Newsletter #1 - AI startup ideas

The AI hype is real. It seems like everyone raising capital is trying to include an AI phrase in their deck just to get their term-sheet signed. In this letter, I want to talk about actual AI and where I see opportunity.

Consumer data platform

AI runs on user-generated content. It’s only a matter of time before users want to be compensated for their content as well as have safeguards to control their privacy.

I should be able to choose what gets fed into a LLM.

How would I do it? I would build a browser extension and a web dashboard where users can toggle what they want to be private and what they don’t mind being fed into a model, and how much they could receive.

The founder would form relationships with the main AI players and create a multisided marketplace, taking a spread in between. Up-to-date training data is valuable, with Google reportedly spending $60mm annually on Reddit’s UCG.

Agentic Workflow Automation

Shifting from consumer data to business workflows, another area ripe for AI disruption is "agentic workflow automation."

What is it and why should you care? LLMs will get smarter in 2024. You'll be able to create a small cluster of agents and sub-agents that can talk and interact, allowing complex tasks to be broken down. What was historically complicated or impossible will become possible.

Instead of having an LLM generate its final output directly, an agentic workflow prompts the LLM multiple times, giving it opportunities to build step by step to higher-quality output.

You can read more about the papers and experiments on DeepLearning.AI (https://www.deeplearning.ai/the-batch/issue-242/).

Back to the business idea. You could approach companies and pitch to automate their workflows (a potential $1-2mm pa. lifestyle business). Or go mass market, find a novel consumer approach to sell subscriptions for access.

As a productivity nerd, there’s potential to 10x the current goal achievement tools with an agentic approach. From setting goals with the main agent, planning with sub-agents, reflective improvement loops, and multi-agent collaboration for overcoming challenges.

The code isn’t too challenging and could be launched over a weekend.

Synthetic Data Generation

As AI models become more complex and data-hungry, the need for large, diverse datasets grows. But collecting real-world data can be time-consuming, expensive, and fraught with privacy concerns.

Enter synthetic data. AI algorithms can generate realistic, simulated data that mimics real-world patterns and statistics. This opens up new possibilities for training AI models in domains where data is scarce, sensitive, or hard to obtain.

Imagine a startup that specialises in generating synthetic medical images for rare diseases, allowing researchers to develop better diagnostic tools. Or a company that creates virtual test environments for self-driving cars, enabling faster, safer training. Synthetic data also has applications in privacy preservation.

By generating synthetic versions of sensitive datasets, companies can share data more freely without compromising individual privacy. As the demand for data grows, I believe synthetic data generation will become an increasingly valuable niche for AI startups to explore.

X catch-up tool

I thought of this because I wanted to explore Anthropic’s Claude ‘Tools’ launched yesterday. This is on my hobby project list. Listen, it’s simple and you might laugh, but I think it has value, or could grow into an app with value.

You give the app a list of 10-15 people who are important to you from your following list. You then task the AI to retrieve the last 20 tweets or replies from each of them, and then task it with providing you with a brief summary.

It explores their mood, topics, and what they were focusing on that day. Then, it's sent as an email.

Personalised Longevity

Let's explore an area where AI could have a profound impact: health and longevity.

Imagine a future where your healthcare is tailored to you as an individual.

That's the promise of personalised longevity, powered by AI and data modelling. By analysing your health data—from genetics to lifestyle factors—AI could identify patterns and risk factors traditional methods might overlook.

This could lead to personalised recommendations for lifestyle changes, treatments, or preventative measures to optimise your healthspan. Data modelling could help predict your health trajectory, allowing proactive interventions to maintain peak wellness. Imagine a longer life filled with vitality and resilience. That's the potential of AI-driven personalised longevity.

From empowering consumers with data control to automating workflows, optimising social media engagement, and personalising health journeys, AI is reshaping our lives.

These were just some ideas that came to mind, hopefully they spark something for you!

Which resonates most with you? How do you see AI shaping our future? Stay tuned for more insights in the next newsletter!