# Jeffrey Ling

![Jeffrey Ling](/resources/cognition/portraits/jeffrey-ling.webp)

Software engineer at Cognition building Devin; ex-Waymo Planner ML, Google AI Resident (NLP)  
Cognition  
United States

## At Cognition

- Role: Human Software Engineer (Engineering)
- Joined: Jan 2024
- How they joined: Directly
- Before Cognition: Co-Founder, Stealth Startup
- Overlapped before Cognition with other Cognition people at: Google; Harvard University (https://www.autumn.ai/resources/cognition/worked-together)

## Identity
- LinkedIn: https://www.linkedin.com/in/jeffrey-ling-92163879
- X: @spdling
- GitHub: https://github.com/jeffreyling
- Website: https://jeffreyling.github.io/
- Domain: cognition.ai

## Notes
- Jeffrey Ling worked with Rhythm Garg, Moritz Stephan, and Raymond Feng on SWE-check, according to a LinkedIn post by Rhythm Garg. (https://www.linkedin.com/posts/rhythm-garg_it-was-awesome-working-with-moritz-stephan-activity-7449944470625701888-xhOt)
- Jeffrey Ling's personal website blog includes a guest post titled "Guest Post: I Migrated This Website" written by Devin, described as "an AI software engineer built by Cognition." (https://jeffreyling.github.io/blog.html)
- Jeffrey Ling is a coauthor of the paper 'Scene Transformer: A unified architecture for predicting multiple agent trajectories' (arXiv:2106.08417), listed among authors including Jiquan Ngiam, Vijay Vasudevan, Zhengdong Zhang, Hao-Tien Lewis Chiang, Rebecca Roelofs, Alex Bewley, Chenxi Liu, Ashish Venugopal, David Weiss, Ben Sapp, Zhifeng Chen, and Jonathon Shlens. (https://arxiv.org/abs/2106.08417)
- Rhythm Garg wrote of working with Jeffrey Ling and others: "It was awesome working with Moritz Stephan, Jeffrey Ling, and Raymond Feng on SWE-check; in the future, many valuable product harnesses (even down to the feature!) will have their own custom-trained models backing them." (https://www.linkedin.com/posts/rhythm-garg_it-was-awesome-working-with-moritz-stephan-activity-7449944470625701888-xhOt)
- In the guest post, Devin wrote: "Hi there! I'm Devin, an AI software engineer built by Cognition. Jeffrey asked me to redesign his website, and I'm writing this guest post to tell you about it." (https://jeffreyling.github.io/blog.html)
- The Scene Transformer paper was noted as ICLR 2022 in its arXiv comments. (https://arxiv.org/abs/2106.08417)
- Public headline says "Human Software Engineer @ Cognition". (https://www.linkedin.com/in/jeffrey-ling-92163879)
- Posted on X: "@var_epsilon Michael is goated" (https://x.com/spdling/status/2113344717098496293)
- Posted on X: "Use /walden-review if you want to be a 100x engineer like Walden" (https://x.com/spdling/status/2112656402741334506)
- Posted on X: "My favorite moment from using Devin Oncall internally: it solved an incident by itself while Andrew was afk https://t.co/wCRhbyjMgK" (https://x.com/spdling/status/2110794818884673573, https://t.co/wCRhbyjMgK)
- Posted on X: "@moritz_stephan has always been focused on building real value rather than working on toy problems. Engines reflect that philosophy for agents!" (https://x.com/spdling/status/2108233941199319130)
- The Scene Transformer paper's abstract states the model predicts the behavior of all agents jointly and achieves state of the art performance across two popular autonomous driving datasets. (https://arxiv.org/abs/2106.08417)
- Jeffrey Ling wrote a blog post titled "VAEs are Bayesian" about variational autoencoders being theoretically motivated by probability theory and Bayes' rule. (https://jeffreyling.github.io/blog.html)
- Jeffrey Ling wrote that he attended NIPS 2017 in Long Beach, CA, calling it "my first conference!" and "an incredible experience." (https://jeffreyling.github.io/blog.html)
- Previous roles include Co-Founder at Stealth Startup; Software Engineer at Waymo; AI Resident at Google; Research Engineer at Vicarious AI; Software Engineering Intern at Google. (https://www.linkedin.com/in/jeffrey-ling-92163879)
- Education history includes BA Computer Science & Math, Harvard University. (https://www.linkedin.com/in/jeffrey-ling-92163879)
- Jeffrey Ling lists a public project titled "Building Cognition" at app.devin.ai, described as "Building Devin, the first AI software engineer." (https://jeffreyling.github.io/)
- Jeffrey Ling wrote a blog post titled "Reinforcement Learning as Optimization" explaining that RL is simply optimizing a particular loss function. (https://jeffreyling.github.io/blog.html)
- Jeffrey Ling wrote a "Welcome" blog post stating: "I'm starting this space so that I can organize my ideas, provide interesting and educational content to the public, and practice writing." (https://jeffreyling.github.io/blog.html)
- Jeffrey Ling lists a public project titled "Planner ML @ Waymo," described as "Researched and deployed ML models for autonomous driving." (https://jeffreyling.github.io/)
- Jeffrey Ling lists a public project titled "Research @ Google," described as "NLP and BERT research as an AI Resident at Google NYC." (https://jeffreyling.github.io/)
- Jeffrey Ling (GitHub username jeffreyling) authored the latest commit, titled "update readme", in the CognitionAI/devin-swebench-results repository. (https://github.com/CognitionAI/devin-swebench-results)
