标签：#ai

Tu repo ya es el contexto de tus agentes, lo hayas diseñado a propósito o no Esa frase me costó entenderla. En este post te ahorro el camino. Era Octubre del 2025, trabajando en el monorepo de Skyward con agentes de IA todos los días. Y todos los días la misma rutina: le decía en el prompt al agente "no uses esto", "no hagas esto así", "reutiliza el componente que ya existe". Lo escribía. Lo repetía. El agente hacía exactamente lo que le dije que no hiciera. No era que no me escuchara. Era que leía el código y ahí veía otra cosa. El agente le cree al código, no a tu prompt Un agente sigue los patrones que ve en el repo, no los que tú le dices en el prompt. Y los subagentes son peores, porque arrancan sin el contexto de la conversación. Toda la pelea que tú diste arriba en el chat, para ellos no existió nunca. Entonces pasaba esto. Creaba un componente nuevo aunque ya había uno que resolvía exactamente el problema. No respetaba las normas de diseño ni usaba los design tokens. Seguía documentación obsoleta porque seguía ahí, viva, sin nada que la marcara como obsoleta. Mi primer instinto fue el de todos, meter más contexto en el prompt. Más reglas, más "por favor no hagas esto", más ejemplos pegados a mano. Funcionaba a medias, y para la siguiente tarea había que volver a agregarlo todo. Hasta que llegaba el siguiente subagente y empezaba de cero. En algún momento, cansado de repetirme, entendí lo obvio. El agente no me estaba desobedeciendo. Estaba leyendo el repo y haciendo caso a lo que el repo decía de sí mismo. Si conviven el componente bueno y tres versiones viejas, no tiene cómo saber cuál es el oficial. Si la doc dice una cosa y el código hace otra, le va a creer al que esté más a mano. Está haciendo justo lo que le pedí. El mismo repo es el contexto que usan los agentes. Si está mal estructurado, las respuestas no van a ser buenas. Punto. No hay prompt que arregle un repo que se contradice a sí mismo. Screaming Architecture me llevó hasta media cancha Lo prim

Architecture Overview Clioloop's Agentic Fusion is not just "run the same prompt 5 times and pick the best." It's a structured pipeline where different models play different roles, with strict security boundaries between them. The Pipeline Step 1: Planning Phase When you run /fusion , up to 5 planner models are dispatched in parallel. Each planner: Receives your original prompt and context Has read-only tool access — they can search the web, read files, but never modify anything Proposes an approach (not an answer — an approach) The planners might suggest different strategies: Planner A: "Search the web for similar problems, then write a script" Planner B: "Read the existing codebase first, then modify in place" Planner C: "Break it into subtasks and use the Kanban system" Step 2: Execution Phase Your main model takes the planners' proposals and does the actual work: Full tool access (file editing, shell, web, browser, image gen, etc.) Fully visible — you watch every file edit, every command, every web search in real time Not a black box — you can intervene at any time This is the key difference from "ensemble" approaches: the main model does real work with real tools, not just text generation. Step 3: Review Phase Up to 5 reviewer models critique the draft: Read-only access — they can see what the main model produced, including generated images They check for errors, suggest improvements, flag problems Each reviewer works independently Step 4: Verdict Loop The draft is revised based on reviewer feedback: If reviewers find issues, the main model gets the feedback and revises The loop continues until reviewers approve You get the final, reviewed answer Step 5: Fusion Everything combines into one answer that has already passed independent review. Security Model The safety comes from schema-level restrictions : Role Can Read Can Write Can Execute Planners ✅ Files, web, images ❌ Nothing ❌ Nothing Main Model ✅ Everything ✅ Files ✅ Commands Reviewers ✅ Draft, files, image

