👋 What’s up friends!
This is Prompt Punk — the best AI sales newsletter 😉
This is my last newsletter of the year. Got some Asian travels ahead. Of course, I’ll be spotting how they use AI.
TL;DR - inside today’s newsletter
Playbook - 4 AI skills you must have in tech sales
AI Winners and Losers - 1 engineer 10x GTM results
Tools - Lead agent, popular AI customer support
Tech and Deals - Code Red, MLOps acquisition, Black Forest
Let’s rip…
🤝 The Playbook: 4 AI Skills You Must Have In Tech Sales
I did a long-form podcast interview on AI in Tech Sales. We discussed…
How do tech sellers survive and thrive in an AI future?
How do you prospect, qualify, validate and close?
What are the greatest career opportunities?
I discuss this all with Jiri Siklar from the Software Sales Formula podcast.
It was fun. Enjoy!
🏆 Who’s winning (and losing) with AI?
🤖 AI GTM: One Engineer Turns 10-Person Sales Team Into a 10x Machine
When Jeanne DeWitt Grosser (ex–Chief Business Officer at Stripe, now COO at Vercel) first tried to automate outbound at Stripe back in 2017, the idea was simple: pull in company data, auto-personalise emails, let machines do the boring bits. The problem? The tech wasn’t there. Even with top-tier data scientists, the system made too many dumb mistakes. The project died.
Fast-forward to Vercel. Same ambition, completely different outcome.
Jeanne hired a single go-to-market engineer and asked a brutal question: “What if an AI agent did 80% of what our sales team does?” In six weeks, that engineer shipped an internal AI agent that now:
Qualifies inbound leads
Runs outbound prospecting
Analyses why deals are lost
The bot costs about $1,000 per year to run. The sales team it effectively replaced cost $1m+ in salaries.
Instead of layoffs, nine reps were redeployed into higher-value work (big deals, strategic accounts, complex expansions). One remaining rep now sells with 10x the efficiency, because the “robot SDR + analyst” handles the grunt work.
It doesn’t stop at pipeline. Vercel also deployed an AI “deal-loss” agent that reviews every email, call transcript and Slack thread on a lost opportunity. In one flagship deal, the rep blamed pricing. The bot called BS: they’d never engaged the budget owner, and when ROI came up, the buyer clearly didn’t believe the value story.
Now, AI watches deals in real time and fires warnings like:
“You’re halfway through the process and haven’t spoken to a budget decision-maker yet.”
Around that, Jeanne’s team rebuilt GTM fundamentals:
Segmentation goes beyond headcount — OpenAI has ~3k employees but behaves like a mega-enterprise account, so they treat it that way.
Sales hires look like PMs for the first 10 minutes in front of engineers, deep product credibility, then commercial instincts.
Custom AI beats off-the-shelf: a GTM engineer shadowed top reps and built an internal deal bot in two days, tailored to Vercel’s exact workflow.
Result
One GTM engineer + one rep doing the work of a 10-person team. AI agents running outbound, qualification and post-mortems for ~$1k/year instead of $1m+.
Reps redeployed to higher-leverage work, and AI surfacing truer reasons for deal outcomes than human gut feel.
Ops is leaner, smarter, and far more honest about what’s actually driving wins and losses.
Why it matters
Most teams still see AI as “better email copy” or “cheaper SDRs.” Jeanne’s playbook shows something bigger:
Ideas that failed pre-AI are suddenly viable, don’t bin them, revisit them.
A single GTM engineer can be force-multiplying, building internal agents that match your exact process instead of duct-taping point solutions.
If 80% of customers buy to avoid risk, AI that flags missing budget owners, weak ROI and shaky champions will outperform dashboards and rep anecdotes.
Product-led growth will take you far, but not to $100b. Pair human sellers who think like PMs with AI agents that never sleep, and you get a sales engine that compounds instead of just hiring in straight lines.
TL;DR: Treat go-to-market like a product, hire GTM engineers, and let AI do the boring, brutally honest work your pipeline has been missing.
🛠️ AI Tools You Can Use
🤖 Vercel Lead Agent – Your 24/7 Inbound SDR on Autopilot
What it does: An inbound lead qualification + research agent you can drop behind your “Contact sales” form. It captures a lead, runs a deep research workflow on the company, auto-qualifies the lead (e.g. QUALIFIED, FOLLOW_UP, SUPPORT), drafts a personalised reply email, and sends it to Slack for human approval before anything goes out. Built with Next.js 16, Vercel AI SDK, Workflow DevKit, Exa search, and the Vercel Slack adapter.
Example: A prospect fills in your site form at 23:17 on a Tuesday. The Lead Agent spins up a background workflow, researches their company, categorises the lead based on fit and intent, and writes a tailored response. Your AE wakes up to a Slack message with the draft email, reasoning for the qualification, and approve/reject buttons. One click and the email is sent; the CRM and Slack thread now have a clean audit trail.
