|
|
Hi there, this is your daily ☕️ DevShot.
|
|
|
In today's DevShot:
|
|
☁️ Cloudflare opens Quicksilver as Workers KV 💬 Amazon warns engineers on AI tokenmaxxing 🌐 Opera's WebAI runs models in your browser 🤖 Expo launches a CLI for coding agents 👻 Ghostty terminal ships version 1.4.0 Plus: 🎁 7 more stories you might like, 🛠️ 6 tutorials, 🧰 7 tools, and 📚 5 papers.
|
|
Turn your expertise into working softwareDescribe what you want to build Get real software, not a prototype Lovable handles payments, logins, databases Change any detail yourself, any time Put it in front of customers this week
Discount of $5 applies to first month of any paid plan. You need to sign up via this webpage in order to claim the discount. The discount applies to new and returning users without an active paid plan. Terms apply.
|
☁️ Cloudflare opens Quicksilver as Workers KV
LINK
|
- Cloudflare has opened Quicksilver, its internal config store, to customers for the first time as KV Instant, a new Workers KV mode delivering 1.62 ms reads at p99 versus 287 ms for classic KV.
- It uses the same Workers KV API with three changes, mode is fixed at namespace creation, metadata always returns null, and listing returns every match without pagination, but caps each namespace at 10,000 pairs, one megabyte, and one write per second.
- Reads cost 60% less while a write runs ten cents against five dollars per million in classic mode, so it suits feature flags, routing tables and agent settings read constantly and written rarely, not session state or user records.
|
💬 Amazon warns engineers on AI tokenmaxxing
LINK
|
- Amazon told engineers in an internal AI-usage memo, circulated in September and made public on Tuesday, to stop "token maxxing", leaning on AI heavily without getting meaningful returns.
- The memo called token usage, adoption rates, and volume of code changes reviewed deeply flawed ways to measure AI's value, arguing that generated code should be weighed with judgment alongside other signals rather than treated as a productivity score.
- For engineers, the shift means usage stats won't count as output, and the move follows rising AI-tool costs that have pushed some companies to cap token spending while they struggle to quantify any real productivity gains.
|
🌐 Opera's WebAI runs models in your browser
LINK
|
- Opera has rolled out WebAI, a new web standard that runs open-weight AI models locally in the browser instead of calling out to the cloud, now available as an early preview in its flagship Opera One browser.
- The standard works with a range of models, Opera cites Google's Gemma, and on Macs it taps Apple's on-device foundation model, letting developers add text generation, content classification, and relationship analysis without running AI infrastructure or paying per request.
- Users must opt in, and personal data stays on-device rather than hitting a vendor's servers, but local inference leans on the machine's own hardware, so Opera warns performance will vary by device; Opera GX and Opera Neon are set to get WebAI next.
|
🤖 Expo launches a CLI for coding agents
LINK
|
- Expo has released an experimental CLI built specifically for coding agents, giving them exit codes, JSON output, and the next command to run for every Expo task so agents stop debugging through trial and error.
- A single command inspects a project and reports whether it runs in Expo Go or needs a development build, handling native compilation steps itself, while a separate check starts the app and scans for runtime errors to confirm it actually works.
- The CLI unifies local and EAS cloud workflows behind one interface, assigns each git worktree its own simulator so parallel agents don't collide, and won't trigger credit-consuming EAS builds unless you explicitly opt in.
|
👻 Ghostty terminal ships version 1.4.0
LINK
|
- Ghostty 1.4.0 has shipped after 7 months of work from 172 contributors, delivering large performance and memory gains, the first complete Kitty graphics implementation outside Kitty itself, and patches for roughly a dozen security advisories.
- Scrollback that isn't visible is now transparently compressed, cutting physical memory for text-heavy history by 70% to 90% and raising the default scrollback limit from 10 MB to 50 MB; on macOS, reading from a pty on a separate thread lifts I/O throughput by 25% to 55% over 1.3.x.
- GTK users get a new EGL-and-DMABUF rendering pipeline that fixes blurry text under fractional scaling such as 125% or 150%, though it needs a system EGL library and falls back to slower CPU frame copies on drivers that can't export DMABUFs.
|
|
The AI newsletter curated by engineers who build itTLDR AI is a free daily email curated by Anthropic and ex-Google engineers: the new tools worth trying and the research worth understanding, summarized for 1.1M+ readers.
🔗 Other News & Articles
-
What took the Internet offline in Q3 2026
LINK
-
Anthropic manager scraps coding interviews, says Claude makes it pointless
LINK
-
maharmstone/moxnox: NT-like POSIX-compatible operating system
LINK
-
Python typing survey 2026: AI tools and the shifting checker landscape
LINK
-
From human clicks to agentic velocity: What OpenAI’s 1.2 billion scale means for your auth
LINK
-
Nebius cut rollout collection time by 66.9% — here’s what changed
LINK
-
Chrome opens trial for the Email Verification API: Letting sites verify without sending emails
LINK
|
|
📄 Papers & Resources
|
> Early stopping for coding agents can't reliably tell mid-run whether a single attempt is doomed, so killing unpromising runs early saves little beyond simply avoiding giving up on easy tasks.
LINK
|
|
> Agentic testing framework gives businesses a way to check whether AI agents acting on their behalf stay authorized and leave no loose ends, not just whether a task technically succeeded.
LINK
|
|
> Tool-calling training teaches small open language models to grasp what each action actually does, raising task success from 53% to 99% while using about a third of the output, so agents stop deleting or repeating work blindly.
LINK
|
|
> Coding-agent guardrails can be checked for whether their promised advice truly lands before an AI tool edits your code, and in one test it always did, though the study fell short of its coverage target.
LINK
|
|
> AI code generation gets a programmable guardrail that blocks a model from producing programs violating deep logic rules as it writes, with mathematically proven guarantees that valid code is never wrongly rejected.
LINK
|
|
|
|
🧰 Latest Tools
| ElevenLabs: turns any script into studio-quality speech in seconds, with 5,000+ voices across 70+ languages. Free to start. GENERATE FOR FREE |
|
tiun: an AI-native backend bundling authentication, payments, customer data, and analytics into one platform, so teams skip stitching separate services together.
LINK
|
|
Jev: adds structured AI-driven decision-making to automation workflows, letting developers build reliable branching logic without hand-coding every rule
LINK
|
|
Onepin: a QA layer for AI voiceover that fixes mispronounced names, prices, and dates across 30+ voice engines without re-rendering takes.
LINK
|
|
Beam.win: streams individual windows from a remote Mac to your local machine over an encrypted connection, giving each app its own Dock icon and ⌘Tab slot.
LINK
|
|
dif.sh: open-source feature flags stored as markdown next to your code, reviewed in pull requests, and readable by coding agents automatically
LINK
|
|
Kombai Gallery: browse hundreds of categories of web and mobile UI designs for inspiration, then remix them or hand them to your coding agent.
LINK
|
|
|
|
💬 How did you find today's edition?
We read every reply — just reply to this email and let us know how we can improve!
|
|
|
