- Joined
- Jun 2025
- How they joined
- Directly
- Before Cognition
- Autonurture AICTO & Co-Founder
- Status
- Current
- Posted that FrontierCode was used in Anthropic's Claude Fable 5 release: "The benchmark used to test it's capability was built by our team at Cognition. We've been working on it for over a year - creating and designing the spec and infrastructure to run it from scratch." He also wrote: "Building FrontierCode alongside this team and seeing it used and trusted by Anthropic has been some of the proudest work so far."Jun 12, 2026Anthropic released Claude Fable 5 this week. The most capable model they've ever made publicly available. The benchmark used to test it's capability was built by our team at Cognition. We've been working on it for over a year - creating and designing the spec and infrastructure
- Joined Cognition in June 2025 in an Engineering role (LinkedIn); his GitHub bio frames it as 'OSS Engineer @CognitionAI'.Jun 2025
- Cognition's SWE-1.6 preview post singles him out among 'members of our contractor team who have made outsized contributions to data & tooling' (with Claudio Costa, Lance Fuchia, Tomer Nosrati, Merlijn Vos).
- Featured maintainer on Cognition's FrontierCode benchmark launch (credited as 'Co-Founder and CTO of Budibase (28k stars)'), quoted: “What sets FrontierCode apart is the attention to detail. Each task is calibrated to a depth that simply hasn’t been seen before in LLM benchmarking. We should be moving away from benchmarks that can be gamed and instead using ones like FrontierCode to demonstrate genuine model intelligence and creativity.”.
- Listed as a pod lead (asterisked) in the FrontierCode 2.0 acknowledgments alongside Claudio Costa, Tomer Nosrati, Merlijn Vos and Cristiam Da Silva.
- Co-founded Budibase (open-source low-code/internal tools platform, Belfast, founded 2019) with Joe Johnston and Mike Shanks; he served as CTO Jan 2020 to Sep 2025.
- A newsletter profile describes Budibase as '(acquired)' and lists him as Founder @ Soria Labs with 1.5 years at Cognition.Sep 7, 2026
- Founded Soria Labs in January 2026 and is CTO & co-founder of Autonurture AI (2025); his posts pitch workflow-first AI adoption for recruitment and legal firms.Aug 17, 2026I've been programming since a young age, and I’m conscious of keeping up to date with the latest AI updates and models, which is why I find it particularly interesting the way that talk of the frontier is flooding everyones LinkedIn with noise right now. Every week there's a new
- Same post: "It's humbling and frankly quite insane to see your work referenced in a frontier model release." and described working with 20+ open-source maintainers across 36 repositories, with Diamond-tier scores around 13% for Opus 4.8 and around 30% for Fable.Jun 12, 2026Anthropic released Claude Fable 5 this week. The most capable model they've ever made publicly available. The benchmark used to test it's capability was built by our team at Cognition. We've been working on it for over a year - creating and designing the spec and infrastructure
- Described his current mix of work as: "I see this from all angles; building enterprise software for governments and banks, doing AI research, and running AI agents day to day for recruitment and legal firms."Aug 17, 2026I've been programming since a young age, and I’m conscious of keeping up to date with the latest AI updates and models, which is why I find it particularly interesting the way that talk of the frontier is flooding everyones LinkedIn with noise right now. Every week there's a new
- Public headline says "I help enterprises save 1000s of hours lost to admin, by mapping workflows before introducing AI | 3000 hours saved and counting | Founder @ Soria Labs | Founder @ Autonurture AI | Frontier AI research @ Cognition".
- Press coverage of Cognition's SWE-2 model quoted him (as Budibase CTO) on FrontierCode task depth.Sep 11, 2026
Shogun Digital is a digital products agency that specialises in turning ideas and concepts from entrepreneurs into fully functional, demoable prototypes and MVPs and iterating on them over time to become fully fledged, high quality software products. We work hard to understand your business and work very closely with founders through every step of the process.
Rapid7 is a leading provider of security data and analytics solutions that enable organizations to implement an active, analytics-driven approach to cyber security. We combine our extensive experience in security data and analytics and deep insight into attacker behaviors and techniques to make sense of the wealth of data available to organizations about their IT environments and users. Our solutions empower organizations to prevent attacks by providing visibility into vulnerabilities and to rapidly detect compromises, respond to breaches, and correct the underlying causes of attacks.
External Qualifications/Awards gained here: - Certified Microsoft Technology Associate (Security Fundamentals), - Microsoft Certified Professional
I've been programming since a young age, and I’m conscious of keeping up to date with the latest AI updates and models, which is why I find it particularly interesting the way that talk of the frontier is flooding everyones LinkedIn with noise right now. Every week there's a new
Anthropic released Claude Fable 5 this week. The most capable model they've ever made publicly available. The benchmark used to test it's capability was built by our team at Cognition. We've been working on it for over a year - creating and designing the spec and infrastructure
Opus 4.8 dropped yesterday - but the most useful change isn't obvious from the benchmarks. 4.8 is more willing to say "I don't know." LLMs are engineered to give you an answer at all costs. They generally tend to surface information with a similar apparent sense of certainty, r
Anthropic pulled Claude Code from cheaper plans, saw enormous backlash, and reversed it. It's a sign of what's coming. Like many, I use frontier models every day, but the reality is that the economics of these models are propped up by an enormous amount of VC money. Anthropics
Anthropic built a model that found a 27 year old security flaw that millions of attempts missed, then they decided not to release it. Claude Mythos Preview is apparently better than almost any human security researcher at finding software vulnerabilities. On top of the 27 year o
Everyone wants to know which model is “better.” So I tested it properly. Same agent, same prompts, four admin jobs: a complex engineering task, turning a messy CV into a template, following a compliance script on a call, and building a UI with a design system. I changed one line
Is AI going to kill us all? That’s the debate filling the headlines, but it’s not what you should be worrying about as a business. Nobody wants to die at the hands of AI. From a business point of view, though, the risk to watch is what AI is doing in your workflows right now, be
We saved an energy company in Portugal $90,000 a year. They were manually calling every inbound lead. That’s common at admin-heavy firms, where time goes on tasks that could be automated. We mapped the workflow, sorted each step by whether it needed automation, AI or a human, a
“AI is getting smarter while humans get dumber” is all over my feed right now. I think half of it is wrong, and that is the half that’s going viral. If a doctor thinks you have a heart condition when you actually have diabetes, the treatment isn’t going to make you better. My po
You’re probably wasting money on mindless admin as you read this. If you’re an ops or finance lead at an admin-heavy firm, this video is for you. An energy company I worked with had its team chasing late payers one call at a time, reading the compliance script word for word, whi
Ignore this entire post if you don’t care about your company’s privacy with AI… Your team is already pasting customer data into whatever LLM you use. You just can’t see it. That’s my starting assumption with every firm I work with. There are a couple of ways that companies usua
We saved a major energy company over $90,000 by mapping their workflows. Before you add AI to your admin work, find out which parts actually need it. That’s where we started. This energy company, here in Portugal, was manually calling every inbound lead. We broke that workflow i
