- Joined Cognition
- Jul 2025
- How they joined
- With Windsurf
- Before Windsurf
- Startup Building AI AssistantMember of Technical Staff
- Now
- River AI
- Left Cognition after SWE-1.7 (released 2026-07-08; LinkedIn shows Cognition Jul 2025 - Jul 2026) and became "GPU Burner" at River AI from Aug 2026.Aug 2026
- On SWE-1.6 Preview he wrote: "I'm incredibly proud to have contributed to the release of SWE-1-6-preview. While it might be a small step for the industry, it is a massive milestone for Cognition. We are just getting started, and the potential is limitless. Try it out in Windsurf today!"Mar 1, 2026I'm incredibly proud to have contributed to the release of SWE-1-6-preview. While it might be a small step for the industry, it is a massive milestone for Cognition. We are just getting started, and the potential is limitless. Try it out in Windsurf today! https://lnkd.in/gdCFRRmY
- Joined Cognition in July 2025 via the Windsurf acquisition, in his own words: "I joined Cognition last July following the company's acquisition of Windsurf."Jan 5, 2026Published: 2026-01-05T02:07:48.214Z Before starting regular work of 2026, this feels like the ideal time to summarize the past year and prepare for the next chapter. First, a long overdue career update: I joined Cognition last July following the company's acquisition of Windsurf. I'm now working with a great team, including Silas Alberti, Cheng-Yuan Lee, Ray Wang, Carlo Baronio, Connor Fogarty Ben Pan Emile Timothy Anand and so on.... When I joined Codeium/Windsurf in August 2024, I shared my thoughts about the future of software engineering on LinkedIn : " I am convinced that the field of software engineering is on the brink of a major transformation over the next 2-5 years. ... Programming languages, with their strict semantics, stand to gain significantly from the recent breakthroughs in large language models (LLMs). " (source https://lnkd.in/gbEY_Kfs) Just two months after my start, and prior to the official launch of the Windsurf IDE, the momentum within the internal team's adoption of Windsurf made it very clear that we were on the verge of a breakthrough. This trend was further solidified in early 2025 when Andrew Kapathy coined the term "vibe coding" which perfectly captured the essence of this shift. This trajectory eventually led to the Windsurf acquisition SAGA. 2025 was a huge year for AI and AI coding, with a ton of exciting stuff happening. To name a few: (1) the awesome gains from RLVF, (2) the shockwave of Deepseek R1/GRPO, (3) the long list of SOTA open weight models and various model architecture innovations, (4) the wide adoption of toolcalls/agentic capability and reasoning capability of top models. ... I'm super thankful I got to work at Scale AI, Windsurf, and then Cognition over the last 5 years, all right in the middle of this AI boom. 2026 is here now, it's full of possibilities, and the competition is definitely going to get tougher. One thing to call out is that model training is no longer exclusive to elite AI labs. (1) GPUs are much more accessible and affordable now than years ago. (2) In the academic research section, training algorithms/recipes are widely researched, shared and discussed. (3) More than that, the open-source training/inference frameworks also gained significant momentum. (4) On the data front, the data collection/preparation pipelines gradually become an industry de-facto standard. Collectively, these advancements empower startups to move into model training, allowing them to develop and vertically optimize models best suited for their specific product scenarios. Cognition is a leader in this area, and we are just starting. and most importantly, we are hiring. :) (https://lnkd.in/gdjGDMJw
- Listed as an author on Cognition's SWE-1.5 launch post (2025-10-29), byline "By Jacob Teo, Nikhil Jha, Connor Fogarty, Gary Chang ...".Oct 29, 2025
- On X after the SWE-1.6 GA launch: "SWE-1.6 is the best alternative solution in the market right now, if you feel the pain the high token cost. It gets significant boost from SWE 1.5 in terms of problem solving capability and model "user experience", Please do give it a try."Apr 7, 2026SWE-1.6 is the best alternative solution in the market right now, if you feel the pain the high token cost. It gets significant boost from SWE 1.5 in terms of problem solving capability and model "user experience", Please do give it a try.
