Notas
Índice cronológico de todas as notas do Scholion.
5359 notas · última:
Deep Agents
The package wraps planning, subagents, a mocked file system, and a detailed prompt into a general-purpose deep agent. It shows how to create and customize one with tools, instructions, subagents, and a model.
I Automated My Entire Dev Workflow with AI (You Won't Believe How Easy It Is)
The author describes how MCP and FastMCP let Claude interact with APIs, files, terminals, and databases to automate keyword research, support summaries, task prioritization, and user feedback reading.
tokens are getting more expensive
The post argues that cheaper models do not fix AI margins because demand shifts to the newest frontier model and agents now consume far more tokens, making flat-rate subscriptions unsustainable.
Will data centers crash the economy?
Noah Smith argues that the AI data-center boom could slow the economy and, if its debt is concentrated in private credit, banks, and insurers, create conditions for a financial crisis.
Meta pirated and seeded porn for years to train AI, lawsuit says
A new lawsuit says Meta used BitTorrent to pirate and seed adult videos for years, possibly to train AI models, and that the pattern may have exposed minors to those files.
Underused Techniques for Effective Emails
The page argues that effective emails save time, reduce misunderstandings, and earn recognition. It recommends leading with the key point, writing complete replies, splitting drifting threads, and using structure and Markdown.
Unlocking high software engineering pace: Build less, ship frequently, learn faster – Jim Grey on software management
The article argues that teams should ship the smallest useful version early, get real-user feedback, and learn faster instead of spending many sprints building a full feature before anyone can use it.
If Writing is Thinking…
The page argues that in business, science, and Wall Street, people rarely read deeply, so AI-written or AI-summarized text can amplify a long-standing problem when even authors do not fully know what was written.
How to Eat Breakfast Like a Wellness Blogger
The piece gathers breakfast routines from wellness bloggers and shows how they work with different amounts of time, from overnight chia pudding to Starbucks fallback orders, while also explaining how they stage the food for Instagram.
Introducing Phi-4: Microsoft’s Newest Small Language Model Specializing in Complex Reasoning
Microsoft introduces Phi-4, a 14B-parameter small language model that focuses on complex reasoning, especially math, and is available on Azure AI Foundry and Hugging Face.
Recado ao Sr. 903
Um morador do 1003 responde à reclamação do 903 sobre barulho, aceita as regras do prédio e a ordem dos números, mas sonha com uma convivência em que a porta aberta substitui a queixa.
Enhancing Your Elixir Codebase with Gleam
The article shows how to add Gleam to an Elixir/Phoenix project, move core enrollment logic into Gleam, call it from Elixir, and extend the setup to handle a waitlist.
How I Stopped Worrying and Started Loving the Umbrella – TAGBASE
The author explains why Tagbase moved from separate Node.js and Elixir apps to an Elixir umbrella app, focusing on shared code, clearer boundaries, faster local development, and safer separation between portal and back office.
Set up HTTP Tunneling in minutes with Tunnelmole
The article explains HTTP tunneling, how it forwards public requests to a local server, and why developers use it for webhooks, demos, device testing, API work, and bypassing network blocks.
Run a Raspberry Pi server at home in minutes with Tunnelmole
The article explains why Raspberry Pi works well as a home server and how Tunnelmole exposes a local service to the internet through a secure tunnel, avoiding port forwarding, static IPs, and CGNAT.
How to Write Effective Incident Post-Mortems: A Complete Guide
The guide explains what an incident post-mortem is, why it matters, and how to structure it with summary, timeline, root cause analysis, action items, and blameless follow-up.
Logging Performance Improvements with Source Generators in C# .NET
The post explains how source-generated logging in .NET uses `LoggerMessage` on partial methods to reduce overhead, improve type safety, cut boilerplate, and keep log messages consistent.
Local-first software: You own your data, in spite of the cloud
The essay argues that software can keep cloud-like collaboration and access from any device while giving users local ownership, offline use, longevity, privacy, and control over their data.
Using AI Without Leaving the Terminal: A Guide to llm
The guide presents llm as a terminal-based interface to multiple language models, with logging, plugins, pipes, local models, embeddings, templates, and workflow automation.
Understanding Managed Services in Software Engineering
The post contrasts managed and unmanaged infrastructure as a tradeoff between control and convenience, then uses IaaS, PaaS, and FaaS to frame how that choice affects cost, scaling, security, and team focus.
