Posts

The JetBrains ACP Agent Registry is a curated, one-click marketplace for AI coding agents inside IntelliJ IDEA, PyCharm, WebStorm, and other JetBrains IDEs. Launched January 28, 2026, it lets you install Claude Code, Cursor, Gemini CLI, and 30+ other agents in seconds — no manual JSON editing required. What Is the JetBrains ACP Agent Registry? The JetBrains ACP Agent Registry is the world’s first open, cross-editor AI agent marketplace, jointly built by JetBrains and Zed Industries and launched on January 28, 2026. It solves a problem that frustrated developers for years: every AI coding agent had its own proprietary installation process — download a binary, edit JSON config files, restart the IDE, repeat. The registry replaces that friction with a browser-like “one-click install” for any ACP-compatible agent directly inside IntelliJ IDEA, PyCharm, WebStorm, GoLand, and other JetBrains IDEs running version 2025.3 or later. As of mid-2026, the registry lists 30+ agents including Claude Code, Cursor, Gemini CLI, GitHub Copilot, OpenHands, Kimi CLI, Goose, Cline, and Koog (JetBrains’ own Junie agent). The registry is open — any developer or company can submit an ACP-compatible agent for inclusion. Both JetBrains and Zed share the same backend registry, meaning an agent listed there works in both editors without duplication. ...

Ollama is an open-source local LLM runtime that exposes a REST API on http://localhost:11434, letting you run Llama 4, Qwen3, DeepSeek R1, Gemma 4, and 4,500+ other models entirely on your machine — with zero per-token cost and no data leaving your network. The OpenAI-compatible /v1/ layer means most existing SDK code works after a one-line base_url change. Why Local LLMs Went Mainstream in 2026 Local LLM adoption crossed a meaningful threshold in 2026, driven by economics, privacy regulation, and dramatically improved model quality in small footprints. Ollama surpassed 170,000 GitHub stars — the most starred local LLM runtime project on the platform — and monthly downloads grew from 100K in Q1 2023 to 52 million in Q1 2026, a 520x increase in three years. The stat that matters most for developer decision-making: 42% of developers now run at least some LLM workloads entirely on local machines, up from single digits in 2023. The economic case is straightforward — a team of five developers can spend $3,000–$30,000 in cloud LLM API costs over a three-month development cycle before shipping a single production feature. Local inference eliminates that cost entirely during the iteration phase. HuggingFace now hosts 135,000 GGUF-formatted models optimized for local inference, up from just 200 three years ago, giving developers access to a deep catalog. For regulated industries — healthcare, finance, government — local deployment isn’t just economical, it’s frequently mandatory: patient data, financial records, and classified documents cannot traverse cloud APIs. Ollama handles this by design. ...

JetBrains’ HAX team tracked 800 developers and 151,904,543 IDE events over two years and presented findings at ICSE 2026 in Rio de Janeiro. The headline: AI doesn’t just speed up development — it redistributes and reshapes how developers work in ways their own perceptions consistently miss. 74% of AI-assisted developers didn’t notice increased window switching, yet telemetry confirmed it was happening the entire time. What JetBrains’ April 2026 Research Actually Found (And Why It Matters) JetBrains’ April 2026 research is significant not because it reports new productivity statistics — the ecosystem has plenty of those — but because it is one of the first large-scale longitudinal studies to compare what developers believe about their AI-augmented workflows against what objective behavioral telemetry actually shows. The study, conducted by JetBrains’ Human-AI Experience (HAX) team and presented at ICSE 2026, analyzed 151,904,543 logged IDE events from 800 developers over two years (October 2022 to October 2024). Sixty-two developers completed follow-up surveys and interviews. The core finding challenges the dominant narrative: AI tools do not primarily speed up the same work. They redistribute it. Tasks that previously required focused writing time shift toward validation, review, orchestration, and context-switching. The net effect is a fundamentally different developer rhythm — more output, more deletion, more cognitive overhead — that developers themselves systematically underestimate. For engineering teams planning AI tool adoption or evaluating current tooling, this data is more actionable than headline productivity percentages. It names the actual mechanism of change so teams can measure and manage it. ...

TanStack Query v5 is the server state library for React that handles caching, background refetching, and stale-while-revalidate out of the box — with 12M+ weekly downloads in 2026, it’s become the default choice for teams building AI-powered applications that need real-time data and LLM streaming. Why TanStack Query v5 Is the Default Choice for AI-Powered React Apps in 2026 TanStack Query v5 (formerly React Query) is a server state management library that handles all the complexity between your React components and your data sources — caching, deduplication, background synchronization, loading states, and error recovery — with minimal configuration. As of June 2026, the library ships at version 5.100.14, has 48K+ GitHub stars (overtaking SWR’s 32K in 2024), and sits at 12.3M weekly npm downloads — a 2.5x lead over SWR’s 4.9M. That adoption reflects a real shift: AI-powered React apps need capabilities beyond simple data fetching. LLM responses stream over seconds, not milliseconds. Dashboards pull from three or more data sources simultaneously. Users open the same AI tool across five browser tabs. TanStack Query v5’s new streamedQuery, broadcastQueryClient, and deep Suspense integration address these patterns directly, which is why teams building AI chatbots, real-time dashboards, and LLM-augmented features in 2026 are reaching for it first. ...