- Named his Cognition teammates publicly: "I'm now working with a great team, including Silas Alberti, Cheng-Yuan Lee, Ray Wang, Carlo Baronio, Connor Fogarty Ben Pan Emile Timothy Anand and so on...."Jan 5, 2026Published: 2026-01-05T02:07:48.214Z Before starting regular work of 2026, this feels like the ideal time to summarize the past year and prepare for the next chapter. First, a long overdue career update: I joined Cognition last July following the company's acquisition of Windsurf. I'm now working with a great team, including Silas Alberti, Cheng-Yuan Lee, Ray Wang, Carlo Baronio, Connor Fogarty Ben Pan Emile Timothy Anand and so on.... When I joined Codeium/Windsurf in August 2024, I shared my thoughts about the future of software engineering on LinkedIn : " I am convinced that the field of software engineering is on the brink of a major transformation over the next 2-5 years. ... Programming languages, with their strict semantics, stand to gain significantly from the recent breakthroughs in large language models (LLMs). " (source https://lnkd.in/gbEY_Kfs) Just two months after my start, and prior to the official launch of the Windsurf IDE, the momentum within the internal team's adoption of Windsurf made it very clear that we were on the verge of a breakthrough. This trend was further solidified in early 2025 when Andrew Kapathy coined the term "vibe coding" which perfectly captured the essence of this shift. This trajectory eventually led to the Windsurf acquisition SAGA. 2025 was a huge year for AI and AI coding, with a ton of exciting stuff happening. To name a few: (1) the awesome gains from RLVF, (2) the shockwave of Deepseek R1/GRPO, (3) the long list of SOTA open weight models and various model architecture innovations, (4) the wide adoption of toolcalls/agentic capability and reasoning capability of top models. ... I'm super thankful I got to work at Scale AI, Windsurf, and then Cognition over the last 5 years, all right in the middle of this AI boom. 2026 is here now, it's full of possibilities, and the competition is definitely going to get tougher. One thing to call out is that model training is no longer exclusive to elite AI labs. (1) GPUs are much more accessible and affordable now than years ago. (2) In the academic research section, training algorithms/recipes are widely researched, shared and discussed. (3) More than that, the open-source training/inference frameworks also gained significant momentum. (4) On the data front, the data collection/preparation pipelines gradually become an industry de-facto standard. Collectively, these advancements empower startups to move into model training, allowing them to develop and vertically optimize models best suited for their specific product scenarios. Cognition is a leader in this area, and we are just starting. and most importantly, we are hiring. :) (https://lnkd.in/gdjGDMJw
- Joined Codeium/Windsurf in August 2024 (LinkedIn dates ML Engineering at Windsurf Sep 2024 - Jul 2025) and recalled that "Just two months after my start, and prior to the official launch of the Windsurf IDE, the momentum within the internal team's adoption of Windsurf made it very clear that we were on the verge of a breakthrough."Jan 5, 2026Published: 2026-01-05T02:07:48.214Z Before starting regular work of 2026, this feels like the ideal time to summarize the past year and prepare for the next chapter. First, a long overdue career update: I joined Cognition last July following the company's acquisition of Windsurf. I'm now working with a great team, including Silas Alberti, Cheng-Yuan Lee, Ray Wang, Carlo Baronio, Connor Fogarty Ben Pan Emile Timothy Anand and so on.... When I joined Codeium/Windsurf in August 2024, I shared my thoughts about the future of software engineering on LinkedIn : " I am convinced that the field of software engineering is on the brink of a major transformation over the next 2-5 years. ... Programming languages, with their strict semantics, stand to gain significantly from the recent breakthroughs in large language models (LLMs). " (source https://lnkd.in/gbEY_Kfs) Just two months after my start, and prior to the official launch of the Windsurf IDE, the momentum within the internal team's adoption of Windsurf made it very clear that we were on the verge of a breakthrough. This trend was further solidified in early 2025 when Andrew Kapathy coined the term "vibe coding" which perfectly captured the essence of this shift. This trajectory eventually led to the Windsurf acquisition SAGA. 2025 was a huge year for AI and AI coding, with a ton of exciting stuff happening. To name a few: (1) the awesome gains from RLVF, (2) the shockwave of Deepseek R1/GRPO, (3) the long list of SOTA open weight models and various model architecture innovations, (4) the wide adoption of toolcalls/agentic capability and reasoning capability of top