Anupam Kundu

Love this amazing large-scale research on Agents, authored by many of the giants of the industry in which they propose a modular, brain-inspired framework for developing more holistic Agents. This whopping 264 page research paper "Advances and Challenges in Foundation Agents" is a collaborative work authored by researchers from a who’s who of leading global institutions across academia, industry, and research labs. Some quick key points to whet your appetite: 𝟏. 𝐌𝐨𝐝𝐮𝐥𝐚𝐫 𝐀𝐠𝐞𝐧𝐭 𝐀𝐫𝐜𝐡𝐢𝐭𝐞𝐜𝐭𝐮𝐫𝐞 Proposes brain-inspired modules for Agents: cognition, memory, world modeling, emotion, perception, and action. Each module maps to human brain functionality, acknowledging areas that are well-developed, moderately explored, or under-researched. 𝟐. 𝐒𝐞𝐥𝐟-𝐄𝐧𝐡𝐚𝐧𝐜𝐞𝐦𝐞𝐧𝐭 𝐚𝐧𝐝 𝐀𝐝𝐚𝐩𝐭𝐚𝐭𝐢𝐨𝐧 Explores how Agents can evolve over time via mechanisms such as: - Prompt optimisation - Workflow and tool learning - Self-improvement loops (online/offline) - Use of LLMs as optimisers - Discusses scientific discovery as a use case for Agent evolution and intelligence growth. 𝟑. 𝐂𝐨𝐥𝐥𝐚𝐛𝐨𝐫𝐚𝐭𝐢𝐯𝐞 𝐚𝐧𝐝 𝐌𝐮𝐥𝐭𝐢-𝐀𝐠𝐞𝐧𝐭 𝐒𝐲𝐬𝐭𝐞𝐦𝐬 Details how Agents can collaborate, communicate, and form societies, including a topology of interaction, protocol design, and Human-Agent collaboration. The authors emphasise the emergence of collective intelligence and adaptability. 𝟒. 𝐒𝐚𝐟𝐞𝐭𝐲 𝐚𝐧𝐝 𝐀𝐥𝐢𝐠𝐧𝐦𝐞𝐧𝐭 A critical discussion of intrinsic and extrinsic safety challenges, such as: Jailbreaks, hallucinations, misalignment, adversarial attacks ,and Agent-to-Agent, and Agent-to-Environment interaction risks. +++++++++++++ I love the general gist of this and their conclusions are EXACTLY what I have been saying here in my many posts, namely, the path forward involves multidisciplinary collaboration, especially combining insights from AI, neuroscience, cognitive science, and ethics. The focus should be on human-centered AI that augments rather than replaces human intelligence, promoting a hybrid society of Humans and Agents. Few seem to recognise - or acknowledge - we are in the earliest nascent days of Agents with much to be figured out, and we really needed something like this done on a global scale. No one institution will be able to figure this all out in isolation. It's simply too big and complex.

63

9 commentaires

RIP Prof Boden “Computer science helps us to understand what a generative system is, how it’s possible to have a set of rules — which may be a very, very short, briefly statable set of rules — but which has the potential to generate infinitely many different structures.” Sound familiar? https://lnkd.in/gPAmwkm3

10

We build systems to make things easier. But too often, what we call “automation” ends up feeling like digital red tape, frustrating, rigid, and impossible to reason with. In this piece, I reflect on the difference between good and bad automation, why metrics matter more than ever, and how AI can quietly make the worst patterns harder to detect and fix. Read it here: https://lnkd.in/gR2kiUsw

21

💡 Fintech promised disruption — but the system pulled them back in. For a decade, start-ups claimed they would reinvent banking. Agile, digital-first, customer-obsessed — they vowed to build what incumbents could not. Yet most of that “innovation” stayed skin-deep. The rails, the regulators, the clearinghouses — all the same. To scale, every challenger eventually had to plug back into the very system it wanted to replace. The result? A revolution that became a renovation. The dream of frictionless finance collided with the gravity of compliance, licensing, and legacy code. This is the Mirage of Innovation in Banking — the first full chapter in The Future of Finance series, following the Prologue published last week. In this piece, I explore why change in finance always stalls just before the core, and why the next wave of disruption will finally need to rebuild the system itself. #FutureOfFinance #Banking #Fintech #DigitalBanking #Payments #Innovation

7

"Design is not just a service to business. It is a way of constructing reality – a lens, a language, a cultural force. To reduce it to business utility is not just a missed opportunity – it’s a profound misunderstanding. So here we are: designers fluent in business and executives blind to design. The asymmetry is now complete. And still, the profession keeps asking itself how to be more useful, more strategic, more like them.” — Natasha Jen, Pentagram

33

8 commentaires

Very nice paper by my former Apple colleagues - The Illusion of Thinking (https://lnkd.in/gkjJi2G2). The heart of the paper lies in distinguishing “reasoning” from “pattern matching” in large AI models, and demonstrating why puzzle-based evaluations are crucial for stress-testing these capabilities. Pattern matching refers to a model’s ability to recognize and reproduce statistical regularities from its training data. Reasoning, in contrast, involves abstract multi-step problem-solving, logical inference, and the ability to generalize solutions to novel problems. The paper demonstrates that LRMs' apparent reasoning is an "illusion”—similar to LLMs, they break down at higher complexity and are unable to truly reason (i.e., break down a complex task, plan, adapt, and apply logic beyond memorized patterns). This has sparked a huge debate in the AI community, with folks taking sides. My view is that the paper is an exceptional contribution precisely for this reason — it sparks the right conversation and challenges thinking on both sides. In research, often the biggest contribution is posing an important question with clarity. It’s a very accessible read, and I expect there will be lots of follow-up work stress-testing these findings in various ways (with different models, tasks beyond puzzles, novel problems not seen in training data, evaluation methods, etc.). Would love to hear about others’ experiences in practical settings.

64

8 commentaires

We publish Arjo Basu and Debleena Majumdar’s first column in Founding Fuel today. It’s a thoughtful piece on something many leaders are grappling with right now — what happens when stories move faster than the systems meant to support them, and how AI accelerates that gap. Anchored in recent events in Indian aviation, and contrasted with the long arc of UPI, the essay makes a quiet but important point: leadership today is less about telling better stories, and more about ensuring that systems can actually carry the narratives we put into the world. ⏱️ Read Time: 3 minutes 🔗 Link in comments. Do read — and please share if it sparks reflection. #Leadership #AI #Organisations #Narrative #FoundingFuel

19

2 commentaires

In this special Humans Being Human episode of KnolShare with Dr. Dave, we explore the democratization of coaching through the lens of cultural and collective identity. Hasan Rafiq, PCC, shares how coaching can honor diverse traditions, histories, and lived experiences while expanding access across boundaries of class, culture, and geography. We examine how coaching empowers belonging, challenges displacement, and creates space for all voices to be seen, heard, and valued in the human experience, from migration to technology. This episode explores strategies to responsibly scale coaching while maintaining its human-centered essence, from workplace inclusion to community resilience. Discover how coaching can become a tool for collective upliftment and systemic change.

8