Cogna

“Fewer than 100 firms in our sample of 8,300—a group that we have dubbed Standouts—accounted for about two-thirds of the positive productivity gains in each of the three country samples we analyzed.” Wouldn’t it be great to build a company that made it on this list!? 🚀 https://lnkd.in/eB7CrnCA

When hiring for high-impact roles, filtering for agency and action is only the start - we also evaluate first-principles thinking versus pattern matching. Ideal candidates default to first-principles reasoning but have enough domain experience to leverage relevant patterns. Most folks with great CVs can't attack problems from first principles. If you can, we're hiring. https://lnkd.in/emGF9DhW

We are super excited to announce and welcome Cogna as sponsor for the #FutureVision #Innovate #Regenerate annual conference 2025 where they will be featuring at the Digital & #OperationalExcellence Networking Live taking place on day 1 and exhibiting across both days. Read the full article to find out more: https://lnkd.in/esGjR8qa and look forward to introducing them to our membership community on November 11th and 12th at Chester Cathedral.

Back in Austin next week on the hunt for exceptional sales talent. If that's you, you know anyone or you're just interested in what we do at Cogna, please get in touch. 🚀 #AustinTech #TexasTech #SiliconHill

Usually I post complete research papers, but this time I wanted to post something different. Over the last few months I talked to a lot of people in the field and I think I have come up with a list of research questions that we as a field need to tackle: ❓ If the AI can produce any code on demand, is it necessary to have modules and dependencies anymore? Why bother with package management, versioning, supply chain management, API incompatibilities and other problems caused by the dependency hell if you can just ask the AI to implement exactly the functionality you need in your codebase? In other words, do the basic principles of modularity and separation of concerns need to be reconsidered? ❓If the AI context is big enough that entire documentations can be sent to the AI with every request, and the AI is able to perfectly recall every interface detail, are abstractions necessary anymore? Do we need to build more abstract APIs that hide complex interactions of more and let the AI do the rest? Is the principle of abstraction still relevant? ❓Can the AI just be given data access directly? Do we need encapsulation? If the AI can just access the entire dataset at once, especially with agents and tools, do we need to protect the internal state of objects or enforce loose coupling? ❓Is “Keep it simple, stupid” still relevant if only the AI is going to read the code? Can we ask the AI to build hard to read but more robust code? What would happen if, for example, we ask the AI to start every single function with an exhaustive list of validity checks for inputs? All of us were taught to do this, and we all generally agree it would be a good idea but nobody does it because it takes far too long to write and produces unreadable, unmaintainable code. ❓What types of tests do we need if the AI can just generate entire categories of tests perfectly every time? Developers write a lot of automated tests that AI can just replace, do we still need to write those? And if not, do we need to write some other tests now? And then there is the following meta-question: ⁉️ Which parts of the codebase will be shared between the human and the LLM, which will be human only and which will be LLM only? The shared parts of the codebase obviously being the most interesting ones because there the two sets of limits interact the most. You can see more in-depth analysis of these questions and how I arrived at them on my blog https://lnkd.in/ep4Mx6fS And If you would like to help me answer some of these questions, my company is hiring, feel free to contact me here on LinkedIn or on [email protected]. #llm #ai #software #softwareengineering #ArtificialIntelligence #GenerativeAI

It's been a real joy to partner with Graham Billsborough and the OCU team to improve operational performance. We’ve achieved some remarkable results - an 83% improvement in processing time across 50,000 job packs is a big deal! Read more about what this involved below ⬇️

Coding is one of AI’s true killer apps. 🔥 Copilots are on fire: → Cursor: $200m ARR → GitHub Copilot: $400m ARR (our estimate) → Windsurf: $40m ARR 🧱 And app builders aren’t far behind, with Bolt and Lovable are both scaling to ~$40m. Why? ✅ Code is deterministic 💸 Devs are expensive 🚀 Tools like Anthropic Sonnet 3.5 can generate thousands of lines of coherent code in one go But what about the messy bits? Think about: 🧑💼 Requirement gathering, which is often fuzzy, revealed only after whiteboards and hard conversations 🧩 Unique environments with integrations, constraints, and legacy stacks varying wildly across companies 🤝 The desire for outcomes, with trust and reliability mattering more than raw output We believe this non-deterministic, unglamorous work is here to stay. In this blog post, together with my brilliant colleagues Owen Brooks and Nils Howland, we dig into the parts of software that still need a human touch — and the startups rising to the challenge. Cc: Cogna Databutton Lovable bolt.new Dust Convergence Article linked in comments below. #AI #B2B #SoftwareDevelopment #Copilots #LegacyTech #LLMs #EnterpriseTech

High-value knowledge work boils down to four steps: 1️⃣ Intuiting the right things to care about 2️⃣ Fusing data across disparate sources 3️⃣ Analysis, optimization & scenario planning 4️⃣ Delivering derived insights in various ways that drive action Today, 99% of folks time is swallowed by steps 2–4. In the next few years, purpose-built AI tools like those we build at Cogna will flip that dynamic and the businesses that attract and champion the high-agency people who master step 1 will leave everyone else in the dust.

A huge thank you to friend of Cogna and angel investor Guy Podjarny for joining us at Cogna HQ and inspiring us with his experiences in building Blaze, Snyk and Tessl 🙏🏼 Guy’s leadership has already reshaped how software is secured… and now he’s championing the AI-driven, spec-first future of development. At Cogna, we’re pioneering an outcome-based services model that fully leverages AI-generated code. At Tessl, Guy and team are building tooling, ecosystem and community to power that shift. We’re both London-based 🇬🇧 both global-minded 🌎 and both on a mission to bring the transformative power of software to billions more users 📈 Guy nailed it when he said: “Cogna is so exciting because it’s building the next generation of SI… going head-to-head with Wipro, Infosys and Tata Consultancy Services who, trapped in a human-labour model, face the classic innovator’s dilemma: they can’t compete without cannibalizing their huge existing businesses.” We’re proud to fill that gap and pioneer a brand-new service company that delivers more affordable, higher-quality experiences for end customers… and to find ways to work with Guy and Tessl in realising that vision! 🚀

Since 2019, productivity has fallen across 13 sectors that make up 63% of the UK economy… meanwhile US software investment grew 42%, more than double the UK’s 17%, helping drive a 9.6% productivity surge and making it the only G7 country to accelerate growth this decade. As a region, we must invest in software. If you’re in a leadership role and unsure where software can most impact your business, Cogna’s world-class team of experts will show you… and you won’t pay a thing unless the benefits are realised.