Why most AI demand gen reps fail

👋 What’s up, quota crushers —

This is Prompt Punk — the self-appointed best AI sales newsletter in the whole damn world! 😉

TLDR; inside today’s newsletter

Let’s rip…

AI Demand Gen reps don’t suck—sloppy setups suck.

Everybody wants a robot rep: works 24/7, follows your instructions, never complains!

Just like everyone wants a flat stomach before summer.

But let’s keep it real.

Jason Lemkin (VC and sales guru) watched 20+ teams try AI Demand Gen reps; 90% got zero pipeline, zero meetings, zero anything.

The 10% who treat the bot like a $100k hire? They’re booking more qualified meetings than human teams. Some scaling to $10M+ ARR with AI doing 80% of outbound.

Incredible.

Here’s the reality check. You have to treat AI like a new employee, not a slave.

Here’s how to make AI Demand Gen reps work for you:

Three non-negotiables

Pattern to $10M ARR from $0
Month 1: 40+ hrs setup + daily QA.
Months 2–3: weekly tests, persona tracks, dynamic data.
Months 4–6: add LinkedIn + phone, lead scoring, team around what works. Then it becomes predictable pipeline—because you worked the system, not the hype.

Bottom line: AI Demand Gen reps are leverage, not magic.

Put in discipline, get results.

Put in laziness, get failure.

Treat your AI like a top hire—train it, coach it, measure it.

Just like your stomach.

Uber for Business plugged Gong’s revenue AI into the entire GTM stack—onboarding, forecasting, coaching, deal execution.

Result: less admin, faster follow-ups, and reps pitching the right value props to the right personas.

The results

Why it works:
Gong’s AI Tracker agent mines real call data, flags the value props that land for each buyer type, and pushes them into workflows—so reps stop guessing and start mirroring what top performers say on winning deals.

Why it matters:
Precision beats volume. If your pipeline review still starts with vibes, you’re donating revenue to your competitors using AI.

Operationalizing AI (not dabbling) gives reps time back and more shots on goal.

What it does: Finds real emails and phone numbers for the people you actually want to talk to. Pulls data from LinkedIn/Sales Nav, verifies it across multiple sources, and adds useful context (role, company info, tech stack). Pipes straight into your CRM or Zapier so you can hit send, not copy-paste.

Why it’s valuable: Fewer bounces, fewer “who is this?” replies, more booked calls. It’s the difference between throwing spaghetti at the wall and emailing the right person on purpose.

Website: https://useartemis.co/

What it does: An AI rep that finds prospects, writes personalized emails, sends follow-ups, and books meetings inside Salesforge. You can let it run on autopilot or keep a hand on the wheel.

Why it’s valuable: You get consistent outreach without adding headcount or stitching five tools together. Frank handles the grunt work; you handle the calls that matter.

Website: https://www.salesforge.ai/agent/frank

As a European, I want Mistral to thrive. But right now Mistral’s shipping for CIOs, not civilians.

Mistral recently added Memories and 20+ MCP connectors (hello, IT), while there is lot of features released to satisfy the “on-prem” crowd.

The problem? I think Mistral is being too rational.

Chasing enterprise dollars because that’s where the money is perceived to be.

Look at Anthropic. They weren’t winning mindshare and almost felt like a little brother of OpenAI. Emphasizing “safety” when nothing bad had happened, yet.

But then they did something different.

They focused on coding before it was hot. Dropped Claude Sonnet 3.5 in the summer of 2024. Boom!…then agentic coding hit, Boom!

The result?

Anthropic rocketed to >$5B annualized revenue, 300k+ business customers, and ~$500M run-rate from Claude Code.

That’s a consumer-ish wedge turning into enterprise dollars.

I want the same for Mistral.

But how?

What to ship next (like, now):

I would do to email and meetings what Claude did to coding:

Mistral Meets — a dead-simple, privacy-first voice + email app. Record calls/meetings, auto-summarize, then draft and queue emails instantly: subject lines, TL;DR, decision bullets, and next steps pulled from the transcript. One tap to send via Gmail/Outlook, cc the thread, attach call notes, log to CRM, and schedule auto-nudges if nobody replies. Local-mode option, EU data posture by default, $10/mo prosumer plan.

Make email as easy as talking.

Why this wins:
Prosumer > enterprise in 2025. Consumer-grade dopamine (speed, taste, shareability) builds distribution and mindshare; distribution converts to enterprise line items.

Selling to the Enterprise one-by-one is too slow.

Wow, my AI dreams are coming true! Perplexity’s Comet browser just went hands-free.

Now you can talk to your browser—search, summarize, switch tabs, and navigate without touching the mouse.

Admittedly it’s a bit slow. See demo below. But there future is here.

Comet’s assistant also steps beyond Q&A to act in the page and across tabs when you say “take control of my browser”—think shopping, reservations, even email and social chores.

Net-net: a voice-driven, agentic browser built to keep you in flow.

Why it matters:
Voice-first agents pull AI out of chatboxes and into workflows—fewer clicks, faster admin, and real accessibility wins for anyone juggling work on the go.

Anthropic closed a $13B investment round at a $183B valuation—nearly triple spring valuation levels. Investors reportedly wanted to put in ~$25B.

Revenue is exploding too: an annualized >$5B run-rate, up from ~$1B earlier this year, driven by 300k+ business customers.

Claude Code, their AI coding product, has grown to $500M ARR in 6 months. Real value not PoCs.

Translation: Claude isn’t just clever—it’s commercial.

Why it matters:
That war chest buys chips, talent, and enterprise land grabs. If you’re selling into big logos, expect the RFP line to read: “ChatGPT or Claude—or both?”

xAI just launched grok-code-fast-1, a low-latency, low-cost model tuned for agentic coding. Spec, edit, run, repeat—built to live inside dev workflows.

xAI thinks coding is not just about intelligence but speed. I think they are on to something. I hate to wait while my AI thinks.

It’s available through xAI’s API plus Cursor, Windsurf and the usual places.

Why it matters:
If code agents feel instant, teams ship more with fewer humans in the loop—and velocity becomes the moat.

OpenAI is acquiring Statsig—the A/B testing + feature-flag platform. Its founder, Vijaye Raji will become CTO of Applications (primarily chatGPT).

Signal received from OpenAI: less headline R&D, more ruthless optimization.

ChatGPT and GPT-5–era apps will be instrumented end-to-end with UX variants, staged rollouts, guardrails, traffic routing, and kill-switches—to lift retention while squeezing latency and cloud spend.

Fewer “ta-da” magic launches; more continuous experiments that quietly compound.

Why it matters:
As frontier gains get pricier, advantage shifts to experimentation + unit economics—ship, measure, cut costs, repeat. It does work 🤷

How Microsoft CEO Satya Nadella uses AI (Microsoft 365 + GPT-5) to run his business.

He prompts his email, calendar and document data in one place👇

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