models. ... I'm super thankful I got to work at Scale AI, Windsurf, and then Cognition over the last 5 years, all right in the middle of this AI boom. 2026 is here now, it's full of possibilities, and the competition is definitely going to get tougher. One thing to call out is that model training is no longer exclusive to elite AI labs. (1) GPUs are much more accessible and affordable now than years ago. (2) In the academic research section, training algorithms/recipes are widely researched, shared and discussed. (3) More than that, the open-source training/inference frameworks also gained significant momentum. (4) On the data front, the data collection/preparation pipelines gradually become an industry de-facto standard. Collectively, these advancements empower startups to move into model training, allowing them to develop and vertically optimize models best suited for their specific product scenarios. Cognition is a leader in this area, and we are just starting. and most importantly, we are hiring. :) (https://lnkd.in/gdjGDMJw
- River AI was founded by xAI co-founder Igor Babuschkin and raised $1.1B in a round led by General Catalyst, announced 2026-08-11, the same month Chang joined.Aug 11, 2026
- His view on Cognition training its own models: "Collectively, these advancements empower startups to move into model training, allowing them to develop and vertically optimize models best suited for their specific product scenarios. Cognition is a leader in this area, and we are just starting." He closed with "and most importantly, we are hiring. :)"Jan 5, 2026Published: 2026-01-05T02:07:48.214Z Before starting regular work of 2026, this feels like the ideal time to summarize the past year and prepare for the next chapter. First, a long overdue career update: I joined Cognition last July following the company's acquisition of Windsurf. I'm now working with a great team, including Silas Alberti, Cheng-Yuan Lee, Ray Wang, Carlo Baronio, Connor Fogarty Ben Pan Emile Timothy Anand and so on.... When I joined Codeium/Windsurf in August 2024, I shared my thoughts about the future of software engineering on LinkedIn : " I am convinced that the field of software engineering is on the brink of a major transformation over the next 2-5 years. ... Programming languages, with their strict semantics, stand to gain significantly from the recent breakthroughs in large language models (LLMs). " (source https://lnkd.in/gbEY_Kfs) Just two months after my start, and prior to the official launch of the Windsurf IDE, the momentum within the internal team's adoption of Windsurf made it very clear that we were on the verge of a breakthrough. This trend was further solidified in early 2025 when Andrew Kapathy coined the term "vibe coding" which perfectly captured the essence of this shift. This trajectory eventually led to the Windsurf acquisition SAGA. 2025 was a huge year for AI and AI coding, with a ton of exciting stuff happening. To name a few: (1) the awesome gains from RLVF, (2) the shockwave of Deepseek R1/GRPO, (3) the long list of SOTA open weight models and various model architecture innovations, (4) the wide adoption of toolcalls/agentic capability and reasoning capability of top models. ... I'm super thankful I got to work at Scale AI, Windsurf, and then Cognition over the last 5 years, all right in the middle of this AI boom. 2026 is here now, it's full of possibilities, and the competition is definitely going to get tougher. One thing to call out is that model training is no longer exclusive to elite AI labs. (1) GPUs are much more accessible and affordable now than years ago. (2) In the academic research section, training algorithms/recipes are widely researched, shared and discussed. (3) More than that, the open-source training/inference frameworks also gained significant momentum. (4) On the data front, the data collection/preparation pipelines gradually become an industry de-facto standard. Collectively, these advancements empower startups to move into model training, allowing them to develop and vertically optimize models best suited for their specific product scenarios. Cognition is a leader in this area, and we are just starting. and most importantly, we are hiring. :) (https://lnkd.in/gdjGDMJw
