- Joined
- Sep 2025
- How they joined
- Directly
- Before Cognition
- BriqsCo-Founder
- Status
- Current
- LinkedIn post (verbatim): "Working on FrontierCode has been my daily bread and butter for almost a year, covering everything from creating some of the biggest tasks in the set, hiring, guiding, and onboarding Open Source maintainers, to improving the tooling and end-to-end authoring workflow."Jun 9, 2026Working on FrontierCode has been my daily bread and butter for almost a year, covering everything from creating some of the biggest tasks in the set, hiring, guiding, and onboarding Open Source maintainers, to improving the tooling and end-to-end authoring workflow. One unique and time-consuming aspect was working with the team on increasingly open-ended, complex tasks. To avoid false negatives for correctness, we sometimes had to push the limits of our already sophisticated rubric tools, which made room for truly real-world scenarios in our dataset. I'm also grateful to have my own team of leading experts from the Open Source community. We often had deep discussions on correctness versus quality and what mergeability means in the context of this repository and task, taking into account the philosophy of the maintainers of the project. FrontierCode is a milestone for AI models respecting subjective quality in the real world.
- Same LinkedIn post (verbatim): "I'm also grateful to have my own team of leading experts from the Open Source community. We often had deep discussions on correctness versus quality and what mergeability means in the context of this repository and task, taking into account the philosophy of the maintainers of the project.".Jun 9, 2026Working on FrontierCode has been my daily bread and butter for almost a year, covering everything from creating some of the biggest tasks in the set, hiring, guiding, and onboarding Open Source maintainers, to improving the tooling and end-to-end authoring workflow. One unique and time-consuming aspect was working with the team on increasingly open-ended, complex tasks. To avoid false negatives for correctness, we sometimes had to push the limits of our already sophisticated rubric tools, which made room for truly real-world scenarios in our dataset. I'm also grateful to have my own team of leading experts from the Open Source community. We often had deep discussions on correctness versus quality and what mergeability means in the context of this repository and task, taking into account the philosophy of the maintainers of the project. FrontierCode is a milestone for AI models respecting subjective quality in the real world.
- X thread (verbatim): "FrontierCode finally dropped, a coding agents benchmark for the real world. Human-verified through an extensive hardening process, with a new continuous scoring model. Working on it has been my daily bread and butter for almost a year."Jun 8, 2026FrontierCode finally dropped, a coding agents benchmark for the real world. Human-verified through an extensive hardening process, with a new continuous scoring model. Working on it has been my daily bread and butter for almost a year. Here's how this benchmark is unlike others 👇
- Cognition's FrontierCode launch post quotes him verbatim as 'Merlijn Vos, Core Maintainer of uppy (30.8k stars)': "FrontierCode is a milestone for AI models respecting subjective quality in the real world." He was recruited for evals in his capacity as an open-source maintainer.Jun 2026
- FrontierCode 2.0 acknowledgments list him among pod leads (marked *) alongside Claudio Costa (Mattermost), Martin McKeaveney, Tomer Nosrati (Celery) and Cristiam Da Silva.Oct 2026
- Cognition's SWE-1.6 preview gives 'extra credit' to contractor-team members with outsized contributions to data & tooling: Claudio Costa, Martin McKeaveney, Lance Fuchia, Tomer Nosrati, Merlijn Vos.2026
- X (verbatim): "FrontierCode 2.0 is a massive leap in benchmark quality over 1.1, reflecting real-world usage a lot more. Let's see if the vibes confirm it. We desperately need some AI relevance in Europe." (on Mistral's FrontierCode 2.0 result).Oct 30, 2026For a long time I thought it's over for @MistralAI, but they're so back. FrontierCode 2.0 is a massive leap in benchmark quality over 1.1, reflecting real-world usage a lot more. Let's see if the vibes confirm it. We desperately need some AI relevance in Europe. https://t.co/aq4Ygeuzrt
- Public headline says "User-centric development, AI evaluations, and leadership".
- Spoke at TechFems Barcelona (hosted by Happy Scribe) on 'Open Source Evolution: from Past to Contribution'; the speaker bio calls him an award-winning Senior Full-Stack Engineer who helps maintain over 400 packages downloaded hundreds of millions of times.Apr 15, 2024
- • Applied AI Engineering. Deep expertise in AI (coding) evaluations and agentic engineering.
• Knowledgeable across the stack. Familiar and comfortable in building full-stack web apps in all forms
and sizes, cloud infrastructure, and developer tooling.
• Proactive without boundaries. I’m a good engineer with great habits. I have a passion that drives me to help out and elevate others, and the product, beyond the minimum requirements. I can take ownership from the earliest stages of talking to; this comes from the profile summary and preserves the person's own positioning.
- Previous roles include Applied Engineer, Evaluations at Cognition; Full Stack Developer at Transloadit; Co-Founder at Briqs; Frontend Developer at Adyen; Frontend Developer at CLEVER°FRANKE; Frontend Developer at De Voorhoede.
