your AI crash course

👋 What’s up peeps!

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

TLDR; inside today’s newsletter

Let’s rip…

For those of you who want to “get into AI” or simply learn more about the field. This is the resource I wish I had.

The best way to get good at AI is to use AI. Just like sex, you should be better at it, the more you do it.

So….practically all your answers and knowledge can be found by prompting an AI.

You can use chatGPT, Gemini, Claude, Grok, Amazon Q, Qwen whatever. It doesn’t matter much. I recommend you pay for it, so you can ask unlimited questions.

So getting good at AI is about asking good questions, not memorizing knowledge. If you don’t understand anything just ask AI to explain it to you like you are 12 years old. Then keep asking questions until your satisfied.

Put the prompt below into your favourite AI and you will have an AI crash course.

Then after the course you need to come up with a small project to use AI in a real use case. You can only learn so much for reading and watching. You have to DO something.

Perhaps ask AI come up with a recipe with 5 ingredients from your fridge or use it to plan a trip or optimize your finances. Add documents and ask it to analyze something. Get going and you will realize you can use it for every possible query.

Good luck!

—————————————————————————————————————

COPY AND PASTE THIS PROMPT INTO YOUR FAVOURITE AI TO GENERATE YOUR CRASH COURSE

—————————————————————————————————————

I want to understand AI so I can pursue new professional opportunities. Turn the below questions into a crash course with detailed answers. Explain every concept in detail, so for example if you use the word “architecture” explain it the context of AI.

—————————————————————————————————————

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

I started off wanting to build an AI agent that can run a sales demo of any website, answer questions and drive sign ups.

Why? So customers don’t need to wait a week to get their questions answered.

I was unsure if it was technically possible, but with a mix of Lovable, Supabase, Browserbase, OpenAI, ElevenLabs, I think it is possible. The question is….will it be good.

Now my agent can…

Next I need to sync the navigation with the speech and get the Q&A functionality tight.

It’s also a bit slow, so I am working on that. Latest screenshot below.

I will share a link once it is presentable 🙂

This is fun!

Sales platform Apollo wired Anthropic’s Claude 3.5 Haiku into its messaging engine and A/B tested the hell out of it.

Result: +35% meetings booked and 76% customer preference for Claude-written copy. Although I must admit, I still like writing my own copy. 😊

Customers support the uplift. FLO EV Charging cut email writing time by ~80%, Qualytics cut prospecting time by ~40%.

Reps report fewer rewrites, more replies, more meetings.

How they win
Apollo didn’t just “turn on AI.” They evaluated models, picked Claude for writing quality, and wired it into workflows.

Why it matters
With AI, reps should be able to carry more accounts AND increase results. The only missing piece is incentive.

Amazon’s finance teams expanded gen-AI to back office chores like tax compliance, transfer pricing, forecasting, and revenue analysis. Reporting ~$260M annual savings and cutting completion time on key tasks by half.

Who said AI can’t be trusted with money?

Tooling: internal generative-AI chatbots/agents (likely Amazon Q) across workflows. Think ticketing, analysis, and document generation.

How they win
Pointing AI towards high-frequency, rules-based tasks and save humans for judgement calls.

Why it matters
More AI uses cases across all business functions: Customer Support, Sales, Marketing, HR, R&D and now Finance. Is there any place we cannot use AI?

What it does: AI-powered people search. Use Happenstance to search across your professional network (Gmail, Outlook, X/Twitter, LinkedIn exports, Slack soon) to find people. Use it for sales, hiring, fundraising, maybe even a date 😉.

Example: You can search “Show me directors of data engineering at pharma companies in the UK.”

Why it’s valuable: You stop cold-pitching strangers and start using warm intros your CRM can’t see.

Website: happenstance.ai

What it does: A general-purpose AI agent that breaks a goal into steps, opens its own “computer,” executes tasks and returns results.

Example: “Draft a landing page, buy a domain, deploy it, and send me analytics”—Manus plans, executes, and reports back.

Why it’s valuable: Moves you from brainstorming to done. Useful for lean teams that need an extra pair of (robot) hands.

Website: manus.im

Here’s my bet: by early 2027 the most used model in the world isn’t GPT or Claude — it’s Qwen.

Not because it wins every benchmark beauty pageant, but because it’s good, fast, cheap, open-source, available and is an antidote to OpenAI.