- Jeffrey Ling contributed to the CognitionAI/devin-swebench-results repository, which hosts "Cognition's results and methodology on SWE-bench" and links to the blog post at https://cognition-labs.com/post/swe-bench-technical-report. (https://github.com/CognitionAI/devin-swebench-results, https://cognition-labs.com/post/swe-bench-technical-report)
- Jeffrey Ling coauthored the paper 'Ling et al. 2020' on entity linking, which is cited in this AKBC 2020 paper as showing that using large-scale pretrained language models like BERT, pretraining on Wikipedia entity links, and fine-tuning on a specific entity linking corpus leads to state-of-the-art performance. (https://www.akbc.ws/2020/assets/pdfs/iHXV8UGYyL.pdf)
- The paper is categorized under Computation and Language (cs.CL) and Artificial Intelligence (cs.AI) with DOI https://doi.org/10.48550/arXiv.1906.03158. (https://arxiv.org/abs/1906.03158, https://doi.org/10.48550/arXiv.1906.03158)
- Jeffrey Ling attended NIPS 2017 in Long Beach, CA, describing it as his first conference. (https://jeffreyling.github.io/posts/nips-2017.html)
- Jeffrey Ling and Rachit presented their short paper at the approximate Bayesian inference workshop at NIPS 2017. (https://jeffreyling.github.io/posts/nips-2017.html)
- The paper was first submitted to arXiv on 16 Sep 2016 (v1) and last revised 13 Jun 2017 (v2). (https://arxiv.org/abs/1609.04938)
- Location is United States. (https://www.linkedin.com/in/jeffrey-ling-92163879)
- Jeffrey Ling asked Devin, an AI software engineer built by Cognition, to redesign his personal website. (https://jeffreyling.github.io/posts/devin-guest-post.html)
- Devin redesigned Jeffrey Ling's site to match the look and feel of a GitHub profile page, replaced the about page with a projects section, moved blog posts to a dedicated page, and removed all Ruby and Jekyll dependencies. (https://jeffreyling.github.io/posts/devin-guest-post.html)
- Devin set up project cards on Jeffrey Ling's site for his work at Cognition, Waymo, and Google. (https://jeffreyling.github.io/posts/devin-guest-post.html)
- The launch integrated MCP into Devin, enabling Devin to use tools such as the Datadog and Sentry MCPs to pick up issues, triage them, and propose root causes based on the codebase. (https://www.linkedin.com/posts/theodormarcu_sara-xiang-federico-marcos-jeffrey-ling-activity-7353898027507441664-wGZO)
- Jeffrey Ling worked at Cognition, the company behind Devin, as of July 2025. (https://www.linkedin.com/posts/theodormarcu_sara-xiang-federico-marcos-jeffrey-ling-activity-7353898027507441664-wGZO)
- The paper's abstract states: "We show that large scale training of neural models allows us to learn high quality entity representations, and we demonstrate successful results on four domains". (https://arxiv.org/abs/2001.03765)
- The paper reports "a 64% error reduction over previous work on TypeNet (Murty et al., 2018)". (https://arxiv.org/abs/2001.03765)
- The paper reports that they "match the state-of-the-art on CoNLL-Aida without linking-specific features and obtain a score of 89.8% on TAC-KBP 2010 without using any alias table, external knowledge base or in domain training data". (https://arxiv.org/abs/2001.03765)
- The paper was submitted to arXiv on 11 Jan 2020 (v1, Sat, 11 Jan 2020 15:30:56 UTC) with DOI https://doi.org/10.48550/arXiv.2001.03765. (https://arxiv.org/abs/2001.03765, https://doi.org/10.48550/arXiv.2001.03765)
- The paper describes Ling et al. 2020 as having focused mainly on constructing general-purpose entity representations. (https://www.akbc.ws/2020/assets/pdfs/iHXV8UGYyL.pdf)
- The paper notes that its method is most similar to Ling et al. 2020 and Broscheit 2019, which also use a transformer, and differs from Ling et al. 2020 by simultaneously considering all mentions and entities in a context. (https://www.akbc.ws/2020/assets/pdfs/iHXV8UGYyL.pdf)
- The paper lists Ling et al. 2020 among prior work (with Yamada et al. 2016, 2017 and Gillick et al. 2019) that pretrains by learning distributed representations of entities directly from Wikipedia text. (https://www.akbc.ws/2020/assets/pdfs/iHXV8UGYyL.pdf)
- The paper's abstract states: 'We show that these representations significantly outperform previous work on exemplar based relation extraction (FewRel) even without using any of that task's training data.' (https://arxiv.org/abs/1906.03158)
- The paper's abstract states: 'We also show that models initialized with our task agnostic representations, and then tuned on supervised relation extraction datasets, significantly outperform the previous methods on SemEval 2010 Task 8, KBP37, and TACRED.' (https://arxiv.org/abs/1906.03158)
- Jeffrey Ling wrote of NIPS 2017: "Overall, it was an incredible experience. Six days of little sleep, skipped meals, and a tremendous amount of talks, posters, and papers to process." (https://jeffreyling.github.io/posts/nips-2017.html)
- The abstract of the paper states: "We propose a novel coarse-to-fine attention model that hierarchically reads a document, using coarse attention to select top-level chunks of text and fine attention to read the words of the chosen chunks." (https://aclanthology.org/W17-4505/)
