I built an AI sales rep. Lovable targets $1B.

👋 What’s up, deal chasers —

Welcome to Prompt Punk — the AI business newsletter for people in the business of selling technology.

Let’s get into it…

No, not the GPT-5 release but what they did to their free tier and legacy models. Sure the power users were not happy.

GPT-5…

But ignore those internet “hot takes”. It’s already fixed. This wasn’t a product update. It was a platform play.

GPT-5 (as opposed to the 6 other models they just retired) makes it easier and cheaper to serve AI with one model to rule them all. OpenAI wants everyone using ChatGPT-5 — for free.

Why? Ads.

OpenAI already has 700M weekly active users. If they can capture just 20% of the online ad market (currently dominated by Google, Meta and Amazon), that’s a potential $90B business. Compare that to their $12B+ subscription revenue and you start to see the angle. It’s easier selling ads to CMOs, than subs to goat farmers in Kazakhstan.

The tradeoff? One-size-fits-all models are hard. That’s why we still have Claude for code, ElevenLabs for voice, Grok for math, etc.

So here’s the real question:
Do you build one good-enough model for everyone, or specialized models that crush specific jobs?

OpenAI picked door one. For the rest of us…the future is sponsored.

Ever built a report in Salesforce? It’s like death by a thousand clicks. Now they’ve acquired Waii — an AI startup that turns plain English into SQL.

Think: “Show me all customers in EMEA with >$100K ARR” → boom, instant report.

Why it matters:
Salesforce doesn’t want to end up like Siebel (Siebel who?). This is one of their smartest bets yet — and they’re still hiring sellers while the rest of Big Tech reins in headcount.

Swedish “vibe coding” star Lovable announced it is now adding $8M-$15M ARR per month, on track for $250M by year-end and $1B within 12 months. They hit $100M ARR just eight months after their first $1M.

Now their new Agent Mode is default — reading your request, understanding your codebase, fixing bugs, running multi-file edits, and plugging into external tools with 91% fewer errors. Real-time app analytics just launched too.

Why it matters:
Lovable’s evolving from “build a demo tool” to a “build a full SaaS tool”, a bit like what Shopify did to e-commerce websites. The TAM (Total Addressable Market) is basically, all of modern SaaS development. Next week, I will share how I got offered $20k to build an AI-native CRM in Lovable.

London-based Humanloop, a YC-backed startup powering AI app development for teams at Duolingo, Gusto and more, is joining Anthropic. Their platform handles prompt management, evaluation, and fine-tuning — the unsexy but critical plumbing for enterprise AI.

Why it matters:
Anthropic isn’t just selling models — it’s buying the tools companies need to deploy them.

Everyone clowned Google for missing the (early) LLM wave they basically created. But here’s what’s cooking and why I am impressed!

Why it matters:
Google doesn’t care about churn, CAC or gross margins. They’re still trying to invent the future. Technology requires visionaries. Demis Hassabis is Google’s Jesus.

Pinterest used AI to reduce the number of clicks it takes to create ads on its platform.
Result: $998M in Q2 revenue, up 17% YoY.

Why it matters:
AI that shortens time-to-value = money. Whatever you sell, fewer steps = higher conversion. Everyone with a website should use AI to shorten the customer journey.

A couple of years ago I met Archie Hollingsworth, a young founder hacking on ChatGPT 3.5 to send automated emails in Gmail. We were both tinkering with Chrome extensions — mine was a hobby project called Rapid Reply. I quit. He didn’t.

Fast forward to today: I’ve have a day job. His company, Fyxer AI, just hit 500,000 users and $10M ARR.

Moral of the story? Maybe I should stick with my hobbies 😜

Why it matters:
People hate email. If your business solves email, calendars, or admin with AI — you’ll win. Fyxer is winning. Check them out.

GitHub’s CEO “resigned” (pushed?) and the company is now folded into Microsoft AI.

Why it matters:
Copilot was supposed to dominate AI coding.
Instead, Cursor, Lovable, Windsurf, Bolt and Replit ate GitHub’s lunch.
Microsoft is not messing around. AI is not just daisies and rainbows.

I see a new job role emerging — part strategist, part operations, part wild west sheriff: the AI Orchestrator.

They don’t build models. They buy them, integrate them, and own the results. See Jason Lemkin’s tweet below.

Why it matters:
Not every org needs prompt engineers
But every org needs someone to make AI actually do something.
If you have in-house AI projects, get someone to own the results, give them a nice title and attract the best. This is not an engineer’s side project.

Apple is quietly testing a new Siri that operates apps with voice. No taps needed.

Want to edit a photo, comment on Instagram, order sushi, buy something on Amazon? Just say it.

Why it matters:
This changes interface design. If Siri works as promised, and your app isn’t voice-operable by 2026… you’re invisible.

Singapore startup Sapient Intelligence — with alumni from DeepMind, DeepSeek, Anthropic, and xAI — is betting big on the Hierarchical Reasoning Model (HRM). A brain-inspired dual-network setup (think endless back and forth between CEO + COO discussions) that loops between each other until it lands at the right answer.

My wife often asks me to rethink, so I like this idea.

Why it matters:
Transformers (like ChatGPT, AmazonQ, Gemini, Claude etc.) make one-speed guesses and move on. HRM can change its mind mid-problem, delivering smarter reasoning on complex tasks. That means cheaper, deeper AI thinking — especially useful for industries that need multi-step planning (finance, logistics, strategy). Let’s see if it works.

Mostly importantly, engineers are getting excited.

I’ve been playing with ElevenLabs, the London-based AI voice company known for eerily realistic voices. They’ve raised $281M from A16Z and friends.

I tested it to build an inbound voice agent for a window-cleaning company and it worked. Sure it was a bit janky, but it kinda did the job.

[VIDEO] Watch my video explaining the build

Why it matters:
AI voice agents are no longer just for support.
They can qualify leads, book meetings, and even close small deals.
If you're in sales and not experimenting with this, you're falling behind.

How good is it?
✅ 70% of customer support? Done.
✅ Basic inbound sales? Almost.
🚫 Outbound prospecting? Not yet.
🌍 Multilingual? English is great. Others need work.

Should you try it?
Yes — if you run a service biz, get too many calls, or hate repeating yourself.

👉 Try my demo (no signup required)
Go talk to my AI — tell me what you think.

If you liked this:

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

See you next week,

— John
Prompt Punk