Don’t treat your AI as a PHD, treat it as a research assistant

The dangerous assumption of "I'm a people person - I'm safe from AI" I was chatting with a friend over lunch on using AI more in her work. For me, it seemed obvious that being AI-savvy was something everyone should be doing to stay competitive. She wasn't convinced so I dug a

The quality of an AI output is dependent on the input (prompt). Focus on quality, thoughtful, prompts and you'll be blown away by the outputs.

Relationships should compound like investments. Cut the underperformers loose.

Trust is built in inches and lost in miles.

Two types of people: Those who need permission and those who need results.

Discipline is about keeping the promises you make to yourself

Invest in knowledge that raises your standards

Take calculated bets with asymmetric upside

Invest in thinking clearly on topics where others struggle

Capital follows clarity. Clarity follows conviction. Conviction follows evidence. Evidence follows action.

Founders who are addicted to customer insights > those addicted to growth

Build the smallest thing that can prove the biggest truth. Then, scale the shit out of it.

The best product managers aren't project managers. They're more like anthropologists studying how humans work.

The best products don't just add features. They remove assumptions.

Most tools help you work faster. Few help you work differently.

Imagine if you could achieve 64% of your results in less than 2 hours a week? Seems like a sweet deal. The Pareto principle (also known as the 80/20 rule) is a phenomenon that states that roughly 80% of outcomes come from 20% of causes. Now if you Pareto x 2, it says that 64%

For the last year, we onboarded new Flow users with a Google Doc with how-tos and tips 😅 Finally, we now have a proper documentation hub! 🤩 docs.enterflow.co

Flow is featured in Daniel Pour Asgharian's 2026 Productivity Stack! 😎 medium.com/@danielasghari…

We cut our Flow OpenAI bill by 75% without sacrificing product experience. 🤑 Previously we were using a fine-tuned model which is more expensive. Instead, we've gone back to using a standard model but with better prompt design for our most expensive use cases like summaries