Nephila Radar [3/26 | Sep 28 – Oct 11]
On our radar
Andrea Daniele Signorelli, a journalist writing for Guerre di Rete, explains why many companies saw their AI spending explode in 2026. After OpenAI and Anthropic moved to usage-based pricing, Uber had already used up its annual budget by April, and Walmart and Meta capped usage, even though the price of a million input tokens fell from $30 (GPT-4, 2023) to $5 (GPT-5.5). What drives the bill is the volume of tokens consumed, multiplied by models that reason in steps.
Worth reading to understand why the AI bill keeps growing while the price per token falls, and what happens if usage-based pricing reaches individual users too: L'economia dei token e il vero prezzo dell'intelligenza artificiale - Andrea Daniele Signorelli
Adam Argyle, a design engineer at Shopify after seven years at Google, lists where AI helps on the front end (scaffolding, token migration, well-worn patterns) and where it stumbles: scroll-driven animations, layout math, accessibility, performance.
Worth reading to understand where AI saves time on the front end and where you still need someone who writes CSS: Why AI Sucks At Front End - Adam Argyle
Simon Willison, co-creator of Django and creator of Datasette, comments on OpenAI's acquisition of Astral, the company behind uv, ruff and ty. The most impactful is uv: more than 126 million downloads in a month. Willison trusts the team and expects the projects to be well maintained, but he flags two risks, the use of uv as leverage against Anthropic and a product-and-talent acquisition that later turns into a talent-only one, and wonders how credible forking is as an exit.
Worth reading to understand what changes for people who use uv now that its development moves to OpenAI, and how much the option to fork matters: Thoughts on OpenAI acquiring Astral and uv/ruff - Simon Willison
Juliana, a backend developer on the Nephila team, took part in Django on the Med, three days of Django development sprints in Pescara, where she was able to contribute actively and directly to the development of the framework. She says she had the opportunity to listen to discussions about important proposals for Django, led by the very contributors who were writing them:
transactions: the integration into Django of django-subatomic, the package by Kraken Tech, and async with transaction.atomic(), which is aiming to land in Django 6.2;
DEP 0022, the proposal to add import maps to Django;
Admin e JavaScript, the proposal to modernize the admin's JavaScript tests.
Worth reading to learn more about the latest developments driven forward by the Django community: Towards Async Transactions in Django - Carlton Gibson | DEP 0022: Adding import maps to Django - Matthias Kestenholz | Modernization of JavaScript unit testing in Django - Valentino Gagliardi
From the Nephila blog
In the third article of the series "AI: The illusion of productivity", Iacopo Spalletti, CTO and Solution Architect at Nephila, explains that running several agents in parallel changes the kind of work, from writing software to coordinating its writing, and multiplies cognitive load before output. To make the point, he draws on a real case, a reviewer-and-implementer loop between two agents that never closed, and on ActivTrak data.
Worth reading to understand what coordinating several agents costs in attention and why the speed of writing code is rarely the real limit: AI: The illusion of productivity - The coordination overhead problem - Iacopo Spalletti
Try it: how many CSS properties do you really know?
CSS or BS? is a game by Keith Cirkel where each turn shows the name of a property. Players have to guess whether it really exists in CSS or was made up. There are twenty rounds, you can answer with the arrow keys, and the difficulty ramps up as you go. Every answer gets a wry comment, and at the end of the game you get a verdict on your score.
Worth trying if you want to put your CSS instincts to the test: CSS or BS?
***
Did you find this list of resources interesting, or want to share your own? Join the conversation on our LinkedIn page: that exchange is the part we enjoy the most.
***
We wrote this article with the support of AI. If you're interested in how we use these tools as part of our writing process, you can read more here.
