Growing wellness businesses with AI • Founder @kenko_ai
San Francisco, CA
Joined May 2018
- Tweets1.3K
- Following1.1K
- Followers191
- Likes7.7K
Vistar Singh retweeted
accidentally said app instead of domain-specific harness and they kicked me out of SF
Vistar Singh retweeted
Astra is very good at 3D modeling, and I can't wait for all of you to experience it, for now here is a little walkthrough on how I built the demo house for our launch blog post. From a Blender scene to a Unreal Engine 5 walkable experience. 🧵
Vistar Singh retweeted
NVIDIA’s strategy: commoditize intelligence
NVIDIA made three large acquisitions / investments in the space of a few days: Hugging Face, Poolside and Perplexity. (Two rumored, but seem likely)
All three target the same goal: help build a robust open source ecosystem as a counterweight to OpenAI and Anthropic.
Hugging Face is the distribution layer — the GitHub of open-source AI, where models, datasets and tooling live.
Poolside is the model-building layer — NVIDIA is paying $6B to license its “Model Factory” technology, while investing another $1B in the company. The obvious application is accelerating NVIDIA’s own Nemotron/open-weight model efforts.
Perplexity is the application layer — a scaled consumer product that proves open models can support products competing directly with the closed-model labs. NVIDIA is reportedly discussing another investment at a 30b+ valuation.
Put them together and the strategy becomes clearer:
Poolside Model Factory → Models → Hugging Face → developers → applications like Perplexity → more demand for NVIDIA GPUs.
NVIDIA doesn’t need to beat OpenAI or Anthropic itself. It needs to make sure there are thousands of credible alternatives to them.
Because a world dominated by a handful of vertically integrated closed-model companies is ultimately bad for the company selling the picks and shovels.
NVIDIA wants intelligence to commoditize.
Compute is where it wants the scarcity to remain.
PS: One of the biggest implications of this is that open source shifts the economics to the inference vendors (who are also nvda customers) from the model labs (who are also customers).
Vistar Singh retweeted
opportunity cost analysis always leads to the same practical conclusion: you should never enjoy anything
Vistar Singh retweeted
2 things are happening at once:
- AI coding models are converging. For 95% of tasks, most people can’t tell whether GPT-5.6 Sol, Fable 5, Kimi K3, GLM-5.2, or Qwen 3.8 Max, is smarter.
- GLM-5.3 dropped today, only 743B parameters. Smaller, cheaper, open models are coming fast.
The center of gravity may be shifting faster than people think.
Vistar Singh retweeted
“Am I shadowbanned?”
“Is X fair?”
“Why am I seeing this post?”
We want the public to be able to answer these themselves. Today we’re giving an unprecedented level of transparency into the X algorithm so people can audit, critique and even help improve it. Enjoy & send feedback.
Have to admire @UseCorgi for going for every opportunity they can find. In the world where only brand and distribution matter, you can layer as much product as possible if you have the talent to support it (re: employer brand).
What is the logic behind building AI product manager startup?
Everyone has that built internally by now.
What is the value-add that I am missing 🧐
Vistar Singh retweeted
If you join a startup and they give you:
- a MacBook
- Codex and Claude subscriptions
- your job offer on a Sunday night
- a random urgent task on your first day
- a founding team with kids
you’re probably in the right place.
If they give you:
- a Windows laptop
- one responsibility for the next six months
- Outlook
- PowerPoint decks
- an office on the outskirts of town
good luck. You’re going to need it.
Vistar Singh retweeted
This argument gets thrown at consumer / prosumer startups all the time.