Why it’s valuable: Turns generic “Thanks, we’ll get back to you” forms into a 24/7 SDR that actually researches every inbound, keeps humans in the loop for brand safety, and gives your GTM engineer a reference architecture they can extend for your own playbooks (extra qualification categories, outbound flows, different data sources). Less manual triage, faster first responses, and more consistent follow-up.
🤖 Parahelp – Your 24/7 End-to-End Support Agent
What it does: Parahelp is an AI support agent that plugs into your existing ticketing system (like Intercom or Zendesk) and actually resolves complex tickets end-to-end, not just chat. It reads the ticket, retrieves relevant context from your knowledge base and tools, executes actions (refunds in Stripe, updates in Retool/Linear, etc.), and only hands off to humans when confidence is low. Teams like Perplexity, Framer, Replit, HeyGen and more are already using it in production.
Example: A user emails support asking for a refund and reporting a bug. Parahelp pulls subscription details, checks eligibility, processes the refund in Stripe, files a bug in Linear, and replies with a human-sounding message, all inside your existing helpdesk. If something’s ambiguous or risky, it routes the ticket to a human with full context instead of guessing.
Why it’s valuable: Turns your support org from “ticket router” into an automated resolution engine. You get higher end-to-end resolution rates, faster response times, and a bot that stays aligned with your policies via their AI Manager (which helps you configure, test, and optimize the agent by just… chatting with it). Less repetitive work for humans, more reliable coverage for customers.
Website: https://parahelp.com
🤖 Fresh Tech, Hot Deals 🔥
🚨 Code Red at OpenAI: New Model, New Business Model
OpenAI just pulled the fire alarm internally. A leaked “code red” memo from Sam Altman says the company is refocusing everything on making ChatGPT faster, more reliable and more “personal” and pushing other projects (health/shopping agents, the Pulse assistant) down the roadmap.
The memo also teases a new reasoning model dropping next week that Altman claims beats Google’s freshly launched Gemini 3 on quality. At the same time, engineers have spotted ad-related code in the ChatGPT Android app, and reporting says OpenAI has been quietly testing search ads inside ChatGPT.
Why it matters
This is OpenAI saying two things: (1) Gemini 3 was great, so the bar for “good enough” chat just moved up again; (2) even at a $500B valuation, they still need a durable revenue engine, and that probably means search-style monetization, not just API and seat licenses.
💾 Snowflake Buys a $200M AI Co-Pilot
Snowflake just signed a $200m multi-year deal with Anthropic to wire Claude models directly into its data cloud for 12,600+ customers. Think “agentic AI on top of your warehouse”: query data in natural language, auto-build apps, and let Snowflake Intelligence bots sit on top of your metrics and logs.
Why it matters
Snowflake is telling every CIO and CRO: your AI strategy should ride on your existing data gravity. If Claude becomes the default “brain” on your warehouse, expect sales teams to get copilots that understand real revenue data, not just pretty dashboards. And for vendors: “agentic AI on your data cloud” is now table stakes in every six-figure analytics deal.
📊 OpenAI Eats Its MLOps: Neptune Acquisition
OpenAI is buying neptune.ai, a Polish experiment-tracking startup that’s basically a metrics command center for model training. Neptune started as an internal tool at deepsense.ai, spun out in 2018, raised ~$18m, and now powers training dashboards for Samsung, Roche, HP and yes, OpenAI itself.
Deal terms aren’t public, but reports peg it at sub-$400m in stock. Neptune will wind down its public SaaS and go all-in on OpenAI, helping researchers compare thousands of training runs, debug weird model behavior, and see how frontier models learn in real time. First serious OpenAI M&A move in Europe, and part of a wider 2025 buying spree on infra tools.
Why it matters
Everyone talks about models, the real moat is the training stack. Creating a model creation factory. OpenAI is quietly pulling critical MLOps primitives in-house. I
🎨 Black Forest Labs – Europe’s Image Money Printer
German AI lab Black Forest Labs just raised $300m at a $3.25b valuation in a Series B co-led by Salesforce Ventures and AMP, with a16z, NVIDIA, General Catalyst and a pile of others piling in. They build the Flux family of image models (now Flux 2) that power image gen and editing in tools like Adobe, Picsart, ElevenLabs, VSCO and Vercel, with 4K output and multi-image style reference.
Why it matters
This is the clearest signal yet that image models are a real business line, not a demo feature. Flux is already hiding inside the tools your designers, growth hackers and content teams use every day, which means:
Budgets that used to go to stock photos, agencies and in-house motion teams will quietly drift toward Flux-powered workflows.
Europe has a flagship foundation model (beyond Mistral) with its own ecosystem, a serious alternative to US APIs for any product that touches pixels.
📭 That’s a wrap
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 👇
What do you think of today's newsletter?
— John
Prompt Punk