React Server Components (RSC) are components that run exclusively on the server, never ship JavaScript to the browser, and can access databases and file systems directly. In Next.js 15 App Router, every component in the app/ directory is a Server Component by default — you opt into client-side interactivity with 'use client', not out of it. This guide covers the complete RSC mental model, data fetching patterns, streaming, Server Actions, caching, Partial Prerendering, and the 7 mistakes that silently wreck bundle size. ...

Salesforce의 AWU(Agentic Work Unit)는 AI 에이전트가 완료한 하나의 개별 작업을 의미합니다. 토큰이 AI가 얼마나 많이 “말했는지"를 측정한다면, AWU는 AI가 실제로 얼마나 많은 작업을 완료했는지를 측정합니다. 개발자에게 AWU는 Agentforce 비용을 이해하고 예측하며 최적화하는 핵심 단위입니다. What Are Salesforce Agentic Work Units (AWU)? An Agentic Work Unit is a discrete, measurable action completed by a Salesforce AI agent — one unit of work executed on behalf of a customer or employee, tracked independently of how many tokens that work consumed. Salesforce CEO Marc Benioff introduced the metric during the Q4 FY2026 earnings call on February 25, 2026, positioning AWUs as the industry-standard way to quantify AI agent productivity rather than raw token volume. As of Q1 FY2027, the platform has processed over 19 trillion AI tokens translating to 3.8 billion total AWUs, with 1.6 billion AWUs generated in a single quarter — a 111% quarter-over-quarter growth. The key insight for developers: AWU is elastic. Salesforce’s stated goal is to deliver more AWUs from fewer tokens as model efficiency improves, meaning the same budget should fund progressively more agent work over time. Whether that promise holds depends directly on how well you architect your agents. ...

Most teams measuring AI coding ROI are looking at the wrong numbers. Developers feel faster, acceptance rates look great, and vendor dashboards show impressive gains — but when you trace those numbers back to shipped features and business outcomes, the story falls apart. The disconnect is real. The METR study found developers felt 24% faster with AI coding tools but were actually 19% slower — and still reported 20% perceived improvement afterward. That gap between perception and reality isn’t just a curiosity; it’s where your ROI evaporates. ...

GPT-6 is not Spud. Spud shipped as GPT-5.5 on April 23, 2026 — a significant but differently-named model. The real GPT-6 is the next-generation system in OpenAI’s pipeline, and Polymarket traders give it 84% odds of releasing by December 31, 2026. Here is exactly what to change in your codebase now so that GPT-6 is a one-config-line upgrade, not a week-long rewrite. What Is GPT-6 (Spud)? Understanding the Naming Confusion GPT-6 (sometimes called “Spud” by developers) refers to the next major OpenAI model after GPT-5.5 — but the Spud codename has caused significant confusion in the developer community. The model internally codenamed “Spud” actually shipped on April 23, 2026 as GPT-5.5, not GPT-6. This naming slip caused many developers to believe GPT-6 was already live. It is not. GPT-5.5 achieved an Intelligence Index score of 60 on Artificial Analysis, topping all 153 reasoning models on the leaderboard at launch. Its API pricing is $5 per 1M input tokens and $30 per 1M output tokens — exactly double GPT-5.4. The real GPT-6 is the next-next model: it is expected to deliver a 40% performance improvement over current models in coding, reasoning, and agentic tasks, and to feature a 2 million token context window (double GPT-5.5’s 1M limit). For developers, the practical takeaway is straightforward: any code that hardcodes "gpt-5.5" or references Spud directly will need to change when GPT-6 lands. Start abstracting now. ...

Tokenmaxxing is the practice of maximizing AI token consumption as a proxy for engineering productivity — and it’s quietly destroying code quality, blowing AI budgets, and making developers measurably less effective. If your team celebrates high token usage without tracking what that code actually does downstream, you’re already in the trap. What Is Tokenmaxxing? The AI Productivity Myth That’s Costing Millions Tokenmaxxing refers to the organizational pattern where engineers and teams treat raw AI token consumption — the volume of text fed to and generated by AI models — as evidence of productivity and AI adoption. First surfaced in enterprise engineering analytics reports in early 2026, the term describes a management antipattern analogous to measuring developer output by lines of code: plausible on the surface, actively harmful in practice. In a Jellyfish Q1 2026 study of 7,548 engineers, teams with the largest AI token budgets achieved only 2x throughput despite spending 10x as many tokens compared to disciplined peers — meaning they paid ten times more for twice the output. Organizations embracing tokenmaxxing have burned through enterprise AI budgets at catastrophic rates. Uber exhausted its entire $3.4 billion annual AI budget in just four months. Meta created a public leaderboard ranking 85,000 employees by token consumption, crowning one developer a “Token Legend” after they burned 281 billion tokens in 30 days. The incentive structure is broken: when token consumption is rewarded, engineers optimize for token consumption rather than outcomes. The result is inflated AI spend, degraded code quality, and a productivity illusion that evaporates the moment you track downstream metrics. ...

A multi-agent coding workflow is a development setup where you orchestrate two or more AI coding tools simultaneously — each handling a different phase of your work — rather than relying on a single tool for everything. In practice, this means Claude Code handles deep codebase reasoning and planning, GitHub Copilot manages real-time inline suggestions, and OpenAI Codex runs async batch tasks in the background. By Q1 2026, 70% of professional developers using AI tools run 2–4 tools simultaneously. Teams that adopted structured multi-agent workflows report wall-clock time cuts from 8 hours to 2 hours on typical feature work — a 4x speedup that’s hard to ignore. ...