Build-in-public. This is the real architecture behind running ~50 local AI agents on 6GB of VRAM — one GPU lock, an eviction watchdog, a resource governor, and a model router. Originally posted on my blog. The question I get most often is some version of "there's no way you run that many agents on a 6GB laptop GPU." The honest answer: not the way you're picturing it. I don't run 50 models at once. I run one model at a time, very deliberately — and most of the engineering is about scheduling, not inference. Here's the actual architecture. The hard constraint: 6GB of VRAM A single consumer GPU with 6GB of VRAM holds roughly one 7B-parameter model at a usable quantization. Two at once? It thrashes — the GPU starts swapping, latency explodes, and eventually a driver out-of-memory can take the whole machine down. I've had the desktop freeze from exactly that. So the first design rule wrote itself: only one heavy model is allowed on the GPU at any moment. That sounds limiting. It isn't — because almost nothing I run is latency-sensitive. A blog post that publishes at 7am doesn't care if it was generated at 6:52 or 6:58. Once you accept that your AI workforce is a batch system, not a chat window, the whole problem changes shape. A lock, not a crowd Every agent that needs the GPU has to take a lock first. It's a simple file-based queue with: FIFO ordering PID-based ownership Stale-lock detection, so a crashed job can't wedge the line forever If an agent can't get the lock within its timeout, it skips gracefully and tries again on its next scheduled run instead of piling up. So at 50 agents, what's really happening is: dozens of cron-scheduled Python workers wake up throughout the day, and the ones that need the model form an orderly line for it. The fleet is huge; the GPU contention is always exactly one. That's the trick. It's less "50 models" and more "50 employees sharing one very busy workstation, politely." Eviction and a VRAM watchdog Even with the lock, idle models l

On June 12, 2026, enterprise developers using the Codex API started seeing an unfamiliar response header: X-Model-Version: kindle-alpha . It appeared on a subset of requests for roughly 18 hours, then vanished. That's the release candidate for GPT-5.6 — OpenAI's next flagship model — leaking through the staging layer. OpenAI's Chief Scientist publicly called the upcoming release "a meaningful leap" the following day. By OpenAI's historically understated communications standards, that's loud. This post covers what the backend traces, developer reports, and Polymarket odds (currently ~80% for a pre-June-30 launch) actually tell you about the model — and what to do before it drops. How the Leak Surfaced Three separate sources converged in the 72 hours after the June 12 header incident. First, developers with ChatGPT Pro OAuth access reported hitting context windows significantly beyond GPT-5.5's supported limit. At least four documented cases logged successful 1.5M-token completions before the backend silently downgraded them to the production model. Second, the Codex enterprise API logs — accessible with full response header exposure enabled — confirmed the kindle-alpha codename across US-east-1 and us-west-2 endpoints. Third, the Polymarket market for "GPT-5.6 public release before July 1, 2026" moved from 61% to 80%+ within 48 hours of the header reports circulating on developer forums. None of this is from OpenAI's press office. No model card, no official benchmark numbers, no pricing. The specifics below are high-confidence inference from multiple corroborating signals — not official spec. Treat it accordingly when making production decisions. The Architecture Shift: Agentic-First, Not Just Smarter GPT-5.5 was trained as a reasoning model with agent capabilities added on top. GPT-5.6 is reportedly designed in the opposite order. The primary optimization target during training was not MMLU or GPQA benchmark scores — it was token efficiency on long-horizon agentic t

The Problem Most AI assistants give you one model's answer. If it's wrong, you catch it or you don't. If you use a cheap model, quality drops. If you use a frontier model, you pay frontier prices for everything — even a simple file rename. What is Agentic Fusion? When you run /fusion , a panel of models collaborates on your task: Planners (up to 5): Read-only models that research and propose routes in parallel. They figure out the best approach but can't touch your files or run commands. Main model : Your chosen model does the actual work — full tool access, fully visible. You watch every step. Not a black box. Reviewers (up to 5): Read-only models that critique the draft. They can see images the main model generated. They check for errors, suggest fixes, flag issues. Verdict loop : The draft is revised until reviewers approve. The answer you get has already passed independent review. Fusion : Everything combines into one reviewed, approved answer. The quality comes from synthesis — not from running the same job 5 times. Cheap open models combine into something that rivals a frontier model at a fraction of the cost. Safety by Construction Planners and reviewers are read-only at the schema level. They can research and critique, but they can never touch your files or execute commands. Only your main model has tool access, and you watch it work live. Beyond Fusion Clioloop is also: Self-improving : Keeps MEMORY.md and USER.md , updated automatically Autonomous : Set a standing goal with /goal and it loops until done Everywhere : Terminal, desktop app, web dashboard, Telegram, Slack, Discord, WhatsApp Multi-agent Kanban : Break big work into tasks with worker agents Tools : File editing, shell, web search, browser, image/video gen, TTS, MCP Scheduled jobs : Run on cron for automated workflows Open-source : Self-host everything, own your data The Omni Loop Portal One OAuth login gives you access to 300+ models. No API keys. An OpenAI-compatible proxy means any tool works

