Notes from San Francisco

On Building, Distribution, and Where AI Actually Compounds

I just returned to India after a few months in San Francisco, and I am still sitting with what that time taught me.

The biggest takeaways did not come from demo days or bold predictions about the future of AI. They came from quieter moments, from building in public, shipping imperfect things, and spending time with people who are trying to turn ideas into something real.

Before sharing the ideas, a bit of context.

In October, I biked across the Bay Area delivering around 60 coffees to sales, marketing, and product leaders. There was no pitch deck and no structured agenda. It was simply a way to meet people and listen.

In November, I was invited into AngelList’s Founders Cafe. What began as a place to work slowly became a community and, over time, a second home.

In December, we built VibeTM v0 and onboarded our first 10 pilot users. The days were long. Onboarding was messy. Early conversations were raw and honest. A few small cheques came in from people who believed early, and they mattered more than the numbers suggested.

What stayed with me was not a single conversation or insight.

But repetition. The same themes surfaced again and again, quietly and consistently.

Once you see them up close, they become hard to ignore.

Here are a few of those ideas.

1. The software layer is becoming table stakes

AI will reshape many layers of the economy, but software is where the impact is most visible today. It is also where differentiation is disappearing the fastest.

With the tooling available now, a small team can build a polished AI product in a matter of weeks. LLM APIs, vector databases, authentication, hosting, onboarding, analytics. The path from idea to a functional MVP is well understood and widely accessible.

As a result, the bar for something that works is low. The bar for something that truly matters is much higher.

What compounds today is not speed alone, but depth.

• Proprietary data loops that improve with use.
• Architectural choices that unlock leverage over time.
• Workflows that embed into daily habits and create real switching costs.
• Revenue quality, not just user count.

You can already see this pattern repeating. Meeting recorders, email copilots, note-taking tools, and summarizers tend to converge on the same features within months. The interface changes. The demo gets sharper. The core value remains largely unchanged.

The advantage no longer sits in the UI.

It lives in what the system learns, what it retains, how it orchestrates decisions, and how that intelligence compounds with continued use.

This has changed how I think about product strategy. Shipping quickly still matters, but meaningful differentiation now lives deeper in the stack than most MVPs ever reach.

2. Distribution is quietly breaking

One pattern came up again and again in conversations with founders and operators. Getting in front of customers has become harder, noisier, and far less predictable.

Open your inbox and you can see it. Cold emails rarely land where they are meant to. If they show up at all, they sit in Promotions, Spam, or some secondary tab that never gets opened. Inbox providers have become extremely good at filtering automated outreach. Domain reputation, sending behavior, engagement history, and consistency now determine whether a message is seen or ignored.

• An arms race is underway.
• Filters keep getting better.
• Spam models get more aggressive.
• Automation tools work harder to bypass them.

Traditional outbound is slowly decaying.

This does not mean distribution is gone. It means generic distribution no longer works.

The companies that break through are building advantages that are hard to replicate. They rely on signals others do not have. They focus on timing instead of volume. They prioritize relevance and context over templates. They develop channels that competitors cannot copy with a tool or a playbook.

The strongest founders I met were not obsessing over subject lines or open rates. They were rethinking how they reach customers altogether.

Going forward, every serious business will need its own distribution engine, designed around its product, its data, and its users rather than borrowed tactics

3. Every company will need to invent its own channel

The old playbook of picking a channel and scaling it is slowly breaking down.

SEO is crowded. Cold outbound is increasingly constrained.

Paid acquisition is expensive and brittle. Social feeds work on rules that change without warning.

What is emerging in its place is more tailored and more intentional.

Companies are building hybrid channels that blend multiple surfaces. They are investing in owned networks rather than rented attention. They are embedding themselves deeply into existing ecosystems through integrations. They are using community, product usage, and collaboration as distribution loops. In many cases, the product itself becomes the channel.

Some teams will stitch together pieces of existing channels. Others will create something entirely new. In both cases, distribution stops being a downstream marketing activity and becomes a core product decision.

This shift feels familiar. Over the last decade, infrastructure thinking moved from the backend into product strategy. Distribution is now following the same path.

4. Hardware is where AI feels underpriced

This was the most unexpected takeaway for me.

AI inside hardware still feels early and underexplored. When it works, the impact is immediate and physical. It changes what people can do in the real world, not just what appears on a screen.

• You see the potential across several fronts.
• Self-driving systems that reduce human load.
• Brain-computer interfaces that restore or extend capability.
• AR devices with real spatial awareness.
• Autonomous drones.
• Robotics transforming logistics and healthcare.

In software, competition often centers on marginal improvements. Better summaries. Faster copilots. Cleaner dashboards. The bar is high, and any advantage tends to be short-lived.

In hardware, the bar is different. Progress is harder, but the surface area for impact is much larger. When a system works, it unlocks new behaviors rather than incremental efficiency.

Building in this space demands more capital, longer timelines, and a higher tolerance for risk. The trade-off is that the upside is also more durable.

My sense is that many of the enduring AI companies of the next decade will not look like traditional SaaS businesses at all.

5. Where this leaves me

I am continuing to build VibeTM, a context-aware orchestration layer for operations teams.

Along the way, we made a deliberate choice to return all angel capital and bootstrap in-house. It gives us the freedom to move at our own pace, go deep on architecture, and compound value over time rather than optimize for short-term optics.

As I see the landscape shift, a few beliefs keep getting reinforced.

Engineering depth matters again. Systems thinking outlasts feature velocity. The most durable companies are built on advantages that are hard to see from the outside.

If anything, this trip sharpened my appetite for hard problems.

Until next time ✌️