John Crickett

Recently, our technical fellow Herb Sutter spoke at a meetup with C++ London to discuss C++’s next decade. Herb discussed the exciting times ahead for C++, as C++26 is poised to do what C++11 did over a decade ago: usher in a new era of C++ code style. After his talk, attendees had the opportunity to connect and discuss what they’d learned. This week, he was the keynote speaker at C++ on Sea, where he shared his thoughts on three key things in C++: features that make our code safer and faster just by recompiling as C++26, reflection as the major game-changing feature of C++26 and the C++26 async framework std::execution.

96

Bio-AI is hungry for structure data, how do we feed it? The UK government might have an answer. For proteins, open resources such as the Protein Data Bank (PDB) empowered AlphaFold-like models to achieve near-experimental accuracy. But the landscape looks very different once you leave canonical protein folds: RNA 3-D structures, membrane complexes, and protein-ligand binding poses remain sparsely explored, leaving even the best models starving for examples and underperforming in practice. Here are some possible solutions. 𝘖𝘱𝘦𝘯-𝘥𝘢𝘵𝘢 𝘮𝘰𝘰𝘯𝘴𝘩𝘰𝘵𝘴 Generating orders-of-magnitude more structures and affinities. OpenBind, from a collaboration between the UK government and industry partners is building the largest public repository of protein–small-molecule complexes ever attempted (500k pairs+binding affinities), while the Structural Genomics Consortium plans to make affinity and structural data for thousands of understudied proteins openly downloadable. These projects follow the PDB’s proven playbook: unrestricted access accelerates every lab’s research. 𝘚𝘺𝘯𝘵𝘩𝘦𝘵𝘪𝘤 𝘢𝘯𝘥 𝘴𝘶𝘳𝘳𝘰𝘨𝘢𝘵𝘦 𝘥𝘢𝘵𝘢 When physical structures are scarce, create proxy information. AlphaFold-predicted models, RNA chemical-mapping megasets, and large-scale generative “hallucinations” supply millions of plausible structures and interactions. While not a replacement for experimental truth, these synthetic datasets broaden model generalization and guide where to invest wet-lab effort. 𝘍𝘦𝘥𝘦𝘳𝘢𝘵𝘦𝘥 𝘢𝘯𝘥 𝘢𝘤𝘵𝘪𝘷𝘦 𝘭𝘦𝘢𝘳𝘯𝘪𝘯𝘨 Instead of generating new data, find ways to use existing data more effectively, without exposing IP. One approach is to focus on consortia and federated learning, by training on internal data anonymously. It's not clear to me how many e.g. internally solved structures are not released in databases like the PDB, have heard conflicting accounts. Otherwise, we can try to use our existing data by building narrow, focused protocols like lab-in-the-loop, or AI-guided directed evolution allow for high quality models without needing more than a 96-plate. 𝘊𝘰𝘯𝘤𝘭𝘶𝘴𝘪𝘰𝘯 Data is a dominant rate-limiting step for Bio-AI. If these initiatives maintain momentum, holes in today's data will be filled, and our prediction capabilities for under-resourced categories might improve as dramatically as structure prediction did for single-chain proteins.

170

5 Comments

The Tech Prosperity Deal signed end of last week between the UK and US could be one of the most important moments for technology in a generation. The numbers are huge: Microsoft’s $30bn (£22bn), Google’s £5bn, and Nvidia’s £500m plus an £11bn supercomputer. But beyond the scale of investment, what excites me most is what it signals: the world is placing enormous bets on STEM. As Jensen Huang, CEO of Nvidia, put it, this is a “Goldilocks moment” for the UK, world-class startups and talent, now matched with world-class infrastructure. And critically, this has been done at speed. Freed from the EU’s slower processes, the UK was able to move fast to strike this deal. In today’s world, moving at the fastest pace in history, that ability to act with urgency is just as important as the capital itself.   This said, Infrastructure and investment are only one side of the story. Without the scientists, engineers, and technologists to design, build and run it all, this deal is just capital on paper. At SThree, as the global STEM workforce consultancy, we see this every day. From supporting businesses building data centres in the UK and US, to helping clients across AI, clean tech and quantum, we’re proud to be enabling the teams who make these investments real. For me, this proves the point I see every day: the future of AI, quantum, and clean tech will be decided by people, not just by investment. That’s why we’re focused on building pipelines of expertise across STEM. Because deals like this can only deliver if the skills are there to make it real. The world is moving faster than ever. The UK has acted quickly on this deal, but the real test is whether we can build the skills and workforce to match the pace. How do we make sure the next generation is ready to lead in AI, clean tech and quantum?

