Short notes, pointers and references.
-
The proliferation of artificial intelligence in software engineering clearly has consequences for how to think about computing education.
[…] with GenAI, the production of code is no longer the bottleneck for students or professional engineers. Instead, the key challenges lie both before and after the production of code.
[…]
We must adapt our education mechanisms to address new metacognitive skills that are needed in a world with GenAI; curiosity, independence of thought, reading critically, and evaluating quality are paramount.
Computing Education When Writing Code Is No Longer the Challenge, by Ibrahim Albluwi, Dennis Bouvier, Claus Brabrand, Michelle Craig, Rodrigo Duran, Christopher Hundhausen, Colleen Lewis, Kevin Lin, Andrew Luxton-Reilly, Leo Porter, Karen Reid, David Smith, Claudia Szabo, Michel Wermelinger, Titus Winters, and Daniel ZingaroThere is the potential of students benefiting tremendously, if they “use GenAI as an ever-present and infinitely-patient tutor.” On the other hand, there is also the clear risk that they might “succumb to the temptation of unreflectively ‘outsourcing’ all effort to GenAI and submitting GenAI slop.”
Indeed.
-
Super-interesting article, wherein the authors reason that moral panic about AI companions (AICs) is unproductive, may even be harmful.
We are at a turning point: AI companionship is no longer niche. Panic-driven bans—in science or in public spheres—risk foreclosing not only research, but also any informed evaluation of potential benefits that must be weighed against risks. Blanket prohibition may offer the illusion of certainty or control, but it risks premature conclusions at a stage when the empirical balance of risks and benefits is not yet understood. And only with those empirics and conversations can we begin to understand where real potential lies, where risks demand action, and where fears may simply be noise.
Don’t Panic about AI Companions, by Jaime Banks and Jessica SzczukaI certainly agree with the authors that “moral panics are outpacing scientific understanding when we need ideologically unburdened research.” Panic is generally not a good position for initiating thoughtful decision-making.
Arguably however, the investment in and deployment of AICs and other AI systems likewise (and perhaps more so) far outpaces our understanding of them.
-
A recent study examined the impact of AI-assisted coding tools on perceived productivity, as measured against the dimensions of developer experience.
This study examines the impact of AI-assisted coding tools through a Developer Experience (DevEx) lens, investigating GitHub Copilot and Windsurf adoption in a fintech organization over three months. Despite productivity gains, developers face validation burdens in prompting and output checking.
S. C. Winckler, J. F. Ribeiro, L. Machado and J. Kroll, AI-Assisted Collaboration: Exploring Developer Experience with GitHub Copilot and Windsurf in IEEE Software, vol. 43, no. 04, pp. 38-46, July-Aug. 2026Flow improved, cognitive load worsened.
-
Some companies are bringing in or keeping fewer early-career professionals in software engineering.
Code may be getting easier and cheaper to produce, […] but maintaining complex software systems is a far more difficult task that, for the time being, still depends heavily on human judgment and expertise. At a growing number of companies, the declining ranks of new hires mean fewer people are gaining access to the kind of hands-on experience that has traditionally produced deep technical expertise.
The Vanishing Apprentice, by Alex WrightThis is not just affecting those early-career professionals who are missing out on learning and career development. The senior members in the engineering team are likewise deprived of opportunities to mentor and grow the more junior talent.
-
The inclusion of AI coding assistance affects the cognitive environment of the development experience:
AI coding assistants alter this [traditional] cognitive landscape [found in programming] by acting as external memory and cognition. This aligns with Hutchins’ theory of distributed cognition, which argued that cognitive systems often extend beyond the individual to include the external environment, which can extend and enhance individual cognition.
AI Didn’t Make Programming Easier. It Just Made It Differently Difficult, by Jeremy OsbornSo AI here allows shifting of cognitive load. It is efficient at generating solutions, but the human developer now incurs a higher cost for understanding and evaluating them.
Or, per the above article, “AI makes programming differently difficult rather than simply easier.”
The second-order effects of that are interesting to contemplate.
-
Deploying agentic AI is categorically different from simply using generative AI.
Generative AI agents can reason, plan, and take actions across enterprise systems, which means deploying them is not just a software installation but a change to how work gets done. When agents gain the ability to execute tasks—updating records, issuing refunds, routing approvals—they introduce operational risks that traditional software tools do not, including unpredictable behavior and unclear responsibility when things go wrong. To use them safely and effectively, organizations must treat them like digital employees, giving each one a defined identity, limited authority, trusted sources of information, clear controls over what it can execute, and audit trails that make its decisions explainable. Companies that adopt this mindset and introduce autonomy gradually will be far more likely to capture the benefits of agentic AI without exposing themselves to costly mistakes.
To Scale AI Agents Successfully, Think of Them Like Team Members, by Rahul Telang, Muhammad Zia Hydari and Raja IqbalThe article presents a thoughtful discussion on responsible deployment of AI agents in a corporate settings, focussed on identity, context, control and accountability.
-
There has been so much progress in AI tools for software engineering in recent years. When you see a paper based on a survey from more than a year ago, you immediately wonder how those surveys would go, if repeated now, with people using the current state of the art.
A key insight is the productivity paradox: AI tools can both enhance and hinder productivity, depending on interaction patterns, cognitive demands, and situational conditions. This duality underscores the need for sociotechnical models that avoid assuming uniformly positive effects.
L. Fortes, G. V. Pereira, A. Coelho, R. Prikladnicki and K. Kohl, “The Productivity Paradox of AI-Powered Development” in IEEE Software, vol. 43, no. 04, pp. 28-37, July-Aug. 2026.That seems safe and intellectually honest: It appears AI-powered development is more productive in some ways. It is not in others.
-
Tristan Harris‘ TED talk Why AI Is Our Ultimate Test and Greatest Invitation:
The comparison to the rollout of social media is powerful.
-
Recent findings show that understanding better how AI art is produced, influences how we think of it from a moral perspective.
When people knew how the AI system operated, they perceived the images it produced as less morally acceptable, especially when the creation of these images involved financial gain and artistic acclaim. But the aesthetic appeal of the images did not change, suggesting that learning how AI works made people reflect on ethics, not aesthetics.
People who know more about AI art find it less ethical, by Ionela Bara.Of course producers or sellers of that art are usually not really incentivized to get their potential customers to engage in that type of reflection.
-
From an exploration of how people are using (generative) AI in 2026, based on research by AI in the Wild:
As the breadth and depth of usage grows, so has the anxiety that people are surrendering their cognitive responsibilities to AI—a trend the authors call “thinkslop.” There’s also a parallel concern that they are relying too much on the technology for emotional support. In the business world, there is a lot of activity that produces marginal rather than game-changing benefits, so far.
How People Are Really Using AI in 2026, by Marc Zao-SandersMore people are using AI for more things. Relinquishing of agency appears to be a common denominator amongst some of the observed negatives.