Ashwin Rajan

Product Guy. Thinker. Tinkerer. Founder at

[Ashwin Rajan @asystemone]

When I hear "AI isn't profitable yet", I think of this:

Uber took 14 years to achieve its first full-year operating profit in 2023, following its 2009 founding. During that time, the company raised $25 billion in private capital from over 95 venture capitalists, sovereign wealth funds, and corporate investors prior to its 2019 IPO. [Source: Gemini]

If investors are willing to progressively sink in a cool 25 bil and wait 13 years on one company to transform one sector, what would their willingness and patience look like betting on a transformational technology that looks like it will impact every known sector, and probably create a few entirely new ones in the process?

[Ashwin Rajan @asystemone]

The core reason why a platform like YouTube or TikTok, or even X, works at scale is because the real value comes from small, unknown, non-brand creators and publishers trying to build a brand. They can't see meaningful engagement if their content isn't good going on great. This drives them to be hyperfocused on quality about all else, to the best of their ability.

Whereas creators who've made brand are now focussed on monetising that brand, an incentive that tends to undermine the quality their content developed to get there. In this setup, it is increasingly the algorithm that cuts through the noise, connecting pure quality to pure user intent and attention, with established brands polluting the waters in trying to intervene. So, paradoxically, an emerging brand strategy might be to simply to "unbrand".

[The Information @theinformation]

Is Anthropic altering model performance to force costly upgrades? Chapter Co-Founder and CEO @CobiBGantz outlines a shift his team recently noticed:

"Anthropic has kind of taken a page out of the Apple playbook, where they have...decreased the quality of the prior models right before they launched the new model, so that the new model feels better."

[Ashwin Rajan @asystemone]

I believe that most advances in AI systems will not come from starting with: "AI is already better than humans at.."; it will come from showing and challenging the limits of these systems compared to humans, and hence give them room - and direction - to develop. So, starting out with the embodiment hypothesis of the human brain and intelligence can be a pretty solid place.

[Ashwin Rajan @asystemone]

For the briefest moment it seemed like it was worth going pro on @GeminiApp for Nano Banana, but with recent updates to the image-gen capabilities on @OpenAI ChatGPT, Gemini still does not have me for now. And I’m thankful 🙏 I don’t have to live with sub-par (although I admit rapidly improving) Google UX. But this points to how fast differentiating edges are being won and lost by the leading players. And to the rapid commoditisation of high value capabilities, which as a user, I’m thrilled about.

[Ashwin Rajan @asystemone]

AI today gives edge to an individual professional - IF three things are in place:

1. Existing alpha / edge in a domain
2. Intent
3. AI skills.

1. AI amplifies the current domain edge of the professional; More capability > better potential outcomes. Or seen half-full: garbage in, garbage out.

2. Intent: this is the big one, you might have the potential but are you genuinely psyched about AI? Anything spanning mild skepticism (which I confess I do have, in certain dimensions) to full-blown paranoia / anger / mistrust / dismissiveness is going to show up and create really sticky, long-term drag. Fix the mindset to orbit-shift the acceleration.

3. This is the work. Usually follows - as a mix of intrinsic and extrinsic motivational factors - if the above two are in place.

Map each of your team members on these dimensions to build a projection of your AI-adoption readiness.

[Ashwin Rajan @asystemone]

Was only a question of time before this wall was hit.

I have no clue why (and how!) anyone came up with the harebrained idea of 'tokenmaxxing' - & am deeply perplexed that some of the smartest people in the world jumped on that bandwagon.

The notion that 'token consumption equates to productivity' just screams - "Never heard of the canonical principle of product development called Outcomes Over Outputs".

Worse, tokens aren't even an output. They are merely an input. Productivity / outcomes / impact moving key needles are two degrees downstream of that: manage token efficiency & cost > output quality > outcomes.

[Gergely Orosz @GergelyOrosz]

I can now probably say this:

Two months ago, inside Anthropic someone suggested building a token leaderboard.

A heated internal debate followed and the decision was made to *never* ever do it… because several people inside Anthropic simply thought ahead of the consequences

[Ashwin Rajan @asystemone]

@ylecun @francoisfleuret What about a designer?

[Ashwin Rajan @asystemone]

@AnthropicAI Claude is bsing about token limits. Used it for 3-4 hours; hit 90% warning. So ran a 172 character prompt req. a 7-bullet summary of content within current context window, & it hit 100% instantly without completing. That math don't math. Is why @Uber says no??

[Ashwin Rajan @asystemone]

@AnthropicAI Claude is literally bsing about token limits. Used it for 3-4 hours; hit 90% warning. So ran a 172 character prompt req. a 7-bullet summary of content within current context window, & it hit 100% instantly without completing. That math don't math.

[Ashwin Rajan @asystemone]

Any friction against the widespread & deep adoption of both consumer & business AI (cost, learning curve, uneven team use etc.) is met by an even more powerful countervailing force: once a user onboards successfully onto AI for a specific use case, there is no going back.