TLDR FOUNDERS 2025-12-12

Find your ideal pricing model >>

8 TAKEAWAYS FROM CARTA'S STATE OF SEED REPORT

Seed funding is back to near-peak levels, but the money is significantly harder to touch. Investors are piling more cash into fewer rounds, essentially ignoring everything outside of SF and NY. The most telling stat is headcount: the average seed startup has slimmed down from 10 people to just 6, likely because AI is picking up the slack. However, don't let the tools fool you into going it alone: VCs still fund co-founders at double the rate of solo acts, despite the data showing that 25% of those partnerships end in a breakup by year four.

HARDWARE IS A FRUIT

The feedback loop for physical products is speeding up. We used to assume that building hardware meant long timelines and high costs for every mistake. That dynamic is fading. Modern simulation tools and AI assistants now let engineers iterate on physical designs almost as fast as they write code. The barrier to entry is shifting from manufacturing logistics to pure design capability.

DRIVING EXTREME CLARITY

At your startup, it's important to recognize that a "messy yes" is often more dangerous than a hard "no." Teams tend to drift into "strategic ambiguity" - nodding along to avoid conflict without actually agreeing on the trade-offs. The fix is to force the friction upfront: make the team stack-rank priorities (no ties allowed) and write down exactly where you disagree.

RL IN REAL LIFE: DURABLE MOATS

Reinforcement learning (RL) has gone from being a promising path towards general intelligence to a taboo phrase over the past year. It has proved highly effective inside AI labs, but outside of the labs, its use cases remain surprisingly underdeveloped. Most efforts appear to be largely experimental and have yet to scratch the surface of the sophisticated techniques used within the frontier labs. This post looks at the intricacies of building effective RL approaches, assesses current use cases, and discusses what founders should consider when commercializing RL-based systems.

context, never miss a follow-up. See how it works

SAFEBOOKS AI

Safebooks is an intelligent automation layer for quote-to-revenue finance operations.

ROARK

Roark is a platform to test, monitor, and improve AI voice agents with simulations and call analytics.

DOOLA: AI CO-FOUNDER

doola is an AI assistant that automates back-office tasks for e-commerce entrepreneurs.

A FEW THOUGHTS ON CONVICTION

The venture capital industry is full of sheep. Funds blindly follow others without thinking for themselves. They lack all conviction. As a result, they don't blaze new trails - and blazing new trails is the only way to earn outsized returns in a hyper-competitive market as an investor or an operator.

WHY STARTUPS DIE

Most startups eventually fail. This is rarely because of competition. Most of the time, their demise is self-inflicted by internal issues. The biggest threats usually come from within the company's own walls, not from the outside world.

IN 2025, I THOUGHT A LOT ABOUT THE THINGS I WANT TO SAY TO MY BOSS

Leadership can't fake that they care because people can feel it.

55% OF ALL DEPARTMENTAL AI SPEND IS NOW ON CODING. AND IT'S NOT SLOWING DOWN

It turns out the real money in AI isn't going to chatbots or marketing copy - it's almost entirely still going to code.

DIFFUSION OF AI ACROSS VERTICAL MARKETS

AI adoption isn't a singular event - it's a staged progression.

HOW TO USE AI AGENTS FOR MARKETING

SafetyCulture leverages AI agents to optimize marketing by achieving near-100% lead enrichment, doubling opportunity creation, and tripling meeting booking rates.

SPECIALIZED STABLECOIN FINTECHS

Stablecoins change the economics of building financial products and make specialization possible.