- His August 2024 thesis for joining Codeium, which he re-quoted: "I am convinced that the field of software engineering is on the brink of a major transformation over the next 2-5 years. ... Programming languages, with their strict semantics, stand to gain significantly from the recent breakthroughs in large language models (LLMs)."Aug 2024Published: 2026-01-05T02:07:48.214Z Before starting regular work of 2026, this feels like the ideal time to summarize the past year and prepare for the next chapter. First, a long overdue career update: I joined Cognition last July following the company's acquisition of Windsurf. I'm now working with a great team, including Silas Alberti, Cheng-Yuan Lee, Ray Wang, Carlo Baronio, Connor Fogarty Ben Pan Emile Timothy Anand and so on.... When I joined Codeium/Windsurf in August 2024, I shared my thoughts about the future of software engineering on LinkedIn : " I am convinced that the field of software engineering is on the brink of a major transformation over the next 2-5 years. ... Programming languages, with their strict semantics, stand to gain significantly from the recent breakthroughs in large language models (LLMs). " (source https://lnkd.in/gbEY_Kfs) Just two months after my start, and prior to the official launch of the Windsurf IDE, the momentum within the internal team's adoption of Windsurf made it very clear that we were on the verge of a breakthrough. This trend was further solidified in early 2025 when Andrew Kapathy coined the term "vibe coding" which perfectly captured the essence of this shift. This trajectory eventually led to the Windsurf acquisition SAGA. 2025 was a huge year for AI and AI coding, with a ton of exciting stuff happening. To name a few: (1) the awesome gains from RLVF, (2) the shockwave of Deepseek R1/GRPO, (3) the long list of SOTA open weight models and various model architecture innovations, (4) the wide adoption of toolcalls/agentic capability and reasoning capability of top models. ... I'm super thankful I got to work at Scale AI, Windsurf, and then Cognition over the last 5 years, all right in the middle of this AI boom. 2026 is here now, it's full of possibilities, and the competition is definitely going to get tougher. One thing to call out is that model training is no longer exclusive to elite AI labs. (1) GPUs are much more accessible and affordable now than years ago. (2) In the academic research section, training algorithms/recipes are widely researched, shared and discussed. (3) More than that, the open-source training/inference frameworks also gained significant momentum. (4) On the data front, the data collection/preparation pipelines gradually become an industry de-facto standard. Collectively, these advancements empower startups to move into model training, allowing them to develop and vertically optimize models best suited for their specific product scenarios. Cognition is a leader in this area, and we are just starting. and most importantly, we are hiring. :) (https://lnkd.in/gdjGDMJw
- At Scale AI (Sep 2020 - Nov 2023), Staff Engineer and Manager who "Led the computer vision (2D) labeling product team", building the web image/video labeling studio, AI pre-labeling, a data collection pipeline "collaborating with Meta Reality Lab", Outlier.ai platform work for GenAI, and a WebGL/PixiJS image segmentation overhaul.Sep 2020
- Open source: his LinkedIn credits him with developing "the open-source PSP integration framework, Sleet" at Bolt (github.com/BoltApp/sleet, a Go library for payment providers such as Adyen, Braintree, Authorize.Net, CyberSource).2019
Developed the teaching mode, which consists of two main components: 1. A smart AI agent system, known as the large action model, that breaks down complex tasks into multiple steps and utilizes advanced AI models to execute each step efficiently. 2. A workflow designed to learn solutions for complex tasks by observing and analyzing how people complete these tasks through a series of actions in a web browser. GeoLocation Platform: Designed and developed a comprehensive GeoLocation platform supporting GPS, cell tower data, and IP-based geolocation. Integrated geocoding and reverse geocoding vendors, providing APIs for feature teams to create location-aware assistant experiences. Observability Platform: Spearheaded the creation of an observability platform by researching and selecting the optimal vendor. Implemented log and metric system integrations, rewrote the logging library, and migrated the existing codebase. This initiative established a robust foundation for long-term DevOps operations and significantly enhanced bug triage efficiency.
Led the computer vision (2D) labeling product team, developing Scale's web-based image/video labeling studio. Managed the product and engineering roadmap to align with key metrics. Built a high-energy engineering team that drove rapid business growth and emphasized quality software development. Key Projects: AI Assistant Labeling: Integrated AI pre-labeling into the data workflow. Data Collection Pipeline: Developed a pipeline from scratch, securing major clients and collaborating with Meta Reality Lab. Outlier.ai Development: Enhanced platform for multi-domain usage, supporting GenAI growth. Image Segmentation Overhaul: Improved product with WebGL and PixiJS, boosting resolution capacity and UX speed. Compensation Platform Extension: Expanded platform for GenAI workloads with new features like direct deposit and tax form generation.
