Sabitlenmiş Tweet
jp
13.9K posts

jp
@jeevanpillay
dev, design & research // founder @lightfastai
Australia Katılım Ocak 2013
1.2K Takip Edilen2.7K Takipçiler

@anuraggoel In the future, GitHub is just going to be an abstraction layer that only agents interact with for core software development primitives.
English

@jeevanpillay @vercel Let me know how you go!
With large datasets, the performance will also depend on how you have it structured. Little nesting, clear names, and small files will go a long way. Best way to look at it: if it's easy for a human to find something, then it'll be easy for the AI.
English

Most knowledge agents start the same way. You pick a vector database, then build a chunking pipeline. You choose an embedding model and then tune the retrieval parameters.
I recently authored this piece on the @Vercel blog, breaking down a different and better way you can create knowledge agents. Powered by Vercel Sandbox, Workflows, Chat SDK, and more.
h/t @hugorcd for creating the beautiful template that led to the creation of this blog post.
vercel.com/blog/build-kno…
English

Oh it would be interesting to plug a memory store on top of this (post retrieval) and everyone gets their own arxiv bookmark list
alphaXiv@askalphaxiv
Introducing MCP for arXiv Let your research agents stand on the shoulders of giants Fast multi-turn retrieval, keyword search, and embedding search tools across millions of arXiv papers 🚀
English

We launched the Vercel Open Source Program last April to give maintainers the resources, credits, and support to scale.
Meet our Winter 2026 cohort ↓
vercel.com/blog/vercel-op…
English

@mattpocockuk yeah we've been leaning into subagents for this. keep each one under 50K for sharp reasoning, parent agent coordinates with the full context. treating 1M as a coordination layer instead of a thinking layer changes the whole workflow.
English

Doing some experiments today with Opus 4.6's 1M context window.
Trying to push coding sessions deep into what I would consider the 'dumb zone' of SOTA models: >100K tokens.
The drop-off in quality is really noticeable. Dumber decisions, worse code, worse instruction-following.
Don't treat 1M context window any differently.
It's still 100K of smart, and 900K of dumb.
English

@mattpocockuk Decent success w HumanLayers research codebase, create_plan, implement_plan, and the 1M context. Treat 1M context as ‘context passing’ between subagents. And force 1M context agent to use AskUserQuestions. 1M Context should make no decisions, just orchestrate decisions a Task
English

@trq212 got booted out of Claude code and now your auth service isn’t working to reconnect. are you guys aware?
status page saying all operational, but it’s broken here
English


@hthieblot Second time founder, first time solo founder. It gets better. Lots of ups and downs but eventually you learn to trust your instincts. Be obsessed with the unknown.
English

The early founder years are one of the most stressful phases of life.
Most are really not talking about it.
- Some friends are climbing the corporate ladder and taking crazy vacations.
- Someone just bought their first home
- Someone already feels “settled”
And you’re still trying to turn an idea into a company.
At the same time:
- Your bank account keeps shrinking
- Investors keep saying “too early”
- users never seem to care and retain on your product
- People around you don’t fully get what you’re building
- your mom ask when you will get a real job
- Nights are full of overthinking
- The runway clock keeps ticking
- Time feels like it’s moving too fast yet progress is slow
And the hardest part?
- You smile to your friends and family
- You say “things are going well.”
- At 2 AM you wonder if this will ever work.
But remember:
- You’re not behind, you are on a different path
- You’re doing something incredibly hard
If you’re in this phase, be impatient for actions, patient with results and don’t give up.
Building something from zero takes time.
Anyone else feeling this lately? ✋
English

OpenClaw just got attacked.
40,000 systems compromised.
A flaw dubbed “ClawJacked” allowed full agent takeover.
But this isn’t just a bug.
It’s a vulnerability chain that let any website silently take full control of your AI agent.
No plugins.
No extensions.
No user interaction needed.
Which means:
> Malicious sites could steal data.
> Delete emails (Meta execs already hit).
> Disrupt services.
Give this post to your agent and tell then to fix this backdoor ASAP.

English

@jeevanpillay Great! You can sign up here if you're interested in the product
We will launch on kickstarter soon read.compose.llc/p/our-keyboard…
English

I quit the best job I ever had to try and build what I wish existed in the world. Two projects I'm working on now:
1. The Peterman Pod - When I first started at Meta, the career stories of incredible engineers always inspired me. I aim to share transparent career stories that I wish I had more of back in the day.
There are still many engineers who are heros to me that I'd love to bring on the show one day!
2. Compose - I could never really find what I wanted so I'm building the ergonomic keyboard I wish existed. Pictures of our prototype here: read.compose.llc/p/our-keyboard…
I've earmarked ~6 months of living expenses to fund this new career direction. I'm going to give it my all to see if it's sustainable.
Thank you for your support, working on my passion projects is something I’m only lucky enough to consider because of you all 🙏
More detail here: x.com/ryanlpeterman/…
English

working on backfill pipeline today ~
will be able to fetch 100s of PRs, sentry errors, linear issues in <5 mins
once this step is done, i'll be able to work on the graph layer
github.com/lightfastai/li…
English