126

2 Comments

Who needs self-service? In this article, Conduktor’s VP of Engineering, Matt Searle, talks about his own experiences attempting to balance developer autonomy alongside platform stability. His key insight? Govern Kafka, or risk havoc. Read his latest blog—and learn more about how Conduktor can help you implement self-service with guardrails. https://lnkd.in/ejWTRctu

14

Today we’re introducing Guide's long-awaited integration with Ashby ATS—bringing Aria, Guide’s AI RC agent, to all Ashby customers. The launch of this integration and our partnership with Ashby will enable more TA teams to: ↳ Automate interview scheduling with Aria, Guide’s AI RC agent, including both Copilot and Autopilot modes ↳ Deliver beautiful, branded candidate portals and interview prep to every candidate seamlessly If you’re an Ashby user looking to free up RC bandwidth and improve the candidate experience, the wait is finally over 🔥

202

42 Comments

From concept to silicon 🚀 Arm partner weeteq is proving how the power of Edge AI can be unleashed to quickly turn bold ideas into reality. The UK-based startup, very quickly progressed from concept to near-commercial silicon—earning recognition as a runner-up in the 2023 Arm Silicon Startups Contest.By tapping into Arm Flexible Access, they accessed Cortex-M55, Ethos-U55, and Corstone-300 to rapidly prototype, validate, and tape out silicon without the overhead of traditional licensing. Arm Flexible Access is helping innovators reduce risk, accelerate timelines, and bring new ideas in edge AI to life! 👉 https://okt.to/4pUIoj.

16

🚀 How do you scale access governance without scaling costs? For Brex, manual access requests, reviews, and elevated permissions were slowing down IT and creating unnecessary OpEx. In a highly regulated fintech environment, that wasn’t sustainable. With C1, Brex automated 50,000+ access requests, added 400 new entitlements in just two days using Terraform, and moved to ephemeral, auditable just-in-time access. The result = a big reduction in operating expenses and a faster, more secure way to manage identity. 🔗 Read the full story: https://lnkd.in/evcFwitG

21

1 Comment

First up at LeadDev London’s new iteration #LDX3 we have Gergely Orosz speaking about what the reality looks like of AI coding. Starts off by addressing some of the “executive hype” around AI — it feels like everyone is trying so hard to sell it. Gergely spoke to a wide variety of engineers to understand what is really happening on the ground, from AI dev startups (like Anthropic) through to independent software engineers. Claude Code has seen usage grow from 40% at launch to 160% growth just weeks later. Talked with a bunch of folks at Google — everything is custom there! Even their IDE used to be a web tool, now it’s a VS-Code fork called Cider-V. They also have Critique for AI review of pull requests. “What I’m hearing from my SRE friends at Google is they’re preparing for 10x the lines of code making it into production” Amazon has Amazon Q Developer Pro, super popular internally for AWS. A year ago, it didn’t really work, 6mo ago folks were meh, now they’re using it a lot. They have an internal Claude as well off Anthropic partnership. If there’s an API it’s easy enough to just bolt on an MCP server; so Amazon are doing that left right & centre. Interviewed incident.io (woo!) who are using AI coding to accelerate their development. Shared some specific tips from their internal AI sharing Slack channel. Entire team now regular users of Claude Code even though just released recently. For balance, spoke to a biotech that is finding LLMs not super helpful in their niche. Building novel software. Independent software engineers who LOVE coding and love the craft eg Armin Ronacher who can’t believe how much he loves AI agents “Claude Code got really good — hallucinations avoided by AI running the code”. 🦄 Peter Steinberger feels we’re at a “just works” inflection point now, where a capable engineer can have a lot more output. Birgitta Boeckeler very thoughtful about exploring & understanding what works & doesn’t. Feels LLMs can be used at any abstraction level, it’s a lateral move across the stack. Simon Willison’s blog of over 23 years is now known as “the LLM blog”. Reckons coding agents actually work now, model improvements in the last 6mo have been a tipping point. [side note: Simon & I (Meri) were at uni together and he was already blogging then!] Still bunch of questions left! Not a slam dunk “the future is here”. 1) Why is it founders/CEOs so much more excited than engineers? 2) (missed it sorry) 3) How much time do we save when LLMs actually used? 4) Why does it work so much better for individuals than teams? “LLMs are not just raising the level of abstraction but we are getting used to non-deterministic outputs for the first time” Martin Fowler (rough quote sorry Gergely goes super fast with his slides!!) Kent Beck interestingly also super excited! Real tipping point again. “It’s time to experiment more with LLMs” is the real take away