Led Bolt's financial product team, building the core engine for its independent credit card payment gateway and integrating major solutions like Stripe, PayPal, Afterpay, and Affirm. Developed the open-source PSP integration framework, Sleet, enhancing Bolt's position as a one-stop solution for payment integrations. Spearheaded the creation of Bolt’s PCI-compliant credit card data vault, optimizing processing costs and user experience. Recruited top talent, doubling the engineering team size in a year. Fostered a collaborative and inclusive environment, mentoring junior engineers and supporting their career growth. Established a long-term architectural vision to scale services and improve code quality continuously.
As the first engineer on the Uber Freight team, I developed the Minimum Viable Product (MVP) from scratch. This included creating the core backend infrastructure supporting freight business logic and building user-facing platforms: web-based interfaces for carriers and shippers, a mobile app for drivers, and an internal web portal for back-office operations. I also engineered the load search and recommendation engine, which utilized both rule-based guidelines and machine learning approaches to optimize freight matching. My work laid a strong technical foundation that contributed to Uber Freight's success in the logistics industry.
As part of Uber's Marketplace Platform Team, I managed one of the largest Node.js worker clusters globally and initiated the transition to Go, scaling the backend to support rapid business expansion. I led the design and implementation of a redundancy and isolation solution across four data centers with seven deployments each, achieving zero global outages. As a key contributor to the platform's re-architecture, I successfully led the migration to a new platform. Additionally, I resolved traffic restriction issues for Uber in China and developed a deep understanding of geolocation-related business logic, enhancing the platform's reliability and scalability.
Published: 2026-01-05T02:07:48.214Z Before starting regular work of 2026, this feels like the ideal time to summarize the past year and prepare for the next chapter. First, a long overdue career update: I joined Cognition last July following the company's acquisition of Windsurf. I'm now working with a great team, including Silas Alberti, Cheng-Yuan Lee, Ray Wang, Carlo Baronio, Connor Fogarty Ben Pan Emile Timothy Anand and so on.... When I joined Codeium/Windsurf in August 2024, I shared my thoughts about the future of software engineering on LinkedIn : " I am convinced that the field of software engineering is on the brink of a major transformation over the next 2-5 years. ... Programming languages, with their strict semantics, stand to gain significantly from the recent breakthroughs in large language models (LLMs). " (source https://lnkd.in/gbEY_Kfs) Just two months after my start, and prior to the official launch of the Windsurf IDE, the momentum within the internal team's adoption of Windsurf made it very clear that we were on the verge of a breakthrough. This trend was further solidified in early 2025 when Andrew Kapathy coined the term "vibe coding" which perfectly captured the essence of this shift. This trajectory eventually led to the Windsurf acquisition SAGA. 2025 was a huge year for AI and AI coding, with a ton of exciting stuff happening. To name a few: (1) the awesome gains from RLVF, (2) the shockwave of Deepseek R1/GRPO, (3) the long list of SOTA open weight models and various model architecture innovations, (4) the wide adoption of toolcalls/agentic capability and reasoning capability of top models. ... I'm super thankful I got to work at Scale AI, Windsurf, and then Cognition over the last 5 years, all right in the middle of this AI boom. 2026 is here now, it's full of possibilities, and the competition is definitely going to get tougher. One thing to call out is that model training is no longer exclusive to elite AI labs. (1) GPUs are much more accessible and affordable now than years ago. (2) In the academic research section, training algorithms/recipes are widely researched, shared and discussed. (3) More than that, the open-source training/inference frameworks also gained significant momentum. (4) On the data front, the data collection/preparation pipelines gradually become an industry de-facto standard. Collectively, these advancements empower startups to move into model training, allowing them to develop and vertically optimize models best suited for their specific product scenarios. Cognition is a leader in this area, and we are just starting. and most importantly, we are hiring. :) (https://lnkd.in/gdjGDMJw
hmm, will there be a day that people use agent to port rust implementations of foobar software to ASM then claim victory?
