hitting.

Why it works:

Now you just have to write a great cold email 😉

This is a new section of the newsletter where I will write about AI I am building. Only when there is something interesting to write.

If I talk about, write and sell AI it makes sense to build it too!

Imagine a chef that talks about, writes and sells food, but doesn’t eat. Bah!

Ok, so I am trying to build an AI Agent that can do website demos. You know, click around, talk through a product and answer questions.

The purpose is stop customers having to wait a week to speak to a rep. Now, businesses can sell faster.

So how did I get started?

Well, I put the requirements in ChatGPT, Claude, Grok and others to see who had the best idea on how to build it.

They came back with complex solutions so I had to simplify, simplify, simplify.

I like to build my v1s in Lovable, so I started building and the very basics are working. Here’s a pic:

I use Browserbase for the AI agent and OpenAI to “understand” the website. I might change to Claude, but I already had a OpenAI API account and couldn’t be bothered to set up a new account. Sorry Claude. This is quick and dirty for now.

Browserbase is complex. I get thrown a lot of errors, but usually just copy and paste to chatGPT or Claude and find a solution.

Two steps forward one step back.

Trial and error ad infinitum.

For now my agent can receive a prompt like “tell me the main features of X”, navigate to the website and click around to what it thinks are the relevent pages.

Next: I want my agent to “talk” about the product it is navigating. Some kind of visual-to-text interpretation. Let’s see 🙂

HubSpot’s 2025 “AI in Startup GTM” says the winners make AI someone’s job. Not everyone’s hobby.

When startups assign real owners, growth jumps and ARR skews higher. As one CEO put it “$100M with <150 people” went from insane to doable.

The results

How they win
Winners centralize AI ownership (team or lead), fund it, and wire agents into GTM workflows—bots handle repetitive service/sales tasks, humans take strategy and edge cases.

Why it matters (winners & losers)

AI is helping sellers match the right job and get recruited faster. No more ghosting!

MetaView (an AI recruiting platform) makes it clear: AI-augmented recruiting cuts interviews ~30% per hire, saves recruiters ~10 hours/week, and boosts decision confidence.

Your likely to get a much fairer shake when your responses are AI recorded, analyzed and fairly weighted vs other candidates. Hopefully less bias.

Case studies show it in the wild: Brex lifted onsite-to-offer rates, EvenUp standardized sales interviews so every candidate was judged on the same rubric, Catawiki used AI Reports to surface what candidates really want (e.g. in-office vs. hybrid).

Why it matters

AI turns interviews from vibe-checks into evidence.

Structured notes + call transcripts mean your answers get captured, compared, and credited—not lost between debriefs.

Every recruiting process should use these tools.

What it does: AI voice “researchers” that call your customers, run natural 1:1 interviews in parallel, and auto-analyze transcripts into themes, charts, and reports—turning weeks of research into days.

Why it’s valuable: Cuts cost/time vs. traditional research while keeping real human nuance.

Website: keplar.io/voice

What it does: Tracks your brand’s “AI search” presence—see how often ChatGPT, Perplexity, Claude etc. mention your company, what they say and where you rank.

Then generate AI-optimized content with workflows to win more citations.

Why it’s valuable: AI answers eat traditional SEO, Profound gives marketing/PR teams a dashboard and tools to capture share-of-voice across AI.

Website: tryprofound.com

AI is where the money moved. Not just in code but in comp.

PwC’s new Global AI Jobs Barometer has the headline stat: wages in AI-exposed industries are rising 2× faster, and workers with AI skills now earn a ~56% premium.

And all you had to do was write “AI” on your LinkedIn profile 😜

But in all seriousness, the divide is accelerating.

Zooming in on GTM teams. At OpenAI, median Sales comp clocks in around $240k, with individual roles running higher. I even heard a story of new hire seller in Europe earning $600k.

Sales Engineers—the field generals wiring RAG, agents, and data plumbing—sit around $198k median across companies.

Anthropic’s own AE postings list $150k–$270k base BEFORE bonus/equity.

These aren’t unicorn outliers; they’re the going rate where AI adoption is real.

Why pay people so much?

AI-native deployments live or die on successful company wide rollouts. Sellers have to run the gauntlet of pilots, pricing, procurement, integrations and change-management.

If the TAM (Total Addressable Market) is nearly all knowledge work, the opportunity is MASSIVE!

And once a company chooses a stack, they ain’t changing so easily.

AI companies will pay up for sellers who can move a customer from “cool demo” to “widely deployed.”

PwC’s dataset also shows productivity (revenue/employee) running hotter in AI-intensive sectors, which gives CFOs cover to fund premium GTM talent.

For those not working in AI, wage growth is slower, the future is uncertain and you might get replaced by a chatbot.

Time to become AI-native.

Last week, I wrote about Sana. This week they got acquired. The AI platform that connects to all your knowledge (Docs, Slack, Wikis, Drive, CRM), then lets you chat, search, and ship AI agents that do the work.

Effectively, your company’s brain.

Well…it was just announced Workday the HR/Finance system-of-record behemoth ($63B market cap) will acquire the Swedish startup.

Why it matters:
Employee AI platforms like Sana, Glean and Amazon Q for Business are quickly becoming the front door of work: a single place where employees ask questions, trigger approvals and launch agents. No more wikis!

Why it matters for sellers: Does your company have an employee AI platform where you can query your CRM intelligently? Learn from past deals and move faster? If not, you’re missing out on meaningful productivity gains.

Groq is an AI-chip startup that builds Language Processing Units (LPUs). These are purpose-built chips to run AI inference fast and cheap. They just raised $750M and claim speeds 18x faster than other clouds.

They compete with NVIDIA, AMD, Cerberus and the hyperscalers. You can use their chips on GroqCloud or deply on-prem with GroqRack.

Groq CEO previously invented the Tensor Processing Unit (TPU), Google’s AI chip. So the team has a ton of credibility.

Why it matters:
Groq is a Plan B versus complex and expensive NVIDIA GPUs that can do many things well, but don’t specialize in inference.

Why it matters for sellers: The AI industry will be charging less and less for tokens as serving costs plummet. How do you thrive when your product gets cheaper every quarter? Something to think about.

Evidence on billions of chats shows sellers aren’t using ChatGPT to “write the email”—they use it to decide what to do next.

The dominant work behaviors map to getting information, making decisions/solving problems, and documenting—together ~80% of work use.

“Writing” is big, but two-thirds is editing your text, not drafting from scratch.

“Asking” (decision support) grew faster than “Doing” and is rated higher quality by users.

GTM reps treat ChatGPT like a deal coach—pressure-testing discovery paths, objection handling, pricing logic, and then polishing the note they already wrote.

Why it matters:
How safe is it for teams to outsource thinking to an AI? What if great ideas are not in the training data and we just do what works well probabilistically with incomplete date?

Personally, I try to think and jot down an answer before asking AI.

I’d rather tell the machine what to do, than have the machine tell me what to do.

😂 It’s true.

Thanks for reading! ✌️

If you liked this:

📨 Forward to a biz techie friend
💬 Hit reply and tell me what you want more (or less) of
📊Vote in the poll below 👇

— John
Prompt Punk