Vin Agarwal

Building @composio | MIT PhD | IITB

[Composio @composio]

For people running Claude Code with `--dangerously-skip-permissions`, we just shipped Enhanced Controls.

It protects you at the tool layer, by blocking your agent from executing destructive actions.

Now available in Claude Code, Codex, and Cursor.

[sarah @sarahfim]

how to let claude cook overnight without it cooking you

[Vin Agarwal @vin_agarwal]

@KaranVaidya6 @aiDotEngineer This is going to be a fire talk!!

[Karan Vaidya @KaranVaidya6]

Excited to be a sponsor

[Timothy Kassis @TimothyKassis]

I'm hiring a few part-time AI engineering interns for @k_dense_ai. Fully remote or in-person in Palo Alto, CA. My agent will interview your agent which must not make up any false information about you. DM me your agent's chat interface or contact info.

[Timothy Kassis @TimothyKassis]

AI co-scientists are getting powerful fast. The real question for scientists is no longer “can this generate an answer?” It’s: can I verify the path from question to conclusion? We wrote about why the next bottleneck for AI in science is verification, not capability.

[Karan Vaidya @KaranVaidya6]

We’ve decided to gather together the best builders in the bay… on the bay. Our 89-foot yacht will bring together the 30 best applicants to have them battle it out as we sail out under the bridge for a final sunset demo.

Is this a good idea? A terrible one? Who knows. But @composio along with @nebiusai and @tavilyai is determined to find out. And the winner is taking home a DGX Spark and $50k in credits.

Apply by posting what you’ve built and tagging @ship_builders, @composio, @nebiusai, @tavilyai

[K-Dense @k_dense_ai]

@antigravity Nice, but better yet, integrate Scientific Agent Skills! with 70+ major scientfic databases and 140+ skills validated by the scientific community. Over 26k GitHub stars and 160k+ scientists using it worldwide! https://t.co/PwNqGKVRpM

[Vin Agarwal @vin_agarwal]

@k_dense_ai @antigravity Agreed

[Timothy Kassis @TimothyKassis]

Pre-registration is the TDD of science. Lock your hypotheses and decision rules before you look at the outcomes, and p-hacking has nowhere to hide. We built Science Superpowers to give research agents that discipline by default. Inspired by Superpowers, reimplemented for doing science with data. https://t.co/2bI1u4ykDE

[K-Dense @k_dense_ai]

Introducing Science Superpowers — a complete computational-science methodology for AI research agents. It makes your agent do science with data the way a careful scientist would, not the way an eager autocomplete would.
https://t.co/3gFVS9iOFh

[K-Dense @k_dense_ai]

New Agent Skill: Nextflow + nf-core (@nextflowio)

Now your coding agent can run community pipelines, write & nf-test modules/subworkflows, configure Docker/Singularity/Conda, scale to SLURM or the cloud, and debug all while implementing best practices.

Try it now in Scientific Agent Skills: https://t.co/7Rl4XNz8NK

[K-Dense @k_dense_ai]

New skill from K-Dense: LiteParse in Scientific Agent Skills — built for fast, local research paper ingestion.

Your AI co-scientist can now:

* Parse PDFs and supplementary files on your machine (no document cloud)
* Pull layout-preserved text with bounding boxes for grounded citations and RAG
* OCR scanned papers, protocols, and figure-heavy supplements
* Batch-ingest literature folders for reviews and evidence synthesis
* Grab page screenshots when the model needs figures, tables, or charts
* Efficient extraction for the papers you already have

https://t.co/7Rl4XNz8NK

[Timothy Kassis @TimothyKassis]

@slavov_n Here is our take: https://t.co/vwhVkw1ToA

[Timothy Kassis @TimothyKassis]

https://t.co/8T6njZQMYh

[Timothy Kassis @TimothyKassis]

We at @k_dense_ai have had early access to Gemini Managed Agents and been testing them for the past several weeks. Great offering competing directly with @AnthropicAI Claude Managed Agents.

[Timothy Kassis @TimothyKassis]

Great work by the @GoogleDeepMind team! https://t.co/TvmsqG3pXH

[K-Dense @k_dense_ai]

"Kosmos is now the first agent that can compress months of drug development into weeks, from the earliest stages of scientific discovery through to FDA approval."

Strong unproven claim, but we agree that platforms like K-Dense and Kosmos are not too far from being able to do that.

Great year ahead!

[K-Dense @k_dense_ai]

https://t.co/vUIH9RTNg7

[Timothy Kassis @TimothyKassis]

@sama That's exactly how we think about our AI co-scientist at @k_dense_ai