There should be a leaderboard of best coding languages for AI Coding. Rust so far leads on this, but just by luck. It is intentionally designed for such effort.
SWE-1.6 is the best alternative solution in the market right now, if you feel the pain the high token cost. It gets significant boost from SWE 1.5 in terms of problem solving capability and model "user experience", Please do give it a try.
Weekend random thoughts. We are in an agentic era, and tool calling is the most critical foundation for communication between LLMs and agents. The JSON Schema of a tool call defines the contract between them. 🧵 1/3 #LLM #AIagents
I have been doing something very similar in the past few months, Deeper in the core, it is a workflow to manage context of your research and work and share it with agent so that it could help you better! This could also be used as your reference note
2026 is the year of agent, large scope enterprise adoption is happening!
I'm incredibly proud to have contributed to the release of SWE-1-6-preview. While it might be a small step for the industry, it is a massive milestone for Cognition. We are just getting started, and the potential is limitless. Try it out in Windsurf today! https://lnkd.in/gdCFRRmY
Try out the new version of Devin, faster and fully remote and async,
Before starting regular work of 2026, this feels like the ideal time to summarize the past year and prepare for the next chapter. First, a long overdue career update: I joined Cognition last July following the company's acquisition of Windsurf. I'm now working with a great team, including Silas Alberti, Cheng-Yuan Lee, Ray Wang, Carlo Baronio, Connor Fogarty Ben Pan Emile Timothy Anand and so on.... When I joined Codeium/Windsurf in August 2024, I shared my thoughts about the future of software engineering on LinkedIn : " I am convinced that the field of software engineering is on the brink of a major transformation over the next 2-5 years. ... Programming languages, with their strict semantics, stand to gain significantly from the recent breakthroughs in large language models (LLMs). " (source https://lnkd.in/gbEY_Kfs) Just two months after my start, and prior to the official launch of the Windsurf IDE, the momentum within the internal team's adoption of Windsurf made it very clear that we were on the verge of a breakthrough. This trend was further solidified in early 2025 when Andrew Kapathy coined the term "vibe coding" which perfectly captured the essence of this shift. This trajectory eventually led to the Windsurf acquisition SAGA. 2025 was a huge year for AI and AI coding, with a ton of exciting stuff happening. To name a few: (1) the awesome gains from RLVF, (2) the shockwave of Deepseek R1/GRPO, (3) the long list of SOTA open weight models and various model architecture innovations, (4) the wide adoption of toolcalls/agentic capability and reasoning capability of top models. ... I'm super thankful I got to work at Scale AI, Windsurf, and then Cognition over the last 5 years, all right in the middle of this AI boom. 2026 is here now, it's full of possibilities, and the competition is definitely going to get tougher. One thing to call out is that model training is no longer exclusive to elite AI labs. (1) GPUs are much more accessible and affordable now than years ago. (2) In the academic research section, training algorithms/recipes are widely researched, shared and discussed. (3) More than that, the open-source training/inference frameworks also gained significant momentum. (4) On the data front, the data collection/preparation pipelines gradually become an industry de-facto standard. Collectively, these advancements empower startups to move into model training, allowing them to develop and vertically optimize models best suited for their specific product scenarios. Cognition is a leader in this area, and we are just starting. and most importantly, we are hiring. :) (https://lnkd.in/gdjGDMJw
DeepSeek recently published a series of papers. DeepSeek Math-V2 features self-verification, while mHC proposes 'manifold-controlled multi-head residual connections' as an alternative to Kaiming He's famous ResNet. The Engram paper highlights the separation of computation and knowledge/memory lookup. Taking these papers together—along with the previous DSA—builds significant hype for the rumored DeepSeek V4/R2. The architectural innovation here is massive. Additionally, the mHC and Engram experimental models are both 27B, similar to the ~27B scale of last year's DeepSeek V3 Lite!
weekend fun facts, QWen updated its model Qwen3-Coder-30B-A3B-Instruct's intermediate_size from 5472 to 6144, (https://lnkd.in/gs92VCaE) that means the model weight size got increased by 12%, through the upgrade of the weight snapshot. It is not a big deal, but pretty interesting to see this type of silent change.
