IT is not magic pixiedust

The Ai Acceleration Whiplash

The AI Engineering Report 2026 is not a survey of how developers feel about AI. It is a measurement of what AI is actually producing across the full software development lifecycle, tracking metric change between periods of lowest and highest AI adoption within each organization.
What it found has a name: the Acceleration Whiplash. AI has flooded a system built around human-paced development and human-quality code with output it was never designed to absorb.

Study finds AI tools made open source software developers 19 percent slower:

Before performing the study, the developers in question expected the AI tools would lead to a 24 percent reduction in the time needed for their assigned tasks. Even after completing those tasks, the developers believed that the AI tools had made them 20 percent faster, on average. In reality, though, the AI-aided tasks ended up being completed 19 percent slower than those completed without AI tools.

Findings backed up by more recent researtch (PDF) from Harvard:

We study the impacts of AI coding assistants and agents on software engineering work, usinga novel proprietary dataset from an engineering analytics platform, covering 300 million work events — including GitHub coding activity, Jira issues, and Google Calendar events — across 718 firms. We use a staggered difference-in-differences design, exploiting variation in firmlevel adoption timing of AI coding assistants and agents. Both technologies increase coding productivity. However, productivity gains do not fully pass through to changes in software output or employment. For AI agents, this incomplete pass-through reflects a bottleneck from code review: review times increase, a larger share of code updates require revisions, and reviews involve more comments. We develop a model of software production to structure these results, in which AI affects both productivity and quality of intermediate outputs, and in turn can generate a review bottleneck and limit pass-through.

AI sitcom accidentally a Lynchian fever dream. The sort of general malaise that only the genius possess and insane lament.

[image or embed]

— 𝔻𝕒𝕟 🏈🍻💀 (@danmynrd.bsky.social) October 29, 2025 at 10:47 PM

AI assistance reduces persistence and impairs independent performance:

AI assistance reduces persistence and impairs independent performance: After brief AI-assisted sessions (~10 minutes), participants were significantly more likely to give up on problems and performed significantly worse once the AI was removed, compared to participants who never had AI assistance.
Effects are concentrated among users who seek direct solutions: Persistence costs were concentrated among participants who prompted AI to solve tasks for them directly. Using AI for hints or clarifications did not produce significant impairments.
Effects generalize across domains: Effects replicated across fraction arithmetic and reading comprehension, suggesting it is a general consequence of AI-assisted problem solving, not specific to any particular task.

The Rise of AI Is Making Life Even Harder for Real People in Gaza :

The 22-year-old has endured almost two years of war in Gaza, as Israel continues an assault that has killed at least 57,000 people and left the territory of 2 million on the brink of starvation. Saeed has been raising money online to feed his family. But many people on social media have been convinced that he isn’t real at all. They believe he’s an AI-generated creation, used by scammers to get donations through crowdfunding platforms from well-meaning people. And in a world where AI videos and images have flooded our digital spaces, it’s easy to understand why some people are so skeptical.

AI-driven productivity gains enable more CO₂ emissions than they avoid:

Under parallel adoption scenarios, net annual CO₂ emissions increase by 0.47–1.8 gigatonnes (1.2–4.8% of 2024 global energy-related CO₂ emissions). Enabled emissions exceed avoided emissions whenever fossil-sector gains are nonzero; net emissions reductions require renewables gains 4–5× greater than fossil fuel gains. Absent policy steering, AI’s modeled effects increase the carbon intensity of the global economy and reinforce fossil fuel incumbency—outcomes that current analytical and governance frameworks do not fully capture.

Pito Salas talks about one of the less obvious pitfalls of using AI in your development work:

The problem is that while it an almost correct piece of code it is often the totally wrong solution to your problem. In other words, it often sets you on the wrong path. A dead end. The trap is that you spend your time debugging an approach that is totally wrong.

When does this happen? I find that it happens when I am tackling something I don’t really understand and want a shortcut to my sugar high. Upon reflection I have found more than once that I spent hours debugging an approach which, once working, I realized was all wrong and I had to start over.

Which reminds me a lot of myself when trying to solve some technical issue by gleaning possible solutions from Stackoverflow. Not being a developer I still run into coding or other IT issues in my job as software tester, or even in real life and what do I do when I can’t fix it on my own? I google it, or rather look at Reddit or Stackoverflow for possible solutions. The number of times I found something that loos like it might work for my own situation, but turned out not to apply… So easy to get stuck on a particular solution when you actually lack the insight to understand why it’s the wrong one. At least with Reddit or Stackoverflow, there’s still the possibility of asking for help when you don’t understand something…

But there are worse things than AI generating bad or unsuitable code. The true danger lies in what happens when AI generated code is good enough to take into production. AI generated code may solve your immediate problem, but how maintainable is it? How understandable?

Let me explain. In my professional life as a software tester I often end up working on projects using code that can be years or even decades old; it’s not unusual that all of the original programmers have left the team. Given developers usual loathing of commenting, let alone documenting, what you end up with, even in the best of circumstances is code that nobody present understands completely anymore, nor understands why certain design decisions were made.

AI generated code turbocharges this situation. You’re now using code where nobody understands the design decisions made because AI cannot make decisions, with code generated in ways humans could never come up with. That’s a recipe for disaster a couple of iterations down the line.

Everybody had their fun mocking this dumb AI shill already, but that won’t stop me:

First off, if mr Future here is drinking coffee for 58 minutes, what’s he doing that other hour? Pissing it all out again?

Second, what kind of (I assume) work related book that you need this complicated AI setup for to get summarised will only take two hours to read anyway?

Third, the weird exactness of “the 17 models I’ve selected”. Why seventeen? Why not sixteen, or eighteen? Why mention a million books too? Just the first random Big Number that popped in his head? Bad enough having seventeen different AIs to summarise a book, but having a separate agent then collating this “raw output” and summarise it even further is begging for errors.

It all sounds very cool and ccyberpunky, something you could’ve read in a late eighties hack sf novel ripping of Gibson or Sterling, but in real life it all sounds a lot like using the fairies to ask Mothra for help. All it does is make it clear you don’t want to do your job, which lords know I can sympathise with, but also that you don’t know how to do your job, if you think reading a summary of a book is the same thing as reading the book itself. What makes you think you can judge what the AI regurgitates for you if you’re too lazy to do the work yourself?

For those who don’t like paying licence fees, Microsoft is offering a new version of Office, but it comes with some limitations:

Office 2024 packs all the modern features Microsoft has added to its popular apps in the past few years, but that’s about it. The suite comes “locked in” and won’t receive any further upgrades. Users will also miss out on real-time collaboration and Copilot AI integration, which are available with Microsoft 365.

No AI? Oh no! How ever will I cope?