At @icmlconf 🇰🇷 this week and excited to host the 5th @DL4Code this Friday! We have an amazing lineup of speakers: - @Diyi_Yang (@stanfordnlp/@humansand) - @syhw (@AIatMeta) - @lschmidt3 (@Stanford/@AnthropicAI) - @LittleXie5 (@hkust/@Alibaba_Qwen) plus great papers and custom DL4C swag!📍HALL B2 Also very proud of two of my past interns presenting at the main conference: - @terryyuezhuo on "Training Language Model Agents to Find Vulnerabilities with CTF-Dojo" Tue 2pm HALL A #3314 - @setlur_amrith on "Reuse your FLOPs: Scaling RL on Hard Problems by Conditioning on Very Off-Policy Prefixes" Wed 10:30am HALL A #1902 Come say hi! ☕️👋

Excited to share that 𝘁𝗵𝗲 𝟱𝘁𝗵 @DL4Code 𝘄𝗼𝗿𝗸𝘀𝗵𝗼𝗽 𝗶𝘀 𝗰𝗼𝗺𝗶𝗻𝗴 𝘁𝗼 #ICML2026🇰🇷. Grateful to everyone who helped make this such an energizing space for the field! Also, a belated update on our position paper 𝗛𝘂𝗺𝗮𝗻𝘀 𝗮𝗿𝗲 𝗠𝗶𝘀𝘀𝗶𝗻𝗴 𝗳𝗿𝗼𝗺 𝗔𝗜 𝗖𝗼𝗱𝗶𝗻𝗴 𝗔𝗴𝗲𝗻𝘁 𝗥𝗲𝘀𝗲𝗮𝗿𝗰𝗵. What I’m especially proud of in the paper is the core argument: the next frontier for coding agents is not just more autonomy, but better human collaboration. As agents get stronger, the real bottleneck is increasingly whether people can align with them, steer them, verify their outputs, and trust them in real workflows. We turn that observation into concrete research agenda for building coding agents that are genuinely more useful. The seed for this paper came from an early conversation with the awesome @KLieret at #NeurIPS. I later discussed the idea with @jyangballin, which led to further conversations with @ZhiruoW @Diyi_Yang, and the final paper was led by these folks together with a fantastic group of co-authors. Parts of the argument were also shaped by wonderful discussions at the #DL4C workshops at #ICLR and #NeurIPS last year. Check it out! zorazrw.github.io/files/posi…