In partnership with
Date: 18-Jan-2025
Hey {{first_name | AI enthusiast}},
We had it good for a while!
We enjoyed powerful AI chatbots like ChatGPT, Perplexity and Claude for free mostly, since 2023 and now it seems ads are coming - at least in ChatGPT. Well, Open AI needs to survive and make money, so I don’t blame them.
Does this mean that the others AI majors will follow suit?
We also have our guest writer, Nisha Pillai, writing about an AI-powered quiz system that she built with Claude.
Also, we showcase Sabrina Ramonov’s “Mega Prompt” which will power your startup forward in 2026!
We round it off with an AI workflow that will create investment memos in a jiffy!
Table of Contents
Hope you enjoy this power packed edition!
PS: If you want to unleash the power of AI agents to grow your business, setup time speak to me, here»
ChatGPT Gets Ads: The Free Internet Model Comes for AI
The free internet always finds a way to pay its bills. This week, that model reached ChatGPT.
OpenAI confirmed it will begin testing ads inside ChatGPT for the first time. The test starts in the United States within weeks.
Ads will appear only for free users and the new $8-per-month ChatGPT Go plan. They will show at the bottom of responses when relevant. Each ad will be clearly labeled as sponsored.
Paid tiers stay ad-free. That includes Plus, Pro, Business, and Enterprise users.
🔹 Where OpenAI Draws the Line
OpenAI stressed several guardrails. Ads will not influence model responses. Conversations remain private from advertisers. I remain skeptical about this claim.
User data will not be sold. Ad personalization can be turned off entirely.
There are also strict exclusions. No ads for users under 18. No ads near health, mental health, or political topics. The company says ads fund free and low-cost access. It also claims it does not optimize for time spent.
🔹 A Clear Shift in Philosophy
This move reverses earlier public comments. In 2024, Sam Altman called ads a last resort. He described ads combined with AI as unsettling.
By mid-2025, that tone softened. Altman said he enjoys ads on platforms like Instagram. He added he was not totally against them. The financial team signaled flexibility earlier. CFO Sarah Friar said OpenAI was open to new revenue paths.
🔹 The Money Pressure Behind the Decision
The timing matters. OpenAI expects to burn $17 billion in 2026. That follows $9 billion burned in 2025.
The company does not expect positive cash flow until 2029 or 2030. Internal forecasts show ads could bring $1 billion in 2026. That figure could reach $25 billion by 2029.
Most users still pay nothing. About 95 percent of ChatGPT usage produces no direct revenue.
🔹 Competition Is Tightening
The market is shifting fast. According to Similarweb, Google Gemini jumped from 5.7 percent to 21.5 percent market share. ChatGPT fell from 86 percent to 64 percent.
The ad announcement landed alongside another move. ChatGPT Go is now live in over 170 countries. Prices are localized to expand adoption.
🔹 The Bigger Takeaway
AI is following the internet’s oldest playbook. Free scales first.
Monetization follows. OpenAI is betting it can add ads without breaking trust.
Founders should watch closely. This is likely the template for consumer AI economics going forward.
AI & Error: When You Can Finally Focus On Systems
I burned through an entire week's worth of Antigravity credits in one morning trying to get a Python script right. The logic was not complicated, I just didn't want to mess with syntax. I wanted the script to work, and I wanted it yesterday.
I was building a production-grade quiz generator that turns markdown into perfectly formatted PowerPoint slides. The quiz I run at my kids' school used to consume weeks of spare time for preparation. Research, difficulty calibration, game design across grade levels—that's the craft work I enjoy. Formatting slides, testing links, verifying sources, making sure styling is consistent - that’s all grunge work. I didn't have the time, and it needed to go.
So I built a system that handles all of it. And I did it in a couple days with Claude Code and Antigravity, despite not being a professional coder.
There were three parts that needed work:
(1) Question generator - making audience-appropriate questions in my style.
(2) Grunge work eliminator - all the formatting and testing and style fixing
(3) Demo - getting a version of online and live so others can see it working
A Few Lessons Learnt Along the Way
Voice mimicry came fast with the question generator. I fed Claude examples of my question style, and it caught on immediately. The hard part was difficulty calibration. Getting Claude to consistently distinguish between "broad trivia" and "expert specialist knowledge" took iteration after iteration. The model wants to collapse into the middle—everything becomes "moderately challenging." Teaching it to stay at certain levels required reference examples, explicit definitions, and constant refinement. I'm still tuning this.
Then came the parsing challenges while setting up the demo. The demo calls Claude's API, Claude responds, and my code has to parse that response into something the UI can display. Except Claude doesn't always format things identically. Sometimes it returns structured JSON. Sometimes it adds markdown formatting. Sometimes it includes explanatory text before the answer. Each variation broke something. To control the chaos, each API call needed to specify that the only acceptable response format is JSON.
Rate limiting turned out to be non-negotiable. Without proper infrastructure, the default solutions are trivially easy to bypass—anyone could drain your API budget in minutes. I implemented IP-based rate limiting. Until a few days ago I’d never heard of Upstash Redis. But it’s simple enough to figure out and set up, Claude does most of the work for you, and it works well for what I need.