92

9 Comments

Please kill the bloody TOON posts. While I value innovative open-source projects, of which this is at least the latter, LinkedIn must surely have an algorithm with something more exciting! The last thing we need is yet another data representation; we’ve had XML, ASN.1, JSON, and many more, and now TOON. Meanwhile, most of the world is still using CSV. The claim is efficiency with AI. LLMs are trained on human text and existing formats like JSON and CSVs. LLMs do not work well with large structured data so changing the format isn’t going to help much. IT might save a few fractions of a cent over a week or mean you can fit a few more lines of data but CVS is more efficient anyway. Today’s LLM challenge is tool-calling, the core of agentic AI. Those tools are typically defined in XML or JSON (mostly the latter for open source LLMs), the majority are the tool and parameter descriptions, TOON is no help here. If someone wants to re-train or fine-tune an LLM to work with TOON defined tools then great, it might save us a few percent of tokens. Until then, I’ve had my fill of TOON posts, thank you!

90

54 Comments

Platform Engineering initiatives are shifting from theory to practice but 𝘄𝗵𝗮𝘁 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝘄𝗼𝗿𝗸𝘀? Thanks to Octopus Deploy platform Engineering Pulse Report, Luckily now we have the data to find patterns which can lead us to successful platform engineering practices and luckily enough to be a contributor alongside Steve Fenton, 𝐂𝐡𝐚𝐫𝐥𝐨𝐭𝐭𝐞 𝐅𝐥𝐞𝐦𝐢𝐧𝐠, 𝐒𝐡𝐢𝐩𝐬 𝐌𝐚𝐡𝐢𝐧𝐝𝐫𝐚, 𝐌𝐚𝐭𝐭 𝐀𝐥𝐥𝐟𝐨𝐫𝐝, 𝐉𝐞𝐝 𝐋𝐞𝐡𝐦𝐚𝐧𝐧. 🔎 𝗤𝘂𝗶𝗰𝗸 𝘁𝗮𝗸𝗲𝗮𝘄𝗮𝘆𝘀 - Most orgs adopt platforms to improve efficiency, standardize processes, and increase productivity not just to reduce cognitive load. - Top features that correlate with successful platforms: build pipelines, deployment automation, infrastructure automation, monitoring/observability. - 𝗔𝗱𝗼𝗽𝘁𝗶𝗼𝗻 𝗺𝗮𝘁𝘁𝗲𝗿𝘀: ~60–63% of platforms are mandated. Platform producers report higher success than consumers. - 𝗠𝗲𝗮𝘀𝘂𝗿𝗲𝗺𝗲𝗻𝘁 𝗴𝗮𝗽: ~25% of orgs rely on subjective assessment; others use 1–8 different metrics (DORA, NPS, onboarding time, etc.). - 𝗖𝗼𝗺𝗺𝗼𝗻 𝗯𝗹𝗼𝗰𝗸𝗲𝗿𝘀: poor adoption strategy, lack of feedback loops, and treating platforms as “internal tools” instead of products. From the key finding you'll get to know a pragmatic playbook: how to prioritize the IDP feature set, design an adoption strategy that increases uptake, choose the right success metrics (𝗗𝗢𝗥𝗔, 𝗼𝗻𝗯𝗼𝗮𝗿𝗱𝗶𝗻𝗴 𝘁𝗶𝗺𝗲, 𝗡𝗣𝗦, 𝗠𝗢𝗡𝗞 𝗮𝗻𝗱 𝗼𝘁𝗵𝗲𝗿𝘀), and navigate the common “J-curve” of platform maturity. Concrete examples, a staged maturity model you can use immediately, and measurable checkpoints to prove platform value to engineering and business stakeholders. This report is aimed at 𝗽𝗹𝗮𝘁𝗳𝗼𝗿𝗺 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝘀, 𝗽𝗹𝗮𝘁𝗳𝗼𝗿𝗺 𝗽𝗿𝗼𝗱𝘂𝗰𝘁 𝗺𝗮𝗻𝗮𝗴𝗲𝗿𝘀, 𝗦𝗥𝗘𝘀, 𝗮𝗻𝗱 𝗰𝗹𝗼𝘂𝗱-𝗻𝗮𝘁𝗶𝘃𝗲 𝘁𝗼𝗼𝗹 𝗮𝘂𝘁𝗵𝗼𝗿𝘀 who want an evidence-based approach to building platforms that actually increase developer productivity and business outcomes. 𝗗𝗼𝘄𝗻𝗹𝗼𝗮𝗱 𝘁𝗵𝗲 𝗿𝗲𝗽𝗼𝗿𝘁 𝗻𝗼𝘄: https://lnkd.in/dAhMGPrm #OctopusDeploy #PlatformEngineering #DevOps #OpenSource #cloudnative #AgenticAI #PlatformEngineeringresearch #InternalDeveloperPlatform #DevEx #IDP

