Anthropic is managing some of the most complex load balancing problems in engineering history.
- Different hardware, locations, standards
- Many models
- Variance across the world, use cases, and distillation attacks
- Order of magnitude jumps in demand
- Physical and political constraints to scale with years-long lead times
People use to criticize top talent going to optimize ads at Meta and Google as a brain drain.
Top talent going to builder smarter AI for everyone feels more like brain rain.
Smarter models every few week.
SoftBank’s investor presentation is one of the greatest things ever made. I’ve been thinking about it all day. These are the real slides shown in a speech where Masayoshi Son said he wouldn’t retire for at least another decade. The goose stuff is perfect.
https://t.co/sk9cDhdWIE
Hey @tankots product feedback:
@WisprFlow should not have an empty keyboard screen during recording.
It's a waste of screen real estate.
Your competitors have already realized this, as shown in the screenshots.
A user sometimes wants to push enter while recording to break into new lines.
The user sometimes wants to move the cursor and add commas or add bullet points before speaking or while in the middle of speaking.
For however the debate goes for the "good will" of @AnthropicAI vs @OpenAI, it's becoming clearer that Anthropic is capitalism-maxing.
Anthropic has made it clear that they do not intend to provide frontier intelligence to everyday people who cannot afford $10,000s of API costs.
Devs who build around Anthropic subscriptions cannot compete with Anthropic's own products that provide tokens at a 1/10th fraction of the cost.
Anthropic's moves like banning @openclaw and creating restrictions around their subscriptions, make this clear.
Now, Anthropic is explicitly removing frontier models like Fable 5 as part of subscriptions.
For what's been said about OpenAI, they have continuously taken the stance of providing frontier model access with a generous, unrestricted subscriptions.
For however the debate goes for the "good will" of @AnthropicAI vs @OpenAI, it's becoming clearer that Anthropic is capitalism-maxing.
Anthropic has made it clear that they do not intend to provide frontier intelligence to everyday people who cannot afford $10,000s of API costs.
Devs who build around Anthropic subscriptions cannot compete with Anthropic's own products that provide tokens at a 1/10th fraction of the cost.
Anthropic's moves like banning @openclaw and creating restrictions around their subscriptions, make this clear.
Now, Anthropic is explicitly removing frontier models like Fable 5 as part of subscriptions.
For what's been said about OpenAI, they have continuously taken the stance of providing frontier model access with a generous, unrestricted subscriptions.
Hey @tankots product feedback:
@WisprFlow should not have an empty keyboard screen during recording.
It's a waste of screen real estate.
Your competitors have already realized this, as shown in the screenshots.
A user sometimes wants to push enter while recording to break into new lines.
The user sometimes wants to move the cursor and add commas or add bullet points before speaking or while in the middle of speaking.
Bit late to the "worst VC stories."
In my experience with 100s of investors and ~maybe over 1k investor interactions.
Most bad experiences fall into the following categories:
1) Indecisive (drags out for months, no wonder YC reviews highly on speed)
2) Trauma (they got burned previously and overly pattern match, weird accusatory vibes)
As someone who has raised from hundreds of investors including a16z, Founders Fund, Jump, etc., I highly recommend partnering with @lior_eth and @edenblockvc if you get the chance.
Lior is sharp, hustles more than anyone for you, and is a genuinely awesome guy.
Congrats Lior and Eden Block!
One caveat on looping
Looping is great for scale, optimization, self improvement, and for systems that are working
But when you’re building something new, I think the founder needs to stay in the loop.
Before PMF the challenge is figuring out what to build, who it’s for, and whether anyone actually wants it.
I technically went through dozens of loops building my MVP, but I was directing the loop the entire time.
My thesis is that pre PMF, a lot of the value comes from founder judgment. The small insights, customer feedback, and raw signals are often what shape the product.
Keeping a human in the loop helps you spot those raw gems and be more opinionated about the problem and solution
After PMF / escape velocity, looping becomes much more valuable because things are working and looking to scale fast.
How I've been looping for a few weeks now.
Create a skill called autopilot or autodrive.
Have the agent provide updates every fifteen mins. This is watchdog skill.
Doing. Done. Restart any stalled agents. Blocked by what, simply.
I check updates for anything I need to unblock.
Most of the time you need auto-merge or commit directly.
Note you need to spend significant upfront time planning and detailing context and end tests/evals to be able to loop for any material length of time.
For anything frontend, you need a mac for the agent to self test.
Recently watched the COIN documentary on Coinbase.
One of the things that @brian_armstrong has said that I haven't heard others say, is to just repeatedly "choose the least worst option."
I've found this mental framework to be helpful in navigating ambiguity.
Working on something ambitious is like climbing a mountain that’s covered in fog.
You can't see a clear path to the top. You have to take a few steps into the unknown to be able to see the next few steps in front of you. Inevitably, sometimes you’ll end up a local maximum and have to backtrack. That’s fine, just keep moving.