Pratik

Hyper-optimist. Building Prajvis, cross platform workflow automation without infra headaches

[Pratik @BasedInfra]

The GPU math still skews in the favourite of datacenter.
The GPUs are too expensive to run models like GLM. And even if you get them, the electricity is still expensive. You get sovereignty, not cost savings.
Unless you invest more for solar power. And then some for cooling.

[Pratik @BasedInfra]

Smart main and specialized sub-agents can be fast can be slow.
Main should also be able to clone itself into n sub-agents

[Beyang @beyang]

Which do you think you’d prefer?

Fast main agent with smart subagents?

Or smart main agent with fast subagents?

[Pratik @BasedInfra]

BCCI assembled such team and team performance actually decreased.
If everyone is a leader, it gets difficult to lead that team.

[Sahil @sahilypatel]

anthropic is assembling the most cracked team in tech

in the past 18 months, they've hired:
- peter bailis (workday cto)
- bryan mccann (you com cto)
- mike krieger (instagram co-founder)
- sam ghods (box cto)
- henry shi (super cto)
- niki parmar (adept ai cto)
- karpathy sensei (openai founding member)

and now john jumper (nobel laureate)

hard to bet against this team

[Pratik @BasedInfra]

1. Show up everyday and keep trying
2. Don’t push things for later
3. Keep smiling

[Pratik @BasedInfra]

Frontier models are becoming regulated infrastructure. If your internal agents depend directly on one model API, your workflows inherit that provider’s policy, availability, pricing, and regulatory risk.

[Pratik @BasedInfra]

Enterprises SHOULD keep Data and Models separate.
Today it was a model, tomorrow it can be a platform. You don’t want your lose your data on Anthropic platform abruptly because government thought so.

[Pratik @BasedInfra]

@MartinShkreli People should get sovereignty over their data and stop depending on model platforms.
Also switching model providers is easy if you’re not locked with your data on their platform.

[Pratik @BasedInfra]

@GergelyOrosz Technically not just another model but platform as well.
Ideally everyone should use model-agnostic platforms and have ability to switch models without losing their data. Intelligence is too important to depend on one platform and model.

[Pratik @BasedInfra]

@ThePrimeagen chat completions was pure LLM play.
OpenAI’s responses is hybrid kind of ? Because it stores things server side ?
Next step is basically what you’re talking, entire agent side on server so you talk with only their ‘agent api’/ managed agents are kind of there.

[Pratik @BasedInfra]

Why will companies use Fable with ZDR disabled ?
Seems like the release is for tech demonstration

[Pratik @BasedInfra]

@martin_casado I’m sure reverting hurt more given the cost

[Pratik @BasedInfra]

@signulll If you’re considering billions of people it’ll probably be WhatsApp and Meta’s game.
iMessage is not that prevalent out of US.

[Pratik @BasedInfra]

@sflorimm They reduced their marketing budgets and other increased theirs

[Pratik @BasedInfra]

@illscience It’ll be exciting though. Agents will be new Apps and you can just ask your Siri to ask some remote agent from an app to do something that Siri cannot do

[Pratik @BasedInfra]

@martin_casado Basically brings the lower quartile up to speed.
But if upper quartile becomes lazy, they suffer.
Those who don’t act lazy will win big basically

[Pratik @BasedInfra]

My agents remember and create jobs for themselves in the future where they either run a workflow or remind me by updating a thread.
Talk with it today and get work done way in the future.
Life is good.

[Pratik @BasedInfra]

@kitarp29 Should be other way round.
Developers will stop working for companies without unlimited token budgets.

[Pratik @BasedInfra]

@thekitze Applicable to developers. I think the executive class will still prefer iOS

[Pratik @BasedInfra]

@signulll We’ve same, but while scrolling X

[Pratik @BasedInfra]

Most of your bugs and security vulnerabilities are due to the fallbacks you added but forgot to remove.