Stop Asking 'Is GAI Here' — Ask 'At What Layer' The GAI debate has a structural problem. Someone says "passing this benchmark means GAI." A model passes it. Then they say "that benchmark wasn't hard enough." The goalpost moves. Someone says "passing the Turing test means GAI." Models pass it. Then they say "the Turing test is too easy." The goalpost moves again. Someone says "inventing new mathematics means GAI." Models do it. Then they say "that's just pattern matching in disguise." Goalpost moves. This isn't bad faith. It's a missing layer definition. We never agreed on what "general" means. Without that, every achievement gets reclassified as "not really general." I've been working on a framework that might fix this. It started as a capability map. Then I realized: this isn't just a map. It's a GAI maturity model. The Five Layers Layer Name Definition L0 Embodied Perceive and operate in the physical world L1 Application Complete single-domain tasks using tools L2 Engineering Build and maintain systems L3 Meta-Domain Abstract and transfer between unrelated domains L4 Meta-Cognition Perceive and control your own thinking process The rule: layers cannot be skipped. It's a maturity sequence, not a checklist. This immediately explains the goalpost problem: some people define GAI as L1. Others define it as L4. They're using different layers for the same word. What About Models Without Bodies? L0 requires embodiment. Text-only models don't have bodies. The cleanest answer: LLMs have no L0. They start at L1 — cognition without embodiment. This isn't a defect. It's an architectural difference. Humans build up from L0 (a baby senses the world before understanding it). LLMs start at L1 (they understand the world directly, skipping physical experience). The result: humans can "feel" when something is wrong — that's L0 feeding signals up to L4. LLMs don't have this channel. The framework forced me to face something uncomfortable: human intelligence cannot exist without a body

Building a high-performance Polymarket trading bot requires mastering its unique hybrid architecture:...

My agents were confidently wrong about the world, and I couldn't tell when. That's the part that got to me — not the wrongness, the confidence. I run my one-person company as a fleet of about twenty AI agents — a content writer, a finance one, a researcher, a security officer, a handful more. They're good at the work I built them for. But every one of them shares a flaw I'd been papering over: when a task needs a fact about the world — how a tax threshold works, what a marketing framework actually says, how a platform bills — the model reaches into its training data and answers in the exact same self-assured tone whether it knows or is improvising. There is no tell. The guess and the fact wear the same face. So this month I built the thing that was missing: a cited, fact-checked knowledgebase the agents have to read before they work, with a gate that keeps me from poisoning my own source of truth. Here's how it's built, the one rule that turned out to matter most, and the honest state of it — which is that I finished it days ago and have no idea yet whether it changes the work. The job I was actually hiring this to do Strip away my setup and the problem is one any solo operator using AI already has. You ask the model for something that depends on a real fact. It answers fluently. You either know enough to catch the error or you don't — and the whole reason you're asking is usually that you don't. The job I needed done wasn't "make my agents smarter." It was narrower and more honest: stop my AI from making things up in the one register where I can't catch it, and let me know which claims I can actually trust. The competition for that job, in my shop, was "just let the model wing it and hope." That had already cost me. A marketing analysis once understated a channel's numbers because an agent trusted a stale figure instead of pulling the live one. Small, recoverable — but it's the recoverable ones you see. The ones you don't see are the ones that scare you. What I bui

Recently, while working on an in-progress open-source framework called Projector, I ran into a (not particularly novel) issue: one of it's internal packages ( core ) had grown during this period, and was not nearly as flyweight as it needed to be in the browser. The result was 10-20kbs of unnecessary machinery getting pulled in. I noticed this while running examples. I was consistently hitting a wall in bundle sizes that was surprisingly difficult to get past, even for someone as stubborn and relentless as I am. Naturally, I turned to Claude and ChatGPT to help me with this, and ended up using ChatGPT 5.5 with Codex as I find that, with the "precise" output mode, it tends to be a little more honest than Opus 4.8 these days. I shared exported HAR network logs with it, having it go through the chunks to confirm where the bulk was; consistently, it confirmed that the issue was around an entangling of authoring/resolution code with runtime code in core that was pulling in too much to the browser. The technical details here aren't really important, but I'm using them to illustrate a larger point. We then iterated through a lot of different solutions—I setup a "goal" in codex with benchmarks to hit, and gave it a bunch of constraints, context, and tooling. Finally, after about 2-3 hours of looping against that goal, it completed. Looking through the git diff, I noticed something odd—it had duplicated the result of the resolved module, so it could skip the resolution machinery and thus drop it from the browser bundle (again, technical details not really relevant). It hit the rough kb benchmarks, respected all constraints, utilized all context and skills available, and avoided importing the machinery that we both aligned on being the core problem. It provided an elegant, coherent, well-written api, implemented a surgical, well-tested, well-designed solution, and convincingly defended its work when I queried about the implementation. That sounds great, right? In fact, I thin

Taiwan's drone spending plans for defense could also boost business overseas.