IMO, it’s almost always deeply flawed. Here’s why:
- Products that feel magical might seem easy to use, but they are almost never easy to build. There is a lot of thought and craft that goes into them. You can reverse engineer the product but not the intuition behind it. This will keep the startup’s product ahead of the game
- If a bigco could do it, they would have done it. I have seen many cases of a bigco attempting to replicate what seems like a “thin” product and failing to the degree that it never even ships. To the Siri example - we’ve been talking about a functional Siri for 15 years, and for three years now post-LLMs. Apple has 170k engineers and a $4.5T market cap…if it was easy to fix Siri, they would have fixed Siri
- Turning a large boat is hard. Sometimes, fixing or improving a legacy product isn’t actually about the technical complexity of doing so. Large companies have infinitely more bureaucratic hurdles and political battles that prevent progress. They also have existing business lines that they need to be careful about cannibalizing, especially when they are accountable to quarterly earnings reports. Google has a whole army of VPs holding back innovation 😂
- Big companies are stuck in their existing product interfaces. When you have 100M users on your product, changing how something little works in that product - or even introducing something new - is a big deal (which, to the point above, could mess up those numbers you’re responsible for!) The risk of screwing up something that is technically “working” at major scale is often not worth it. They are also stuck in their existing operating systems - whereas a startup can ship a product or feature that is device / OS-agnostic. Apple is unlikely to build a Wispr Flow clone that works on Android - so if you have an Apple laptop and a Google phone, you’re out of luck
- New products can achieve lock-in before incumbents catch up. This is especially true for product with network effects or compounding data advantages. Say Apple did perfectly replicate Granola tomorrow. I already have 2.5+ years of very valuable meeting data stored there, and it has context on me that is not transferable or ingestible by Apple. If you’re asking “what if X big co builds this….” the startup product usually has enough psychological, behavioral, or contextual loyalty that it is too late to just rip and replace it
Vistar Singh retweeted
Correct take. 99% of tokens in the world will get consumed in an enterprise context.
They will be for writing code, combing through insane amounts of life sciences research, automating manufacturing, securing enterprises, detecting fraud, and thousands of other tasks like this that have significant economic value. This is both where the cost of AI can be most justified but also where the impact of parallel workers doing tasks on your behalf is most useful.
Even the agents that most impact consumers will often be wrapped up as just end to end services where the consumer doesn’t think about there being an AI system on the other end, like faster insurance or onboarding to a bank.
The key though is it’s going to take years for diffusion of agents across the economy because workflows have to re-engineered to incorporate AI, which is a complete change to how most work happens today. Anyone who’s expecting this to happen overnight should be updating their timelines.
yeah, no.
the average consumer has a near zero need for offloading cognitive work. what they need is their netflix and their doordash — your job as a consumer is literally to chill tf out and stay happy and motivated
if you dont intuit this as obvious you're most likely neck-deep in a muck of motivated reasoning as a consumer founder / VC
okay you can use agents to search the web and synthesize findings on what to buy. okay you have cognitive chores like filing tax returns and sifting thru email. sounds like great features to bake into google and gmail!
on the other hand, your job as a knowledge worker is to convert your cognition into something of economic value — and then because of competitive market dynamics, you'll also need all the cognitive amplification that's available on the market.
the biggest problems need the most cognition. the consumer takes care of problems scoped to themselves. the knowledge workers together take care problems concerning society.
when we look back, we'll remember how AI agents helped us cure cancer, not how your agent helped you order pizza over imessage.
Vistar Singh retweeted
I've now added Wispr, Notion, Granola, Fireflies, Zoom, Gemini, Cluely, Otter, Fathom, Read AI, and Wingrep to every meeting.
Each produces slightly different meeting notes.
I send all of them to Claude and ChatGPT, which push the summaries into Slackbot and Salesforce AI. Linear agents automatically create tickets while Notion agents create follow-ups in Notion. A Zapier workflow then uses 500k tokens to draft a follow up email in Superhuman.
None of it works and all of it is wrong but man, is it beautiful.
AGI is here.
Vistar Singh retweeted
The hardest segment
in B2B:
Sales-led SMB sales
You have to be wildly, wildly efficient and effective
And you have to further fight the inherent churn in SMBs
CAC has to be really, really low, team has to execute truly relentlessly
Most aren't built for this.
No one wants a model monopoly. Not governments. Not enterprises. Not users.
When monopoly is off the table, cost-efficiency and trust become the last battlefield.
That makes models a critical commodity. Powerful, yes. But ultimately fungible.
Fungible models means the middle layer thrives. Not as the winner, but as a critical, enduring part of the stack.
Here's the part most people miss though: the world is complex. Intelligence and the middle layer will constantly balance each other out. Trust, security, policy, and economics won't let either side dominate forever.