Project of the Week: Excalidraw
The page describes Excalidraw as an open-source virtual whiteboard with real-time collaboration, end-to-end encryption, and a maintainer-led workflow shaped by fast reviews, short wait times, and limited bot use.
The Unwritten Rules to Becoming a Senior Developer: 4 Steps to Level Up
The article argues that senior growth comes from solid coding, team-oriented thinking, handling fires, and writing more to share knowledge. Titles matter, but promotion timing and career value are not the same thing.
Tatami Rooms & Japanese Minimalist Bedrooms – How to Create Your Own – Comfort Pure
The page explains how a Japanese minimalist bedroom uses tatami mats, shikibutons, low furniture, and natural materials to create a calm, functional space that supports relaxation, mindfulness, and floor-level living.
Why you should use Ash?
The chapter argues that Ash is worth learning because it centralizes domain logic, reduces duplication, and helps teams manage growing complexity, onboarding, and long-term maintenance.
Senado aprova texto alternativo da Câmara e suspende aumento do IOF
O Senado aprovou PDL que suspende três decretos sobre o aumento do IOF, restabelece a regra de 2007 e expõe a disputa entre arrecadação, diálogo político e previsibilidade fiscal.
The AI lifestyle subsidy is going to end
The article argues that consumer AI is being subsidized now by venture capital, but that this will fade as companies move toward ads, shaping discovery and making user experiences worse.
São João Festival 2025 | Porto's Best Night Of The Year
The page presents São João do Porto as Porto’s biggest annual celebration, with street parties, manjericos, lanterns, martelos, bonfires, boats, fireworks, and Cats Porto’s rooftop party.
Technology, Symbols, and How Reality Loses Its Meaning
The article uses Baudrillard to argue that media, algorithms, and AI push us from reality toward simulacra, overload meaning, and reward sameness, while creativity and agency require resisting that drift.
Entrevista: Banco Central, auto custódia de stablecoin e o futuro do mercado cripto no Brasil
Erik Oioli diz que o Banco Central deve manter a restrição à transferência de stablecoins para autocustódia e que o mercado terá de se adaptar para operar legalmente no Brasil.
How to Design Twitter (X) in a System Design Interview?
The article outlines a structured way to answer a Twitter-like system design interview by covering requirements, capacity estimates, APIs, database tables, caching, load balancing, and failure scenarios.
The Illusion of Thinking
Apple argues that frontier large reasoning models improve on some reasoning benchmarks, but their accuracy collapses past certain problem complexities, their reasoning effort can fall as complexity rises, and they show limits in exact computation.
Why do people disagree about when powerful AI will arrive?
The article contrasts short- and long-timeline arguments for AGI, using benchmarks, task length, AI research automation, scaling, expert surveys, and limits from real-world work to explain why forecasts diverge.
MCP is the coming of Web 2.0 2.0
Anil Dash argues that MCP matters less as a spec than as a sign that open, interoperable protocols can spread again, reviving the Web 2.0 ideal of programmable platforms.
How Facebook Sets Goals - Analytics at Meta
The article explains Facebook’s goal map, how team goals ladder up to org and company goals, how goal tiering and countermetrics shape prioritization, and how targets are set and measured.
System Design — Tips
A practical list of system design choices for scalability, availability, latency, and reliability, with examples showing when to use caches, queues, sharding, replication, and other patterns.
How I Study Consistently With A Full-Time Job
The author says study consistency comes from gamifying goals, using time blocks, and tracking progress, which helped him learn while working full time and move toward cloud engineering.
AI and cognitive offloading: sharing the thinking process with machines
The article argues that AI should be designed as part of distributed cognition, helping users offload effort through familiar patterns, progressive disclosure, contextual guidance, and graceful failure.
The AI behind Uplift
Adyen describes the AI stack behind Uplift, from connected decisioning and offline evaluation to weak supervision, transformers, observability, fairness, and explainability in payment flow models.
How Adyen does AI
Adyen says it uses ML on every transaction for fraud, authentication, routing, and conversion, while its GenAI work focuses on support, human-in-the-loop workflows, and open-source models hosted in-house.
I'd rather read the prompt
The essay argues that language-model writing usually buries original thought, adds vague filler, and should not replace a human voice, especially in school, reviews, and creative writing.