February has been a busy month for me at InfoQ. I wrote three articles that, on the surface, cover different topics: skill formation, open-source sustainability, and Agile methodology. But when I stepped back and looked at them together, a pattern jumped out at me. Each one tells a piece of the same story: AI is transforming how we build software at a pace that exceeds our ability to think about the consequences.

I want to use this post to connect the dots.

The Skill Problem Nobody Wants to Talk About

The first piece I wrote covered an Anthropic study on how AI coding assistance affects skill development. The research was a randomized controlled trial with 52 junior engineers learning a Python library called Trio, which none of them had used before. The findings were stark. Developers who used AI assistance scored 17 percent lower on comprehension tests compared to those who coded by hand. That gap is roughly equivalent to two letter grades.

What struck me most wasn’t the headline number, though. It was the nuance underneath. Participants who used AI as a thinking partner, asking conceptual questions, requesting explanations, and working through problems alongside the tool, retained far more knowledge than those who asked the AI to generate code for them. The dividing line sat around a 65 percent score threshold. Above it, you found the curious developers. Below it are the ones who had delegated the thinking.

I’ve been working in IT for a long time. I’ve seen junior engineers grow into senior architects, and the path always involved struggle. Debugging code you don’t understand at 11 PM on a Tuesday. Reading documentation that makes your eyes glaze over. Writing something that breaks, then figuring out why. That struggle is where the learning happens. What concerns me is not that AI exists; I use it daily and find it genuinely helpful, but that we might be removing the friction that develops competence in the first place.

The full article is here: Anthropic Study: AI Coding Assistance Reduces Developer Skill Mastery by 17%

Open Source Is Drowning in AI Slop

My second article examined a problem I’ve been watching develop for months. Daniel Stenberg shut down cURL’s bug bounty after AI-generated submissions reached 20 percent of the total. Mitchell Hashimoto banned AI-generated code from Ghostty entirely. Steve Ruiz took it even further with tldraw, auto-closing all external pull requests. These aren’t fringe projects. cURL runs on billions of devices. These are maintainers reaching a breaking point.

RedMonk analyst Kate Holterhoff coined the term “AI Slopageddon” for what’s happening, and it captures the reality well. The flood of AI-generated contributions looks plausible at first glance but falls apart on inspection. The problem isn’t just quality, it’s volume. Maintainers are human beings with limited time, and they’re now spending that time sifting through submissions that an AI produced in seconds without any real understanding of the project.

A research paper from the Central European University and the Kiel Institute for the World Economy modeled the bigger structural risk here. Open-source projects depend on user engagement, documentation views, bug reports, and community recognition as a return on the maintainer’s investment. When AI agents assemble packages without developers ever reading the docs or filing bugs, that feedback loop breaks. The researchers tried to model a “Spotify-style” revenue redistribution. Still, the numbers didn’t work: vibe-coded users would need to generate 84 percent of the engagement that direct users currently provide. That’s not realistic.

I keep thinking about this one. My entire career has been built on open source, from the tools I integrate at work to the libraries I rely on for InfoQ articles. If the ecosystem that produces and maintains these tools becomes unsustainable because AI-generated noise overwhelms the people doing the actual work, we all lose. Not eventually. Soon.

More details here: AI “Vibe Coding” Threatens Open Source as Maintainers Face Crisis.

Is the Agile Manifesto Dead? Not So Fast.

The third article I wrote covered a debate sparked by Steve Jones, an executive VP at Capgemini, who declared that AI has killed the Agile Manifesto. His argument: when agentic SDLC systems can build applications in hours, the Manifesto’s human-centric principles no longer apply. If the tooling matters as much as or more than the people using it, then the Manifesto’s preference for “individuals and interactions over processes and tools” breaks down.

It’s a provocative claim that generated a lot of discussion. Casey West proposed an “Agentic Manifesto” that shifts the focus from verification to validation. AWS’s 2026 prescriptive guidance suggests “Intent Design” should replace sprint planning. Kent Beck, one of the original Manifesto signatories, has been talking about “augmented coding” as a new paradigm.

But here’s the counterpoint that keeps sticking with me. Forrester’s 2025 State of Agile Development report found that 95 percent of professionals still consider Agile critically relevant to their work. That’s not a methodology on its deathbed. And as one commenter noted in the discussion thread, bureaucracy killed Agile long before AI agents came along.

I think the question isn’t whether the Agile Manifesto is obsolete. It’s whether we’ve ever fully lived by its principles in the first place. The Manifesto says “responding to change over following a plan.” If there’s ever been a moment that demands responsiveness and adaptation, it’s right now. The irony of declaring Agile dead precisely when we need its core philosophy the most isn’t lost on me.

Full article: Does AI Make the Agile Manifesto Obsolete?

Connecting the Threads

When I look at these three stories together, I see a common tension. AI is accelerating what we can measure, lines of code produced, pull requests submitted, and applications prototyped, while eroding what is harder to quantify. Deep understanding of a codebase. Thoughtful engagement with an open-source community. The human judgment that sits at the heart of iterative development.

The Anthropic study shows that speed and learning pull in opposite directions, at least for developers acquiring new skills. The open-source crisis tells us that volume and quality are diverging at an alarming rate. The Agile debate tells us that our existing frameworks for organizing human work are straining under the weight of AI-driven change.

None of this means we should reject AI tools. I certainly won’t. But I think we need to be far more intentional about how we deploy them. That means designing AI assistants that support learning rather than replace it. It means building platforms that protect maintainers from low-quality noise. It means evolving our methodologies rather than abandoning them.

As someone who has spent years exploring new technologies, it’s one of the things I enjoy most about working in this field. I remain optimistic about where AI can take us. But optimism without caution is just naivety. The choices we make in the next year or two about how AI integrates into our development practices will shape the industry for a decade.

We should probably pay attention.