It’s kind of like Android vs Apple’s iOS.

Qwen is now stocked by AWS Bedrock, live in Google Cloud’s Vertex AI and turning up in Azure’s catalog.

That’s optionality for CTOs and a lifeline for procurement teams who don’t want to be locked into OpenAI.

Distribution wins markets, especially when it arrives through partnerships you already have.

Meanwhile, the product keeps getting better. Alibaba just unveiled Qwen3-Max, a trillion-parameter flex aimed squarely at frontier use cases. It also shipped Qwen3-Omni as open source so enterprises can audit, fine-tune, and deploy without legal heartburn. There’s no auditing GPT-5 or Claude 4.1.

None of Qwen’s products need to be perfect; it needs to be close, timely, and everywhere.

Plus Alibaba is a brand many know, versus Deepseek, that sounds like a shadowy Chinese Communist Party trojan horse.

And then there’s price. Qwen is typically 30-80% cheaper that OpenAI and Anthropic.

Qwen is well positioned to be the world’s Android vs OpenAI’s iOs.

The only way I see this not happening is if Google or Grok go FULL open-source. But I doubt it. Google appears timid and Grok doesn’t have the cash. But then there is a famous saying in technology - “never bet against Elon”.

What it means for you:

If you’re in Europe or APAC get comfortable with Chinese tech, use Qwen become an expert. Your customers will ask you about it.

Barring geopolitics tripping the cord, Qwen’s playbook of being good, cheap and everywhere, will win quietly…and then all at once.

OpenAI and Nvidia signed a partnership to light up 10 Gigawatts of Nvidia systems. Millions of GPUs, enough to power 7.5M homes. Doubling the entire world’s existing AI datacenter footprint. The first Gigawatt drops in late 2026.

NVIDIA and OpenAI are thinking like nation states not companies.

Meanwhile, OpenAI + Oracle + SoftBank unveiled five new Stargate datacenter sites. OpenAI’s CFO has already admitted it makes sense for them to go after the core AI infrastructure business of AWS, Azure and Google Cloud. Are the Big Tech hyperscalers getting disrupted? Everyday OpenAI’s dependence on Microsoft shrinks.

Why it matters:
In AI, electricity has become the product. Whoever controls cheap, reliable Gigawatts can run more experiments, develop advanced models, sets prices and lead the market.

Alibaba launched Qwen3-Max, a trillion-scale model aimed at the same class as GPT-5 and Claude 4.1. Bigger models mean more capability, and cost. 😉

They also shipped Qwen3-Omni, a multimodal model that handles text, images, audio, and video.

Alibaba is not messing around.

The smart move is now distribution: Qwen models are showing up on AWS and Google Cloud, letting Western teams evaluate and buy with standard tooling. No friction.

Why it matters:
We now have increased competition at the top end: GPT-5, Claude 4.1, Gemini 2.5, Grok 4, Qwen 3, Deepseek 3.1. With OpenAI and Anthropic pulling media attention, talent and budget, hyperscalers benefit from stocking an alternative. Watch for the softening stance on Chinese AI.

Google just jammed Gemini into Chrome: a one-click AI sidekick that can answer questions across your open tabs, summarize pages, and soon do chores like booking haircuts. It’s rolling out to U.S. users now, the rest of us in October.

Perplexity has long since done the same with its AI browser Comet, but with a bit more attitude, pushing the boundary on automations. Chrome AI appears more like a helpful sidebar than a full-time research assistant. But let’s see.

Why it matters:
Chrome has distribution, Perplexity has opinions. If you want a zero-friction rollout to your whole org, Chrome’s the safe bet. If you want agentic workflows today, Comet feels spicier.

Cohere just raised $100M lifting their valuation to $7B. Remember Cohere? Founded by a co-author of the original “Attention Is All You Need” transformer white paper that set off the LLM craze. Well...they are still at it.

They just struck a deal with AMD to reduce dependency on NVIDIA GPUs and have now surpassed $100M ARR. Trending towards $200M by year end.

That’s modest next to OpenAI’s $12B run-rate and Anthropic’s $5B run-rate. Different scale, but real revenue.

Why it matters:
More options for buyers. The siege on the NVIDIA-OpenAI duopoly continues.

Yep!

In case you are wondering where the first $100B came from, that would be VCs, Microsoft, Google, Amazon, and Meta buying GPUs from NVIDIA.

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