Sohil Kshirsagar

love when the audience asks thoughtful questions, s/o to @patricia_dugan for hosting last week’s Agents & APIs meetup pictured: @sohilkshirsagar telling war stories and showing how we repro traffic

Super cool learning @sohilkshirsagar’s backstory and what he and @marcel7an are building at @usetusk @ the @getpostman Agents and APIs meetup tn!

Sharing about execution boundaries for agents here next week. If you love thinking deeply about agent infra, come say hi!

growing up i've organized nightclub events, duolingo streak parties, and even a music festival now i'm hosting a meetup series for the people building infra for coding agents. first event is Thu, May 7 in SF featuring: - @aditabrm at @reductoai - @MelkeyDev at @vercel - @juecd__ at @usekernel - @jyt4n at @usetusk

AI is transforming software development, but more code means more pull requests, more edge cases, and more QA pressure on engineering teams. Tusk, a Y Combinator startup, catches bugs that slip past both AI agents and humans using AI-enabled tests based on real production traffic. Built on Amazon Bedrock, Tusk flags issues before code merge so teams can focus on building great products.

🌉 April 30 - Agents & APIs SF Developer Meetup ✨ Ben Warren @bwarrn, Founder, @mesa_dot_dev ✨ Ben Backus @BenBackus4, SWE, @greptile ✨ Nihar Kumara, Sr. SWE, Postman! ✨ Sohil Kshirsagar, Co-Founder & CTO, @usetusk Space is limited! Register 👉 luma.com/agents-AI-apis…

AI is helping teams ship 10x more code. Testing hasn’t caught up. @usetusk is fixing that by turning production traffic into automated tests. Featuring @marcel7an + how @auth0 secures it all 👇

Your AI-generated tests are expensive assertions against hallucinated data. We ran test generation evals with and without Tusk-recorded API traffic across 5 quality dimensions using LLM-as-a-judge. Traffic-augmented tests won 67% overall, and 100% on mock data realism. Without it, the agent repeatedly hallucinated order IDs and error formats.

hello automated e2e testing 👋 we want Tusk to be the all-in-one AI test engineer that understands how users use your app in the real world we built a CLI that: - pulls real user sessions to determine the test plan - creates a BDD-style yaml - spins up a browser agent to hit your critical paths simulate thousands of realistic user sessions and test the living hell out of your app. beta soon.

Nobody starts a company to write tests. But when 300k+ devs and a quarter of the Fortune 500 run production workloads on your open-source repo, every commit goes under a microscope. How @promptfoo added 2k+ high-quality unit tests without slowing shipping velocity 🧵

Ran our benchmarks after moving our Python SDK's core instrumentation paths to Rust. Median throughput gains: → FastAPI: +92.6% → Django: +88.7% → Flask: +66.5% Teams using Tusk for live traffic testing will now see even better performance. Upgrade to `tusk-drift-python-sdk[rust]==0.1.22` to experience it.

Recently moved core instrumentation/serialization paths in our Python SDK to Rust. Median throughput vs non-Rust in framework benchmarks: - Django: +88.7% - FastAPI: +92.6% - Flask: +66.5% Profiling showed lower CPU time in key span-processing paths, consistent with the throughput gains. If you already use the Python SDK, upgrade to `tusk-drift-python-sdk[rust]==0.1.22` to get on the fast path. If you run a Python API and want tests recorded from live traffic, details here: github.com/Use-Tusk/drift…

RT @usetusk: Catch @marcel7an's talk on how to catch edge cases with automated API testing using recorded traffic Bonus live demo of our M…

scripted talk 🙅 live demo 🙋‍♂️ had a wonderful time presenting at @auth0 Ski Camp AI, chatting with devs, and handing out swag full video available

Our Drift MCP is out 🤘 Connect Claude and Cursor to traces and spans recorded by Tusk

Our team is doing a live demo of the Drift MCP at @auth0's Ski Camp AI event today. Come through, say hi, and get some swag 👇

Code review should factor in your live traffic, so we built a suite of tools to power the next frontier of code verification. Has been super fun working on this! Try it out, grab a coffee while our CLI agent sets it up for you.

Introducing Tusk 2.0. Tests pass. AI code review looks good. But your product still breaks. The future of code review is a deterministic tool that tests changes against real user behavior. @usetusk turns your production traffic into a self-maintaining API test suite. Cover thousands of real-world edge cases without writing a test again.

product so hot off the press even the sf fire department had to check it out tomorrow, tue 9am PT big reveal for @usetusk 👀