22

1 Comment

🦄 𝟲 𝗨𝗞 𝗔𝗜 𝗦𝘁𝗮𝗿𝘁𝘂𝗽𝘀 𝗼𝗻 𝘁𝗵𝗲 𝗙𝗮𝘀𝘁 𝗧𝗿𝗮𝗰𝗸 𝘁𝗼 𝗨𝗻𝗶𝗰𝗼𝗿𝗻 𝗦𝘁𝗮𝘁𝘂𝘀 The UK is an emerging global hub for AI, here are a couple companies to keep on your radar: Fetch.ai (Cambridge) 📈 Funding: $112M | Founded: 2017 An open marketplace for AI Agents. Fetch.ai empowers developers to build multi-agent systems that can securely access datasets and operate autonomously. Founders: Humayun Sheikh, Toby Simpson, Thomas Hain Nscale (London) 🖥️ Funding: $185M | Founded: 2023 A “hyperscaler” purpose-built for AI and HPC. Nscale is redefining infrastructure with a bold focus on sustainable, high-performance computing for the AI era. Founder: Josh Payne Sylvera (London) 🌎 Funding: $97M | Founded: 2020 Using AI to assess and accelerate climate investments. Sylvera provides high-integrity carbon data and tools to drive better net-zero decisions. Founders: Allister Furey, Samuel Gill QuantoPay (London) 💷 Funding: $168M | Founded: 2016 Blockchain meets banking. QuantoPay is a next-gen fintech platform focused on transparency, security, and global transactions without hidden fees. Founder: Dr. Alexander B. bit.bio (Cambridge) 🔬 Funding: $175M | Founded: 2016 Revolutionising biotech with deterministic cell programming. Bit.bio is using AI to engineer human cells for therapies, research, and regenerative medicine. Founders: Mark Kotter, Florian Schuster UrbanChain (Manchester) ⚡ Funding: $75M | Founded: 2017 Challenging the traditional energy monopoly, UrbanChain offers peer-to-peer green energy trading, making renewable power both affordable and accessible. Founder: Somayeh Taheri #UKTech #AIStartups #AINews #ArtificialIntelligence #Startups

10

1 Comment

More interesting news for the UK AI scene!   Microsoft and NVIDIA announced they are teaming up to launch the Agentic Launchpad - a new hub to back UK startups building the next generation of autonomous and “agentic” AI systems. It comes at a time when investment in dedicated UK AI companies hit £2.9 billion in 2024, proof that the ecosystem is scaling fast.   This means big opportunities for engineers and data talent:   - Generative + agentic AI skills will be hot property - Experience deploying on Azure or NVIDIA stacks will be a major plus - Companies will be hiring fast to keep pace with the innovation The UK isn’t just using AI anymore, but creating the opportunities that come with it! #AI #DataEngineering #Startups #GenerativeAI

26

1 Comment