#Ai-Agents
31 notas · última:
Spec-Driven Development: Make Specs Enforceable with ATDD
The page argues that Markdown specs alone do not enforce behavior. Acceptance tests, especially in ATDD, combine specification and verification so code must satisfy the expected outcome.
I don't know if my job will still exist in ten years
The author argues that software engineering may shrink fast as AI agents get better at writing, reviewing, and maintaining code, leaving engineers to supervise systems or exit the field.
9 Observations from Building with AI Agents
The author shares nine lessons from building AI agent systems, covering model choice, fine-tuning, static typing, team critique loops, prompt management, logging, and the cost-performance tradeoffs of current models.
Everyone is a Staff Engineer Now
The article argues that coding agents are pushing engineering work upward: execution gets cheaper, while planning, context management, architecture, review, and clear intent become the main bottlenecks.
How I use AI agents to write code
The author says AI agents are useful when paired with project guidance, tests, plan mode, and repeated review loops, even if they still make common mistakes and feel frustrating.
The Complete Guide to Building Agents with the Claude Agent SDK
The guide shows how the Claude Agent SDK wraps the agent loop, built-in tools, permissions, subagents, hooks, structured output, and MCP so you can build a code review agent and other agents with less manual orchestration.
Night Shift Agentic Workflow
Jamon Holmgren describes a day-shift/night-shift setup where he prepares specs during the day and lets AI agents work overnight with strict tests, docs, and review loops.
What the Team Behind Cursor Knows About the Future of Code
Cursor’s team argues that code work is shifting from typing in an IDE to directing and reviewing AI agents, with planning, model choice, and long-running workflows becoming central.
Agentic AI Explained: A Philosophical Framework for Understanding AI Agents
The post maps Socrates, Plato, and Aristotle onto AI agent personas to show how questioning, contextualizing, and analyzing can shape multi-agent collaboration and real-world problem solving.
AI Workflows vs AI Agents, Explained with LEGOs
The text compares AI workflows to LEGO-style assemblies with fixed steps and agents to constructions with a goal, showing when to use each and why hybrid approaches make sense.
Durable Objects are Made for Agents
Argues Cloudflare Durable Objects are a near-perfect primitive for building agents, combining a per-ID V8 isolate, paired SQLite storage and native websockets at roughly a tenth of AWS's cost.
AI Is Forcing Us To Write Good Code
A six-person team enforces 100% code coverage, thoughtful file naming, and fast disposable dev environments because coding agents need the guardrails humans used to skip under deadline pressure.
Don't Waste Your Back Pressure
Argues that agent leverage scales with the automated feedback ('back pressure') around them, like build systems, type checkers, and browser tools, not with the model alone.
Programming Languages in the Age of AI Agents
If AI agents write the code, does language choice still matter? Alexandru Nedelcu argues static typing sharpens the agent's feedback loop, but comprehension debt stays a human problem.
How to Write a Good Spec for AI Agents
A framework for specs that keep AI coding agents focused: start high-level, structure like a PRD around six core areas, work in modular prompts, and iterate continuously.
AI Needs Game Designers - David Kaye's Essays
David Kaye argues that running dozens of Claude Code agents in parallel is the same skill RTS players trained for years: attention allocation, not speed, and game designers already solved the interface for it.
Agent Jido: the Weather Agent example
A worked example shows how Jido, an Elixir agent framework, wires an LLM to a weather API tool through three layers: Actions, an AI Skill that renders prompts, and a signal-routed Agent process.
Fragments: February 13
Martin Fowler collects notes from a Thoughtworks retreat: senior developers staying relevant with LLMs, cognitive debt outpacing technical debt, and the drift toward supervisory programming.
Really Good Search is Really Hard
Abhi documents building Index, a hybrid search system for AI agents on Postgres/pgvector and Cloudflare Workers, that matched ChromaDB's retrieval quality with sub-300ms P99 latency.
After two years of vibecoding, I'm back to writing by hand
Mo Bitar traces two years of AI coding through excitement, spec docs, and what he calls slop, and explains why he went back to writing most of his code by hand.
12 Predictions for 2026 | Tomasz Tunguz
A VC's twelve bets for 2026: AI agents outearning people, a record IPO year, agent-first web design, and Cloudflare as gatekeeper of agent payments.
MiniMax M3 and the Cost of Long-Context Agents
MiniMax M3's sparse attention keeps per-step cost flat as context grows, making multi-hour autonomous agents affordable in production.
I don't like LLMs
Fowler feels both fascination and fear about AI, but talking to LLMs directly leaves him with one dominant reaction: dislike, traced mainly to the Silicon Valley culture that shaped these models.
I Never Want to Use Third-Party Software Again
The essay argues that software is moving toward hyperpersonalization: people will increasingly tune tools, workflows, and content for themselves rather than rely on uniform third-party products.
How Domain-Driven Design Changed My Approach to AI agent architecture
The author says Domain-Driven Design changed how he sees software design, helping him isolate the core domain of an AI coding platform and model agents through events instead of direct calls.
Custom GPT conducting a weekly review by voice
A custom ChatGPT with 15–20 questions, run in voice mode every week — turns the review from a static template into an interviewer that makes you think out loud.
The Rise of Agent Computers
The post argues that PCs are evolving into agent computers built to run AI agents full-time, locally and persistently, so people can delegate tasks instead of operating apps step by step.
Lessons from Building AI Agents for Financial Services
Nicolas Bustamante explains that financial AI agents depend on sandboxes, clean context, skills, S3-based storage, streaming, and evals, because small errors can destroy trust and money.
The State of AI: Global Survey 2025
McKinsey reports that AI use is broadening, but most organizations are still in pilots. The strongest results come from firms that redesign workflows, pursue growth and innovation, and scale AI agents.
I Vibe-Coded a Triton Kernel
The author describes trying to make GPT-OSS attention trainable by replacing a forward-only Triton kernel with a working backward pass, using Modal, ChatGPT, Claude Code, and Codex CLI to fix test failures until the harness passed.
Notte: Reliable Browser AI agents
Notte is a full-stack framework for building and deploying web automation agents, combining deterministic scripting with AI to reduce costs, improve reliability, and support structured output, sessions, vaults, personas, and scraping.