🤖 🧑💻 Beyond Developers #197 - There is no pause button
The challenge is no longer whether organisations should adapt to AI, but how quickly they can reinvent themselves before the old operating model breaks
Hello friend,
Following Anthropic’s slightly disingenuous claim that we may need to slow down AI development so humanity has time to adapt, I found myself thinking about Alvin Toffler’s Future Shock.
Back in 1970, Toffler argued that the greatest challenge of the future would not be technology itself, but the inability of individuals and institutions to adapt to an accelerating pace of change. Fifty-five years later, that idea feels more relevant than ever.
The stories in this week’s newsletter aren’t really about AI. They’re about adaptation. How do junior engineers learn when AI performs much of the apprenticeship work? How do organisations preserve knowledge when it doesn’t actually live in people’s heads? How do leaders govern systems where execution is becoming effectively abundant?
One thing I’m fairly certain of is that slowing everyone down is not a viable solution. There is no pause button. The technology will keep improving, competitors will keep adopting it, and expectations will keep rising.
The only sustainable response is to strengthen our ability to adapt. In many ways, the most valuable skill of the AI era may not be coding, prompting, or even building models. It may simply be learning faster than the world changes around us.
Now, let’s dive into this week’s signals.
P.S.: I recently launched Rubbr, an AI software delivery platform for complex enterprise applications. If you’re interested in talking about AI engineering in real-life setups, let’s talk!
Recursive organisations
Anthropic’s latest essay on recursive self-improvement sparked plenty of debate about whether AI systems could eventually improve themselves faster than humans can improve them. My takeaway is different. The most important signal isn’t the speculative future scenario, but what is already happening today. Anthropic reports that Claude now writes most of its code, while engineers are shipping dramatically more software than they were only months ago. The same pattern is emerging across the industry: agents are increasingly capable of coding, testing, researching and executing tasks that previously required human effort. As execution capacity explodes, the fundamental challenge shifts from “how do we get things done?” to “how do we decide what should be done?” The organisations that win won’t necessarily be those with the most powerful agents, but those that develop the best mechanisms for prioritisation, governance, coordination and judgment.
I posted on the topic last week and the comments revealed a sharp divide between those focused on the long-term possibility of AI self-improvement and those focused on today’s operational realities. Skeptics argued that Anthropic’s essay is primarily AGI marketing, pointing out that current systems remain heavily dependent on human guidance and still struggle with maintainability, reliability and verification. Others argued that forms of recursive improvement already exist in production through autonomous agent systems that can inspect, modify and optimize parts of their own behavior within predefined boundaries. But regardless of where people stood on the RSI debate itself, a common theme emerged: execution is accelerating far faster than validation. Multiple commenters argued that the real bottleneck is no longer code generation but review, verification, governance and strategic alignment. In other words, even if recursive AI remains uncertain, recursive organisational challenges are already here.
The knowledge extraction myth
Most enterprise AI initiatives assume that organisational knowledge already exists somewhere and simply needs to be extracted from employees. In practice, that’s rarely how knowledge works. Humans are unreliable knowledge repositories: we forget important details, omit information that feels obvious to us, and often describe processes that no longer reflect reality. The more effective approach is to reconstruct knowledge from the traces organisations leave behind: source code, documentation, tickets, architecture decisions, workflows, production systems, operational data and communication patterns. Humans still play a critical role, but not as the primary source of truth. Instead, they become validators, helping confirm or reject hypotheses generated from the organisation’s actual behaviour.
This shift has important implications for enterprise AI. Many companies are investing heavily in knowledge capture initiatives, expert interviews and documentation projects in the hope of preserving institutional knowledge. But the real opportunity may lie elsewhere. The richest source of organisational knowledge is often the organisation itself: the systems it operates, the decisions it records and the artefacts it produces every day. Building AI systems that can continuously reconstruct, validate and update this understanding may prove far more valuable than trying to manually extract knowledge from people’s heads.
Reinventing the engineering apprenticeship
The debate around AI replacing junior developers misses a more important question: how do junior developers become senior developers in an AI-native world? For decades, software engineering relied on an apprenticeship model where newcomers learned through bug fixes, documentation, testing, maintenance work and small features before gradually taking on larger architectural responsibilities. The challenge is that AI agents are becoming increasingly capable of handling exactly this kind of work. Rather than eliminating the need for junior engineers, this forces a rethink of what they should be learning. The future engineer may spend less time writing boilerplate code and more time understanding systems behaviour, architecture, trade-offs, product requirements and problem decomposition. The companies that build the strongest engineering organisations won’t be those that cut junior hiring, but those that redesign the path from beginner to expert.
After I posted on the topic, comments revealed both optimism and concern about this transition. Several people agreed that software engineering is increasingly becoming a product-building discipline, where understanding customers, business problems and system design matters more than raw coding output. Others worried that removing junior engineers from the “dirty hands” work of debugging and maintenance could deprive them of the practical experience needed to build intuition about how software actually behaves in production. The emerging consensus seemed to be that future engineers will need stronger systems thinking much earlier in their careers, but that companies must remain deliberate about creating opportunities for experiential learning, rather than assuming AI alone can replace the apprenticeship process that has historically produced great engineers.
The Changelog - Week of June 1st, 2026
Last week, 7 companies raised $1.62 billion in 5 countries. Europe-based companies attracted less than 1% of total funding vs 56% for Asia-based companies (including Istrael) and 56% for North America-based companies. One of these companies distribute or contribute to an open-source project. On the M&A side, 2 company were acquired.
Funding Rounds
Supabase, from Singapore 🇸🇬, raised $500 million in Series F funding led by GIC. Supabase is an open-source Postgres development platform providing databases, authentication, storage, edge functions, and vector search for more than 9 million developers and AI coding agents. (more)
DriveNets, from Ra’anana 🇮🇱, raised $410 million in Series D funding led by Bessemer Venture Partners and Atreides Management. DriveNets develops networking software that scales AI infrastructure and cloud systems through a software-defined Ethernet fabric for heterogeneous AI clusters. (more)
Generalist AI, from San Mateo 🇺🇸, raised $400 million in Series B funding led by Radical Ventures. Generalist AI builds embodied foundation models that enable robots to perform general physical tasks across industrial, warehouse, and laboratory environments. (more)
Cyera, from New York 🇺🇸, raised $300 million in Series G funding led by Blackstone. Cyera is an AI-powered data security platform that discovers, classifies, and protects sensitive data across cloud environments. (more)
Kosmos, from Chicago 🇺🇸, raised $5 million in Seed funding led by Norwest. Kosmos is an AI-native operational intelligence platform that correlates customer, engineering, and incident data to identify root causes and accelerate issue resolution. (more)
Oplane, from Malmö 🇸🇪, raised $5.2 million in Seed funding led by Seed Capital. Oplane is an AI-native security platform that automates threat modelling by analysing application architecture and delivering remediation guidance directly to developers. (more)
Kodesage, from Budapest 🇭🇺, raised $6.6 million in Seed funding led by VentureFriends. Kodesage provides an on-premise AI platform that modernises legacy enterprise software by extracting business logic, generating documentation, and automating migration and testing. (more)
M&A Rounds
Kumo AI, from Mountain View 🇺🇸, was acquired by Nvidia for $400 million. Kumo AI develops foundation models for structured enterprise data that help businesses generate predictions for churn, fraud, forecasting, and recommendations. (more)
VoidZero, from Palo Alto 🇺🇸, was acquired by Cloudflare. VoidZero is an open-source JavaScript tooling company behind Vite, Vitest, Rolldown, Oxc, and other core developer infrastructure used across the modern web ecosystem. (more)