- Joined
- Jan 2026
- How they joined
- Directly
- Now
- Thinking Machines Lab
Sources1linkedin.com
Details
LocationUnited States
UniversityMassachusetts Institute of Technology
Notes
- Currently a Member of Technical Staff at Thinking Machines Lab, a role LinkedIn dates to Aug 2026; that is about four months after his Cognition stint ended.Aug 2026
- @terminalbench confirmed his authorship on 2026-06-18: "The first three challenges, written by @andr3w_wang, each involve completing projects that would previously require a team of experts working for several months.".Jun 18, 2026
- His Cognition window (Jan to Apr 2026) covers the whole SWE-1.6 cycle: Cognition published the SWE-1.6 early preview and research update in early March 2026 and released SWE-1.6 generally in April 2026.Apr 2026
- LinkedIn lists Cognition (Devin/Windsurf) from Jan 2026 to Apr 2026 with a placeholder title of '.'.Jan 2026
- He follows Adam Zweiger ('Working on learning @thinkymachines | prev: @cognition @MIT_CSAIL'), whose path MIT -> Cognition -> Thinking Machines matches his own.Nov 2026
- In his launch thread he wrote, word for word: "Challenges are tasks that are not only large in size, but also difficult because of the technical ability needed to solve them. I'm excited to see how these tasks drive harness optimizations, gearing them toward long horizon work. Even more so, I'm excited to see the end artifacts they come up with!"Jun 18, 2026Challenges are tasks that are not only large in size, but also difficult because of the technical ability needed to solve them. I'm excited to see how these tasks drive harness optimizations, gearing them toward long horizon work. Even more so, I'm excited to see the end artifac
- Chris Barber's Frontier List poll of people who best understand recursive self-improvement (RSI) includes 'Andrew Wang, Thinking Machines' with 1 vote.Nov 2026
- He follows Terminal-Bench and Harbor builders Alex Shaw (founding MTS at Laude Institute), Ryan Marten, Steven Dillmann (Terminal-Bench-Science lead), Ludwig Schmidt and Alex Dimakis, plus @terminalbench and @frontierbench. This supports his involvement in agent-benchmark work.Nov 2026
- Cognition's SWE-1.6 preview post says the model 'learns to use bash commands for search instead of pre-defined tools because terminal commands are more ex[pressive]'. Terminal-native agent behavior is the same theme he worked on next with Terminal-Bench.Mar 2, 2026
- He follows Cheng-Yuan (Sam) Lee, whose bio reads 'Research @cognition | prev intern @windsurf | 2x ICPC World Finalist'. Lee is the clearest Windsurf-to-Cognition link in his network, but no source shows that Andrew himself came through Windsurf or Codeium.Nov 2026
- The same day he reposted @terminalbench's announcement 'Introducing Terminal-Bench Challenges! A new capability has emerged at the frontier: agents completing large-scale proj[ects]'. His own launch post got 39 likes and 11 reposts.Jun 18, 2026
- Cognition presented 'How We Scaled RL for Agentic Coding: Lessons from Training SWE-1.6' at NVIDIA GTC San Jose 2026 (session S81825). It is the main public talk on the model family he worked on; the page does not list him as a speaker.Mar 2026
Experience
Member of Technical Staff at Thinking Machines Lab
Aug 2026 - Present
. at Cognition
Jan 2026 - Apr 2026
Education
Massachusetts Institute of Technology
Bachelor's degree
Projects
SWE-1.6 (Cognition, 2026)
Terminal Bench Challenges: author of the first three challenges (launched 2026-06-18)
AchievementsAuthored the first three Terminal-Bench Challenges (2026)
Recent posts
June 2026
Can agents build complete projects that deliver real value? We’re launching Terminal Bench Challenges: 3 unsolved tasks which could make a real impact on the open source community if solved. These tasks provide a testing ground for optimizations both on the model and harness lev
Challenges are tasks that are not only large in size, but also difficult because of the technical ability needed to solve them. I'm excited to see how these tasks drive harness optimizations, gearing them toward long horizon work. Even more so, I'm excited to see the end artifac
We’re looking for more challenge ideas! Especially tasks that have interesting verification methods: adversarial testing, fuzzing with LMs, replicating closed source software, etc. And tasks in specialized domains: Game emulation, FPGAs, GPU driver programming, MoQ, etc. https:
Profiles
