🧑💻 Developer-First #192 - The new role of humans in software engineering
From writing code to orchestrating systems, the centre of gravity is shifting
Hello friend,
There is a growing disconnect in how different parts of the industry see the role of humans in software engineering. On one side, developers push back, rightly, on narratives that reduce engineering to prompt-and-pray. Execution still matters. Systems still break. Someone still has to understand what is happening under the hood. On the other side, most CEOs are looking at the same tools and seeing something else entirely: a path to reduce dependency on scarce, expensive engineering talent. Not eliminate engineers, but fundamentally change where and how they are needed.
Whether we like it or not, generative AI is the biggest shift software engineering has gone through in decades. And organisations will adapt to it. The centre of gravity is moving from execution to orchestration: defining problems, setting constraints, supervising agents, and validating outcomes. Some engineers will resist this shift. Some will thrive in it. But the direction is clear. The question is no longer whether humans stay in the loop, but where they create the most value in a system where machines increasingly handle the doing.
Now, let’s dive into this week’s signals.
P.S.: I recently launched Rubbr, a controlled AI-driven software delivery platform for legacy systems. If you’re interested in talking about AI engineering in real-life setups, let’s talk!
When agents act faster than systems can think
An AI agent deleted a production database in seconds. Not because it went rogue, but because it did exactly what it was designed to do: solve a problem as fast as possible, with the context it had. It didn’t know what was production versus staging, didn’t understand the blast radius, and had no concept of what “safe” meant in that environment. So it made a technically valid decision and an operationally catastrophic one. As I argued on LinkedIn, this is not an AI failure. It is a system design failure. We are giving agents access to systems that are poorly understood even by humans, and relying on scattered documentation and conventions to keep things under control.
The comments to my post reinforced how misunderstood this failure mode still is. Many argued it was a human mistake: bad permissions, poor hygiene, missing approvals. That’s true, but adding more human guardrails is not a solution. It just creates bottlenecks and slows everything down. We end up reintroducing the very constraints AI was supposed to remove. The real shift is that agents amplify whatever system they operate in. If the system is fragile, they break it faster. If it is well-structured, they scale it safely. The answer is not more humans in the loop, but systems with explicit constraints and boundaries that cannot be bypassed.
From coders to owners
Andrew Ng is right: AI-native teams don’t just move faster, they operate differently. Coding agents are shifting engineers up the stack, from execution to ownership. The role is expanding beyond writing code into understanding problems, shaping solutions, and validating outcomes. This is not entirely new, but AI is accelerating a model that already existed: the product engineer. As execution becomes cheaper, the value moves upstream. Engineers who can think in terms of product, systems, and trade-offs become disproportionately more valuable.
I shared these thoughts on LinkedIn last week, and the comments on my post reinforced that this shift is broader than engineering alone.It’s not just engineers expanding into product and design, it’s entire teams converging. Product managers, designers, and even marketers are overlapping more with engineering, because AI compresses execution across all functions. The common pattern is ownership: those who ask the right questions, challenge assumptions, and validate outcomes outperform those who simply “build.” The role that disappears is not the engineer, but the one optimised for execution without context. AI is not replacing developers, it is exposing how narrow that role had become.
Coding is shrinking. Engineering is expanding
I think both Kent Beck and Dario Amodei are right, but they’re talking about different layers of the stack. Coding is becoming less important in relative terms as models take over more of the execution. On that, Amodei is directionally right. But equating coding with software engineering is the real mistake. As Kent points out, the discipline has always been about something deeper: understanding what to build, how systems behave, what can break, and how changes propagate. AI is not removing that. It is exposing it.
The comments on my post made this tension very clear. Many pushed back hard, arguing that code remains fundamental, that execution still matters, and that this is just hype driven by AI vendors. They’re right to challenge the narrative, but they’re missing the shift. Code is not disappearing, just like assembly didn’t disappear. It is being abstracted away. What matters is not who writes it, but who understands the system well enough to guide it. As several practitioners pointed out, the real bottlenecks are now verification, context, and decision-making. The engineers who thrive will not be the ones writing the most code, but the ones who can reason about systems they did not fully write, detect when things go wrong, and decide what should happen next.
The Changelog - Week of April 27th, 2026
Last week, 7 companies raised $1.26 billion in 3 countries. Europe-based companies attracted 88% of total funding vs 12% for North America-based companies. One of these companies distribute or contribute to an open-source project. On the M&A side, 2 companies were acquired.
Funding Rounds
Ineffable Intelligence, from London 🇬🇧, raised $1.1 billion in Seed funding led by Sequoia Capital and Lightspeed Venture Partners. Ineffable Intelligence is an AI research lab building “superlearner” systems that acquire knowledge through reinforcement learning without relying on human-generated data. (more)
Parallel Web Systems, from Palo Alto 🇺🇸, raised $100 million in Series B funding led by Sequoia Capital. Parallel Web Systems provides web search and research API infrastructure for AI agents, offering structured access to the open web through a proprietary index. (more)
Featherless, from San Francisco 🇺🇸, raised $20 million in Series A funding led by AMD Ventures and Airbus Ventures. Featherless is a serverless platform that lets developers deploy and run open-source AI models at scale without managing infrastructure. (more)
Definity, from Chicago 🇺🇸, raised $12 million in Series A funding led by GreatPoint Ventures. Definity builds an agentic data engineering platform that monitors and autonomously manages data pipelines in real time. (more)
General Analysis, from San Francisco 🇺🇸, raised $10 million in Seed funding led by Altos Ventures. General Analysis develops AI security infrastructure to test agents with adversarial scenarios and identify failure modes before deployment. (more)
OpenObserve, from San Francisco 🇺🇸, raised $10 million in Series A funding led by Nexus Venture Partners and Dell Technologies Capital. OpenObserve is an open-source observability platform unifying logs, metrics, and traces to monitor systems and automate incident response. (more)
Redpine, from Stockholm 🇸🇪, raised $8 million in Seed funding led by NordicNinja. Redpine provides an API platform enabling AI agents to access and pay for premium licensed datasets in real time. (more)
M&A Rounds
Eigen AI, from Palo Alto 🇺🇸, was acquired by Nebius for $643 million. Eigen AI optimizes AI model performance on chips to improve inference efficiency and reduce compute costs. (more)
Assured Robot Intelligence, from New York 🇺🇸, was acquired by Meta. Assured Robot Intelligence develops AI models that help robots understand and adapt to human behaviour in real-world environments. (more)