What I Think Has Really Changed
I can read Python. I understand logic. I have an engineering background and I know how to think structurally about systems. But I don't want to fight with syntax. I don't want to debug indentation errors or look up which library handles HTTP requests. That cognitive overhead used to be the barrier between "I should build this" and "it exists." Not anymore.
I told Claude Code what I wanted. It wrote the Python script. When the script failed, I described the error and Claude fixed it. When I needed the web app built, I specified the architecture and Claude generated the Next.js code. When parsing broke, I explained the response format variations and Claude built the fallback logic.
The bottleneck isn't coding knowledge now. It's architectural thinking. Knowing to sandbox experiments so you don't accidentally wipe real data. Deciding between serverless and traditional hosting. Understanding when to use prompt caching—which cut my token costs by 90%— and pushing Claude to implement it versus accepting higher per-request costs for simpler implementation.
These aren’t new problems — they’re the same tradeoffs I’ve made in non-AI systems for years.These decisions still require judgment. The syntax, the library imports, and the boilerplate are handled.The consequences of architectural mistakes are not.
What This Means
The quiz generator is live. It works. You can try it on my site. I shipped a production system with rate limiting, intelligent caching, and a freemium model (two free demos, then bring your own API key) in the time it used to take me to manually format a couple of rounds of quiz slides.
The system isn't perfect. Difficulty calibration needs more work. The UI could be cleaner. But it's deployed, it's useful, and it solved the problem I built it to solve.
This is the shift that matters: AI hasn't eliminated the need for structured thinking or engineering judgment. But it eliminated the gap between "I know what this should do" and "it's live."
If you can think clearly about systems, you can build them now. The rest is just telling the AI what you want and iterating until it works.
Next week: So what does all this mean for tinkerers and builders?
Nisha Pillai transforms complexity into clarity for organizations from Silicon Valley startups to Fortune 10 enterprises. A patent-holding engineer turned MBA strategist, she bridges technical innovation with business execution—driving transformations that deliver measurable impact at scale. Known for her analytical rigor and grounded approach to emerging technologies, Nisha leads with curiosity, discipline, and a bias for results. Here, she is testing AI with healthy skepticism and real constraints—including limited time, privacy concerns, and an allergy to hype. Some experiments work. Most don't. All get documented here.
AI Agent Creates Investment Memos in Minutes (Nvidia Memo Demo)
Watch an AI workflow transform how wealth managers and investment analysts create research memos. This system uploads earnings call transcripts and financial statements. Searches for latest market news. Delivers a consolidated client-ready investment memo in under 5 minutes.
WHAT THIS VIDEO SHOWS:
• Live demo using Nvidia earnings data
• Automated transcript analysis and financial review
• Real-time news aggregation
• Complete investment memo generation
• Time saved: 60+ minutes per memo
THE TRADITIONAL PROCESS:
Analysts spend 90 minutes per memo. Reading transcripts manually. Pulling financial data. Searching multiple news sources. Formatting everything into client presentations.
THE AI-POWERED PROCESS:
Upload source documents. System analyzes earnings calls and financials. Pulls latest analyst updates and market news. Generates formatted memo ready for client delivery.
TIME SAVINGS BREAKDOWN:
• Manual research: 90 minutes
• AI-assisted research: 4-5 minutes
• Time recovered: 85+ minutes per memo
• Monthly impact: 25-40 hours per analyst
Learn Prompting for founders: MEGA Prompt to make 2026 your best year!
This prompt is courtesy of Sabrina Ramanov.
MEGA Prompt - Step by Step Breakdown
First, let’s break this mega prompt down into sections so you understand why it works.
"Part 1: Role
You are a top 0.1% tech founder of a multi-billion dollar rocket ship startup. (Note: Change “tech founder” to match yourindustry, e.g. “Real Estate” or “E-commerce Expert”).
Part 2: Context
Here is the context on my business:
Product: [Explain what you sell]
Customers: [Who is your highest retention user group?]
Goal: [What is your revenue or growth goal for 2026?]
Vision: [Where do you want the company to be in 5 years?]
Part 3: Data
(Note: Don’t lie to your AI co-founder. Input your real numbers.)
Here are my metrics:
Revenue: [Insert 2025 Revenue]
Churn: [Insert Churn Rate]
Funnel: [Visitor -> Lead -> Sale conversion rates]
Expenses: [Major cost categories]
Part 4: Activities
(Note: This is where you dump everything you’ve been doing in 2025 and everything you think you should do in 2026. I literally copy-pasted my massive TODO list from google doc into the prompt.)
What worked in 2025:
[List marketing activities that drove actual ROI]
What you’re considering for 2026:
[Dump your entire overwhelming to-do list here]
# TASK
Your task is to identify the TOP 3 MOST IMPORTANT things I should focus on, in order to reach $ [ ] monthly recurring revenue by the end of 2026.
There’s always a million things I could be doing, but I need to know the 3 most important things that will multiply my current monthly revenue.
Ask me clarifying questions, one at a time, until you are 95% confident in your answer."Obviously, tweak the prompt for your business and use case. This was tailored for a product-led growth SaaS startup where they don’t have sales touch points, they have recurring monthly revenue, and they currently don’t run ads, etc.
Try this one and you will create an AI co-founder for your startup or business.