- The abstract of the paper states: "Empirically, we find that while coarse-to-fine attention models lag behind state-of-the-art baselines, our method achieves the desired behavior of sparsely attending to subsets of the document for generation." (https://aclanthology.org/W17-4505/)
- The paper presents a neural encoder-decoder model to convert images into presentational markup based on a scalable coarse-to-fine attention mechanism, evaluated on image-to-LaTeX generation with a new dataset of real-world rendered mathematical expressions paired with LaTeX markup. (https://arxiv.org/abs/1609.04938)
- The paper's abstract states: "Our approach outperforms classical mathematical OCR systems by a large margin on in-domain rendered data, and, with pretraining, also performs well on out-of-domain handwritten data." (https://arxiv.org/abs/1609.04938)
- The paper's abstract states: "To reduce the inference complexity associated with the attention-based approaches, we introduce a new coarse-to-fine attention layer that selects a support region before applying attention." (https://arxiv.org/abs/1609.04938)
- Jeffrey Ling's redesigned site is plain HTML and CSS with no build step, no dependencies, and no Gemfile, and all four blog posts were converted from Markdown to HTML with MathJax equations in the RL and VAE posts still rendering. (https://jeffreyling.github.io/posts/devin-guest-post.html)
- Devin wrote that the entire site redesign, from start to finish, was a single conversation in which Jeffrey told Devin what he wanted, Devin explored the existing codebase, built the new site, tested it visually, fixed a few issues, and opened a PR. (https://jeffreyling.github.io/posts/devin-guest-post.html)
- Jeffrey Ling (@spdling) replied to Alex Kaplan's post about growth in AI with the word-for-word comment "spoken by the goat himself". (https://x.com/alexkaplan0/status/2065483246394118443)
- Jeffrey Ling, along with Sara Xiang and Federico Marcos, worked on the launch of integrating MCP into Devin at Cognition. (https://www.linkedin.com/posts/theodormarcu_sara-xiang-federico-marcos-jeffrey-ling-activity-7353898027507441664-wGZO)
- Theodor Marcu wrote that "Sara Xiang + Federico Marcos + Jeffrey Ling really cooked with this launch" regarding the MCP integration into Devin. (https://www.linkedin.com/posts/theodormarcu_sara-xiang-federico-marcos-jeffrey-ling-activity-7353898027507441664-wGZO)
- Jeffrey Ling is the first author of the paper "Learning Cross-Context Entity Representations from Text" (arXiv:2001.03765, cs.CL). (https://arxiv.org/abs/2001.03765)
- The paper "Learning Cross-Context Entity Representations from Text" was coauthored by Jeffrey Ling, Nicholas FitzGerald, Zifei Shan, Livio Baldini Soares, Thibault Févry, David Weiss, and Tom Kwiatkowski. (https://arxiv.org/abs/2001.03765)
- A footnote in the paper states 'Work conducted during Google AI Residency,'. (https://www.akbc.ws/2020/assets/pdfs/iHXV8UGYyL.pdf)
- The coauthors of 'Matching the Blanks: Distributional Similarity for Relation Learning' are Livio Baldini Soares, Nicholas FitzGerald, Jeffrey Ling, and Tom Kwiatkowski. (https://arxiv.org/abs/1906.03158)
- Jeffrey Ling wrote: "The poster sessions were undoubtedly the highlight of the conference." (https://jeffreyling.github.io/posts/nips-2017.html)
- Jeffrey Ling wrote: "The discussions at the posters are what research is really about - sharing awesome work with people who are interested in the fine-grained details." (https://jeffreyling.github.io/posts/nips-2017.html)
- Jeffrey Ling wrote: "Turns out everyone is doing machine learning these days!" (https://jeffreyling.github.io/posts/nips-2017.html)
- Jeffrey Ling wrote: "At NIPS, I was finally able to put a face and voice to many of these people :)" (https://jeffreyling.github.io/posts/nips-2017.html)
- Jeffrey Ling wrote: "If you're interested in a blow-by-blow recap of my NIPS experience, read below (warning: lots of details)." (https://jeffreyling.github.io/posts/nips-2017.html)
- Jeffrey Ling coauthored the paper "Coarse-to-Fine Attention Models for Document Summarization" with Alexander Rush. (https://aclanthology.org/W17-4505/)
- The paper "Coarse-to-Fine Attention Models for Document Summarization" was published in the Proceedings of the Workshop on New Frontiers in Summarization (NewSum 2017), pages 33-42, Copenhagen, Denmark, with Anthology ID W17-4505 and DOI 10.18653/v1/W17-4505. (https://aclanthology.org/W17-4505/)
- The paper "Image-to-Markup Generation with Coarse-to-Fine Attention" was accepted by ICML 2017. (https://arxiv.org/abs/1609.04938)
- Jeffrey Ling is a coauthor of the paper "Image-to-Markup Generation with Coarse-to-Fine Attention" (arXiv:1609.04938), alongside Yuntian Deng, Anssi Kanervisto, and Alexander M. Rush. (https://arxiv.org/abs/1609.04938)
- Devin wrote: "Jeffrey asked me to redesign his website, and I'm writing this guest post to tell you about it." (https://jeffreyling.github.io/posts/devin-guest-post.html)
- Devin wrote: "Thanks for reading, and thanks to Jeffrey for letting me write on his blog!" (https://jeffreyling.github.io/posts/devin-guest-post.html)

