We're launching EnIGMA, our state-of-the-art AI agent for offensive cybersec! It uses tools like Ghidra & pwntools, can debug, connect to servers, and exploit vulnerabilities to solve CTF challenges. Built with researchers from Princeton, NYU, and TAU. enigma-agent.github.io

In a year or two from now, 'fine-tuning' will become synonymous with 'training' (as used in the good old ML days). LLMs will be seen more widely as starting points, just like weight initialization or choosing the number of layers for a Transformer. Pick a starting point, curate

Sierra partnered with Casper to launch Luna 2.0, their AI agent delivering 24/7 personalized customer support. From helping with mattress purchases to driving lifelong loyalty, Luna 2.0 is transforming the shopping experience!💤✨️ Learn more: sierra.ai/customers/casp…

We're launching SWE-bench Multimodal to eval agents' ability to solve visual GitHub issues. - 617 *brand new* tasks from 17 JavaScript repos - Each task has an image! Existing agents struggle here! We present SWE-agent Multimodal to remedy some issues Led w/ carlos 🧵

Today we're excited to announce a new way to interact with Sierra agents: voice. Learn more about how this new capability is transforming customer interactions in our latest blog post.: sierra.ai/blog/sierra-sp…

Today we're releasing Common Sense Agents, a new backbone for agentic creative computing: 💻 Windows VMs for safe and repeatable workflows 🔧 Long workflows broken down into reusable tasks 🦾Support for off the shelf agents like Claude ⌛️ Data recording + finetuning infra

The biggest mistake we can make right now is not dreaming big enough, especially w.r.t AI

SWE-agent 1.0 is the open-source SOTA on SWE-bench Lite! Tons of new features: massively parallel runs; cloud-based deployment; extensive configurability with tool bundles; new command line interface & utilities.

The best thing about SWE-agents and tools like cursor is the amount of additional agency they provide us

In the AI age, agent reliability is key, and Sierra’s 𝜏-bench is setting the standard—shaping academic research, industry applications and next-generation development. Read more: sierra.ai/blog/tau-bench….

Interesting tidbits on using dedicated "thinking" steps in agents from Anthropic Also loved seeing full pass^k curves for τ-bench - measuring this was the primary motivation of the benchmark, not just avg scores!

I’m at ICLR to present a poster and give a talk, both related to the second half blogpost. See you there if you wanna chat about it :)

Humans evolved to communicate so we could coordinate better. But these days, it feels like we communicate so much, yet coordinate so little.

Successful agents are the result of collaboration between teams: engineering, operations, customer experience, and marketing. Yet every platform available today except Sierra forces businesses to optimize for one group over another. Our Agent OS enables both no code and

Like all great products, the best agents are the product of many teams working together — some technical, some non-technical. Sierra’s Agent OS uniquely supports both no code and programmatic agent development, enabling customer experience and engineering teams alike to build

Can GPT, Claude, and Gemini play video games like Zelda, Civ, and Doom II? 𝗩𝗶𝗱𝗲𝗼𝗚𝗮𝗺𝗲𝗕𝗲𝗻𝗰𝗵 evaluates VLMs on Game Boy & MS-DOS games given only raw screen input, just like how a human would play. The best model (Gemini) completes just 0.48% of the benchmark! 🧵👇

As we optimize model reasoning over verifiable objectives, how does this affect human understanding of said reasoning to achieve superior collaborative outcomes? In our new preprint, we investigate human-centric model reasoning for knowledge transfer 🧵:

Last year, we introduced 𝜏-bench, a benchmark for evaluating AI agents on realistic, multi-step tasks involving tool use and domain-specific constraints. It surfaced a critical limitation in LLM-based agents: low repeatability, even under identical conditions. Now, we’re

Learn more: sierra.ai/blog/benchmark…

Today we announced a set of major advances to our agent benchmark, 𝜏-bench. This new benchmark, 𝜏², introduces the notion of "dual control", where AI agents are challenged not just to reason and act, but to coordinate, guide, and assist a user in achieving a shared objective.