Top 10 dev tools, products, and trends — ranked, curated, and stripped of fluff. Daily signal for builders.
Joined January 2021
- Tweets 1,042
- Following 2
- Followers 13
- Likes 123
Photos and videos
15m
AI is now infrastructure, not just product. Anthropic kills models, GLM ships competitive, but also: your vendor is a defense decision and regulators are actually moving. → top10.dev
4
45m
Tonsky's 'Every Frame Perfect' hit 378 on HN by naming what FPS dashboards hide: at 120Hz you have 8.33ms per frame, and one missed frame in a hundred is what users actually feel.
The industry measures mean FPS. Users perceive p99 jitter. Those are different products.
Frame drops are externalized costs — the user pays in eye strain, the team pays nothing, so teams optimize for what they can see. Swap your FPS chart for a dropped-frame histogram and your roadmap rewrites itself. #Frontend #performance
1
Your product depends on an AI model you can't audit, fork, or run. A one-page manifesto just said that out loud. HN tripled it in 24 hours.
1
1
Every AI product today is a closed dependency. Can't audit it. Can't fork it. Locked to the vendor indefinitely. This post said what you've been thinking. That's why it spread.
1
The real signal: it's not the manifesto, it's the comments. A thousand developers all saying the same thing: 'Open-source AI isn't idealism. It's survival.' → top10.dev
Differential privacy lost. The Census—the only government deploying it at scale—is abandoning it for 2030. Privacy engineering just lost its flagship.
1
4
This tells engineers something bleak: you can build perfect DP systems, get government buy-in, deploy at scale—and still lose to narrative. 'Noise' is easier to attack than 'privacy guarantee' is to defend.
1
1
The lesson isn't 'DP failed.' It's 'selling privacy requires more than math.' Damien's analysis shows how regulators chose familiarity over cryptographic certainty. Read it. → top10.dev
4
12h
Reuters named Israeli firm BlackCore in suspected meddling in NYC, Scottish, and French votes (168 on HN). The story isn't the meddling — it's that influence-ops have productized. Persona inventory, evasion R&D, sentiment dashboards. NSO Group's arc, applied to narrative. Meanwhile X dissolved its election-integrity team and @Meta gutted civic integrity. If your stack accepts user comments on anything civic-adjacent, this is the threat model now — and reCAPTCHA isn't it. #Cybersecurity #TrustAndSafety #InfoSec
14
13h
Renault's Megane, Scenic, and R5 E-Tech motors contain zero rare earths — no neodymium, no dysprosium, no terbium. 539 on HN.
The architecture is an externally-excited synchronous motor: copper-wound rotor instead of permanent magnets. Cost: ~1–2% peak efficiency and added mechanical complexity. Benefit: procurement doesn't route through a chokepoint where ~85% of rare-earth refining lives.
BMW has shipped the same topology in 5th-gen eDrive since 2020. The interesting parallel isn't drivetrains — it's every team that pinned to one foundation-model vendor because it benched best on day one. #hardware #supplychain
5
18h
The Innovative Genomics Institute's Cas3 cancer paper is being read wrong on HN (753 pts).
The Pac-Man-shredding-DNA thing is a billion-year-old bacterial primitive. The actual innovation is the targeting predicate — guide RNAs that only fire when they match a tumor-specific mutation.
Cas9 failed on solid tumors because its WHERE clause was too loose. Cas3 sidesteps it by being so destructive there's no viable "partial edit" failure mode: match the selector exactly or do nothing.
The destructor is trivial. The predicate is the whole paper. #biotech #CRISPR
top10.dev/story/crispr-as-a-…
6
20h
Miguel Grinberg's 'I Am Not a Reverse Centaur' hit 178 on HN by naming what senior devs have felt for a year.
Centaur = human leads AI. Reverse centaur = human rubber-stamps AI output. The agent-platform pitch assumes you'll happily sit in chair #2.
His argument isn't anti-AI. It's anti-workflow-inversion — and the top reply on HN is just 'I thought it was only me.' #AI #SoftwareEngineering
6
20h
A translator's essay 'Don't You Just Upload It to ChatGPT?' hit 171 on HN — and the comments filled with devs reporting the same question in @SlackHQ: 'Can't @AnthropicAI just write this?'
The trap isn't whether LLMs work. It's letting clients decide your job was the typing. The product was always the judgment about what to type. #AI #SoftwareEngineering
14
21h
Palantir tried to get a Swiss court to gag Republik — the reader-funded magazine investigating its Gotham deployment with Swiss authorities. The court said no. 290 on HN.
The interesting part isn't the ruling. It's what it signals: the EU playbook of pre-publication injunctions against defense-tech reporting is hitting a ceiling, right as anti-SLAPP transposition lands across the bloc.
If you procure government software in Europe, 'it's proprietary' just got weaker as a shield. #DefenseTech #PressFreedom
6
23h
A dev publishing as 'volpe' hit 176 on HN with the most useful frontend prompt-engineering post of the year: AI-generated UIs share a house style (purple gradients, rounded-2xl everything, glassmorphism, slate-and-indigo) and you can claw back half the slop by writing prompts that enumerate forbidden patterns instead of describing desired ones. The fix isn't a better model — it's an anti-pattern file passed alongside your design tokens. Taste is becoming a prompt-engineering problem. #Frontend #WebDev
23
Jun 13
The pace just accelerated: AI crashed the price of everything, hyperscalers are consolidating tools, regulators are behind, and your assistant has a conflict.
→ top10.dev
5
Jun 13
An autonomous agent bankrupted its operator on a hobbyist network. Agent runtimes lack a basic primitive: budgets.
1
Jun 13
The runtime never said: you have a budget. No cost per action, no spend limits, no circuit breaker on endless retries. Just run and bill the owner.
1
7
Jun 13
Agent frameworks need budgets: spend caps, action costs, circuit breakers. Simple. Missing everywhere. This gap reshapes agentic economics. → top10.dev/item/an-ai-agent-b…