Ashwin Rajan

Product Guy. Thinker. Tinkerer. Founder at

[Ashwin Rajan @asystemone]

@sama is going to win.

[Ashwin Rajan @asystemone]

My sentiments exactly 👇

[edwin 🐸 @pescatios]

If AI adoption is actually this low, as someone who’s worked in crypto for years, I can’t help but see the same mistake happening all over again.

We spent years trying to get normal people to care about wallets, chains, tokens, gas, bridging, and all the machinery underneath the experience instead of abstracting it away behind products people simply wanted to use. AI increasingly feels like it’s doing the same thing with models, prompts, tokens, context windows, agents, tools, memory, harnesses, and all the other technical know-how you increasingly need to actually squeeze the full value out of these systems.

The current AI paradigm asks an insane amount from the user. You have to recognize what problems AI can solve in the first place, imagine what the system might be capable of, figure out what you want, translate an abstract goal into instructions, break complex tasks down into something it can act on, provide the right context, decide what to delegate, and then evaluate, verify, and iterate on the output. Before AI can reduce your cognitive load, it often requires a ton of cognitive load just to figure out how to make it useful.

And believe it or not, a lot of people simply do not think at that level of abstraction. A blank prompt box might be one of the most flexible interfaces we’ve ever created, but flexibility is not the same thing as usability. You have to imagine the capability before you can benefit from it. We’ve effectively made the user responsible for discovering the product’s capabilities, designing their own workflow, and inventing their own interface to the intelligence underneath it.

I also think AI power users have already developed the exact same blind spot crypto power users had. Once you spend enough time with this stuff, it all starts feeling obvious. We mistake learned fluency for intuitive UX and forget how completely alien it still is to everyone else.

And the gap might actually be bigger than it was with crypto. The average person isn’t just less knowledgeable about AI than the people on here. A lot of people have almost no mental model of what these systems are capable of, and plenty actively view AI as threatening, inauthentic, or even morally wrong. At its worst, crypto was scammy, complicated, or cringe. AI touches people’s sense of work, creativity, truth, agency, and what it even means to be human.

So I don’t buy the idea that AI kills apps. There are really two ways out of this. Either agents become good enough to understand us, anticipate what we need, and proactively act on our behalf without fucking anything up, or apps continue doing what good apps have always done: constrain the possibility space and turn complicated technology into an obvious experience. The first is much harder than people want to admit.

If anything, AI massively expands what apps can do. The opportunity is not to keep building “AI products.” Find net new consumer experiences that are only possible because of AI, put the problem and use case first, hide the underlying complexity, and see how far you can get without mentioning AI at all.

There’s almost a pre- and post-AI way of experiencing computers now, and most of us on here are already designing from the other side of that divide, while the rest of the world is still living pre-singularity, unaware of what the singularity even is, let alone that they may already be living through its beginning.

[Ashwin Rajan @asystemone]

The harness is the product.

[a16z @a16z]

Call center jobs grew ~4% annually for 15 years

Now they’re shrinking 4% annually

Charts of the Week: https://t.co/6UOlZZPpiE

[Ashwin Rajan @asystemone]

The central tension in figuring out the value of generative AI is between relationship and content.

[Ashwin Rajan @asystemone]

Is it time for @Apple to buy @Bose? -- I think so.

[a16z @a16z]

not trying is the only guaranteed failure

[Ashwin Rajan @asystemone]

Design today for the world in which you will be disrupted.

[Lyn Alden @LynAldenContact]

People want Jarvis.

Or an AI version of Alfred.

They want Jarvis/Alfred to handle all mundane details in their life, and we are nearing that.

Right now, they have localized chat and decent agents that still struggle with basics but are on the trajectory.

[Ashwin Rajan @asystemone]

In the age of AI, teams and execution might be the only real moats left.

[Ashwin Rajan @asystemone]

Good gosh, Clippy has reincarnated ...but this time its cuter and more useful.
#codex

[Alex Veremeyenko @alex_verem]

The Chinese president stood on a stage in Shanghai and laid out China's entire AI playbook in one speech.