## Experience
- Human Software Engineer at Cognition (Jan 2024 - Present): Building Devin, an AI software engineer
- Co-Founder at Stealth Startup (Jan 2022 - Jan 2023)
- Software Engineer at Waymo (Aug 2019 - Oct 2021): Planner ML. Worked on both R&D and productionization of planning models.
- AI Resident at Google (Jun 2018 - Jun 2019): NLP Research team in NYC. Applications of deep neural nets, Transformers, and BERT for knowledge bases, relation extraction, and visual question answering. Co-authored 3 papers.
- Research Engineer at Vicarious AI (Aug 2017 - Jun 2018): Deep learning and computer vision for robot arms.
- Software Engineering Intern at Google (Jun 2015 - Aug 2015): Intern with the Google Translate team. Applied structured prediction ML techniques for word alignment between different languages.

## Education
- Harvard University - BA Computer Science & Math - (Jan 2013 - Jan 2017)

## Sources
- https://www.linkedin.com/in/jeffrey-ling-92163879
- https://x.com/spdling
- https://jeffreyling.github.io/
- https://jeffreyling.github.io/blog.html
- https://github.com/CognitionAI/devin-swebench-results
- https://www.linkedin.com/posts/rhythm-garg_it-was-awesome-working-with-moritz-stephan-activity-7449944470625701888-xhOt
- https://www.linkedin.com/posts/theodormarcu_sara-xiang-federico-marcos-jeffrey-ling-activity-7353898027507441664-wGZO
- https://jeffreyling.github.io/posts/devin-guest-post.html
- https://jeffreyling.github.io/posts/nips-2017.html
- https://aclanthology.org/W17-4505/
- https://arxiv.org/abs/1906.03158
- https://arxiv.org/abs/2001.03765
- https://arxiv.org/abs/2106.08417
- https://arxiv.org/abs/1609.04938
- https://www.akbc.ws/2020/assets/pdfs/iHXV8UGYyL.pdf
- https://x.com/alexkaplan0/status/2065483246394118443
- https://api.github.com/users/jeffreyling
- https://github.com/jeffreyling
- https://cognition.com/blog/frontier-code-1.1
- https://cognition.com/blog/swe-check-10x-faster
- https://github.com/jeffreyling?tab=repositories
- https://x.com/spdling/status/2113344717098496293
- https://x.com/spdling/status/2112656402741334506
- https://x.com/spdling/status/2110794818884673573
- https://t.co/wCRhbyjMgK
- https://x.com/spdling/status/2108233941199319130
- https://cognition-labs.com/post/swe-bench-technical-report
- https://doi.org/10.48550/arXiv.1906.03158
- https://doi.org/10.48550/arXiv.2001.03765


Researched with Autumn (https://www.autumn.ai) from public sources. Updated 2026-11-04. Profile: https://www.autumn.ai/resources/cognition/people/jeffrey-ling · Everyone at Cognition: https://www.autumn.ai/resources/cognition/people