- Education history includes Bachelor of Science - BS, Communication & Multimedia Design, Amsterdam University of Applied Sciences; Minor, Web development, Amsterdam University of Applied Sciences.
Leading a team within the bigger evaluations initiative at Cognition. Covering everything from creating some of the biggest tasks in the set, hiring, guiding, and onboarding people, to improving the tooling and end-to-end authoring workflow.
Working on Open Source Software on the GitHub platform (not employed by GitHub): unified core team member • https://unifiedjs.com The core team is a collective team responsible for governing all other teams (unified, remark, rehype, retext, redot, mdx, micromark, syntax-tree, vfile) tus core team member • https://tus.io Open protocol for resumable uploads. uppy core team member • https://uppy.io The next open source file uploader for web browsers
Working on evaluations and tooling to improve Devin
• Maintained a globally distributed file progressing API (frontend/backend). • Defined and actively implemented long-term visions for open source libraries. • Dealt and prioritized with a constant flow of issues across multiple projects, our community forum, and through customer support. • Campaigned for quarterly goals and guided the founders in the process to help us pick work that aligns with business interest. • Spearheaded the effort to collect all concerns of the Uppy documentation website, rewrote most docs based on that, did the visual design for the website, and the development work • Spearheaded the creation of tus Node.js, a server for resumable uploads. • Delivered succinct, engaging writing in blog posts, job posts, and docs.
Building Dime (later renamed to Briqs), an embedded lending service for platforms and marketplaces. Leading the infrastructure, front-end, and backend of our integrations and AI underwriting engine.
• Building the terminal fleet manager, the software that manages all the payment enabled devices in the field for all customers. • Contributed to the design system components (Vue). • Part of the hiring process (interviews, onboarding).
• Pushed the boundaries of data visualization on the web with D3.js. • Worked on an accessibility critical government pandemic website. • For the Opus project I eventually transitioned into a temporary Tech Lead role. I focussed on onboarding new developers, steering the technical direction, and eventually the handover of the project to a new internal Warner Music Group team • Part of the hiring process (interviews, onboarding)
Worked directly with The Things Network, from requirements gathering, to UX research/design, and eventually delivering a storytelling frontend experience.
At its heart, the degree programme is about gaining mastery of the design process: from understanding the design problem via developing and building the product all the way through to the official launch – and then working on a second version that is a little better still.
A half year minor on web development with a focus on progressive enhancement, accessibility, CSS, semantic HTML, and JavaScript. Projects were always provided by companies with real world use cases.
Working on FrontierCode has been my daily bread and butter for almost a year, covering everything from creating some of the biggest tasks in the set, hiring, guiding, and onboarding Open Source maintainers, to improving the tooling and end-to-end authoring workflow. One unique and time-consuming aspect was working with the team on increasingly open-ended, complex tasks. To avoid false negatives for correctness, we sometimes had to push the limits of our already sophisticated rubric tools, which made room for truly real-world scenarios in our dataset. I'm also grateful to have my own team of leading experts from the Open Source community. We often had deep discussions on correctness versus quality and what mergeability means in the context of this repository and task, taking into account the philosophy of the maintainers of the project. FrontierCode is a milestone for AI models respecting subjective quality in the real world.
For a long time I thought it's over for @MistralAI, but they're so back. FrontierCode 2.0 is a massive leap in benchmark quality over 1.1, reflecting real-world usage a lot more. Let's see if the vibes confirm it. We desperately need some AI relevance in Europe. https://t.co/aq4Ygeuzrt
@MichaelArnaldi Maybe you got (wrongly) downgraded to Opus if you got flagged for doing AI research or security stuff 🤔 For FrontierCode, which tries to capture real world mergeability as much as possible, the jump was meaningful. https://t.co/dbHfuwdjf9
FrontierCode measures *mergeability* by evaluating code along the following axes: behavioral correctness, regression safety, mechanical cleanliness (build, lint, and style checks), test correctness, scope, and code quality.
I’m grateful to have my own team of leading experts in the Open Source community. We had deep discussions on correctness versus quality and what mergeability means in the context of their repository. FrontierCode is a milestone for AI models respecting subjective quality in the real world.
FrontierCode finally dropped, a coding agents benchmark for the real world. Human-verified through an extensive hardening process, with a new continuous scoring model. Working on it has been my daily bread and butter for almost a year. Here's how this benchmark is unlike others 👇
Truly the best of all worlds - Keep your heavily subsidized OpenAI tokens - Run ALL sessions in the cloud for a flat $20/month. No unpredictable compute costs. - Use Devin Fusion (e.g SWE-2 + Astra) for the same intelligence while burning 40% less of your ChatGPT sub - Devin Review on your PRs + Security scans
@jaredpalmer @cognition @asha_shar @ScottWu46 Very cool, congrats!
If Zed is as good as Cursor at agents I would much prefer it over a VS Code experience. But hard to imagine it can be as good seeing the difference in funding and AI-focussed hiring at Cursor/Windsurf 🤔 https://t.co/h3Jjimtep7