It was his first-ever in-person appearance at the World AI Conference. I went through the whole thing and pulled out everything that matters.

- he opened with his signature maxim that great changes unseen in a century are unfolding across the world.

- he said AI development should not be a solo performance by a single country but a symphony of international cooperation.

- he reaffirmed China's commitment to open source AI in the name of openness and shared benefit.

- he warned against overstretching the concept of national security in AI, where one country puts its own security above everyone else's.

- he said China opposes the emergence of new historical injustices in AI, one of the strongest-worded lines in the speech.

- he pledged 5,000 AI training opportunities for developing countries over the next five years, naming ASEAN, the Arab League, the African Union, CELAC, the SCO, and BRICS.

- he committed to giving 30 countries access to a Chinese AI weather system that provides early disaster warnings.

- a day before the speech, 29 countries signed the agreement creating a new World AI Cooperation Organization headquartered in Shanghai.

strip away the politics and one thing stands out to me. he didn't pitch benchmarks or chatbots. he pitched AI as infrastructure, weather warnings for countries that lose thousands of lives to storms they never saw coming, and training programs for regions the AI boom has skipped.

meanwhile most of the Western AI conversation revolves around which lab ships the next frontier model.

I don't care who wins the race. I care whether the computing power reaches the people who need it.

The full speech is below, and it's worth your time.

[Ashwin Rajan @asystemone]

Finally. We have shifted from the world of technical architecture to a world of information architecture.

🚀

[Ashwin Rajan @asystemone]

One of the stickiest moats in product, which I think will remain valid in the AI era, is "functional fixedness".

Functional fixedness is the idea that users become entrenched into a specific pattern in using an interface. Once that pattern becomes crystallised, it becomes progressively harder to break - even when newer, simpler patterns to do the same things are surfaced to them.

That's why what started as a workaround out of a sub-optimal UI pattern entrenches users into becoming their default use pattern, even after the core UI problem the workaround solved for is eliminated.

Nowhere are we more creatures of habit than in our use of technology.

[Ashwin Rajan @asystemone]

Perhaps the most persuasive evidence of an individual's ability to build is the quality of their professional portfolio of work.

After all, in an time when anyone can build, being able to tell one's own professional story compellingly - and represent their body of work in the most articulate way possible - can be seen as the foundation for their ability to build.

[Ashwin Rajan @asystemone]

#tokenminiming

[Bill Gurley @bgurley]

This is what’s causing Anthropic to aggressively beg for govt protection (see below). Customers are finding cheaper alternatives. Keeping employees requires continuing ultra-rich secondaries ($$$) that are dependent on revenue growth. When you can’t win on the field go to DC.

[Kunal Shah @kunalb11]

The benefits of GLP-1 drugs extend far beyond weight loss, especially in a country like India.

Given India’s high burden of metabolic disease, wider adoption as these drugs become more affordable and go off patent could trigger a significant wave of usage and significant second order effects.

[Ashwin Rajan @asystemone]

Payment failure is a telling indicator of potential churn in a number of verticals. In others, it is a rather poor indicator. Often it need to be combined with other signals for confidence.

Conversational AI products solve for this intrinsically, in that their product usage provides an unprecedented wealth of context to qualify the payment failue signal.

For instance, if I fail my payment on @GammaApp, (where I am being subsidised to retain month after month, and not really using the product for value work, unfortunately) payment failure can be a high quality signal of impending churn.

On @ChatGPTapp on the other hand, a payment failure - seen in the context of all the high-value cases I use the product for, combined with my stickiness within them - means @sama needs to lose no sleep about his money going anywhere.

[Ashwin Rajan @asystemone]

The smaller and tighter the agent harness, the more you gain trust, control and confidence for scalability.

Said differently - failure modes and error rates compound at the jagged frontier, so aim to maintain the smoothest possible, most lossless frontier at the earliest stages of harness development - because it’s exactly those tiny early losses that add up - and constrain tightly around those initial boundary conditions while scaling from there.