Copilot’s biggest issue isn’t cost. It isn’t product. It isn’t even the brand dilution.
They have the insurmountable problem of having the actual stupidest user base of any AI product.
The average Lovable user understands basic software development better than Copilot users.
How many merchant AI accelerator companies will there be?
Two, of course.
And capitalism needs the loser to have enough market share. Such is the nature of capitalism.
Everything is multi-sourced against everything.
Such is the nature of capitalism.
There is a reason why there are only two turbocharger companies, two airplane companies, and two rating agencies.
My understading is ~all hyperscalers have multiple ASIC programs, and many ASIC programs have multi-sourcing particularly at large volumes.
Makes for a bad narrative, and maybe correct to question $AVGO competitive advantage over time (rising competition is never good for stocks), but $GOOG partnering with Mediatek for parts of TPU doesn't 1) mean they're replacing $AVGO or 2) mean that substantial volume is at risk of moving.
As Gavin Baker said: "What is preventing the bubble from forming and us from achieving smoother for longer are the conservative folks in the memory shitcos."
@PythiaR I tried to say that more as a forward-looking thing, as I think their revenues will accelerate as people need to buy MCP, more seats, more data, etc.
And that is only starting.
What total comp does yours need to be for your boss to think $1,500 is more than enough in AI spending for you?
$UBER median comp is $203,000 per year.
Basically, Dara thinks 10% is enough to pay as the tithe for Anthropic.
God, Cloudflare is the fucking AGI play
In three hours since I started Clawdbot, it has already created an extension for me that will 10x the amount of api calls I do when I visit a particular website.
I think everyone should build tool-shaped objects
1- There’s a lot of beauty in tool shaped objects
2- You’ll eventually get bored of building tool-shaped objects and you’ll do real work
3- Building a tool-shaped object will help you a lot to build a decent mental model of AI
Ok I would like to clarify
There is a risk that Opus 4.6 is like 4o for white collar workers
If you spend too much time w/ it, you can feel really good
But it's easy to get worm-holed into building tools that don't actually improve productivity
Proceed with caution traveler
@ShanuMathew93 Probably years away from this being useful.
Computers are already slow just because of the many things require your computer to do. Imagine also using multiple LLMs in parallel.