Startup Baseten is reportedly close to finalizing a $1.5 billion round at a $13 billion as the “inference gold rush" marches on.

AI Observability for Lovable Apps: Monitor Prompts, Traces, and Evaluations with Currai Building AI applications has never been easier. Tools like Lovable allow developers and founders to create AI-powered products in minutes. Whether you're building a chatbot, AI assistant, recommendation engine, AI agent, or prediction app, generating the application is often the easy part. The real challenge starts after launch. How do you know what prompts are being sent to the model? How do you debug unexpected AI responses? How do you compare prompt variations and determine which performs better? How do you evaluate output quality over time? How do you track token usage and costs? This is exactly why we built Currai . What is Currai? Currai is an AI observability platform that helps teams understand, test, and improve AI applications in production. It provides: Prompt tracing AI request monitoring Session tracking Prompt versioning A/B testing LLM evaluations Cost and token analytics OpenTelemetry support Instead of guessing why your AI application produced a particular response, Currai lets you inspect the entire execution flow. The Problem With AI Applications Traditional monitoring tools were built for APIs, databases, and backend services. AI applications introduce a completely different set of challenges: Prompt changes can significantly impact output quality Model updates can affect behavior Hallucinations are difficult to track User conversations are hard to debug Prompt experiments are often unmanaged Quality evaluation is usually manual When something goes wrong, application logs alone don't provide enough visibility. You need observability designed specifically for AI systems. Trace Every AI Request Currai captures every prompt, model response, latency metric, token usage, and cost. You can inspect: System prompts User prompts Model outputs Execution traces Tool calls Metadata This makes debugging AI applications dramatically easier. Run Prompt A/B Tests Prompt engin

The Snapchat maker is spinning off yet another internal unit. Dotmo will be comprised of current Snap staff who are leaving the social media company to focus on AI video development.

Valve has some good news and bad news about Steam Controllers. The good news: if you make a reservation for a Steam Controller, the company will now show you one of three estimates of when you'll be able to actually order your gamepad: by September 2026, by December 2026, or sometime in 2027. The bad […]

OpenAI is bulking up before its IPO, landing Transformer co-inventor Noam Shazeer from Google DeepMind and former Trump AI policy official Dean Ball in the same week.

About 47% of singles look negatively at the use of AI in dating -- but, many dating app users are open to AI helping with profile punch-ups and conversation starters.

AI agents get better the more they run. Every conversation turn, every task completed, every prompt refined adds to a growing body of context that shapes the next output. The compounding effect is real: an agent with 100 turns of memory and a versioned prompt history behaves meaningfully differently from one starting cold. This post walks through using a structured, versioned, API-accessible store as the memory layer for AI agents, with TypeScript examples. Agent messages, system prompts, findings, and instructions are all stored as structured, versioned, API-accessible Objects. Each new turn adds to the record. Each prompt edit is tracked. What Agent Memory Actually Needs The compounding loop only works if the memory layer has the right properties. Most agent frameworks handle working memory well. The gap is episodic and semantic memory: what the agent learned, did, and produced across sessions. Researchers at Elastic recently published a breakdown of agent memory tiers : working memory (in-context), episodic memory (past interactions), semantic memory (knowledge), and procedural memory (learned behaviors). Good persistent agent memory needs four properties: Structured : queryable by type, status, date, or custom field, not just full-text search Versioned : you need to know what the agent wrote at each point in time, not just the latest state API-accessible : any model, any framework, any language should be able to read and write it Human-reviewable : agents make mistakes; a human needs to inspect and correct outputs without touching a database Objects as Agent Outputs When an agent produces output, storing it as a structured Object gives you a queryable record with typed fields, a draft/published workflow so a human can review before promoting to production, a full audit trail of every change, REST API access from any runtime, and a dashboard UI where non-technical team members can inspect, edit, or approve agent outputs. Here's a simple research agent that stores

I was building a feature that needed to say something useful about a stock — not just print its P/E, but actually read the situation: is this cheap or expensive, what's the bull case, is the insider buying real or routine. I went looking for an API. Every finance API I found sold me raw data . Alpha Vantage, Twelve Data, Yahoo Finance, FMP — they'll hand you fundamentals, prices, filings, all of it. Great. Now I get to write the part that turns 40 metrics into "this looks expensive but the moat is widening." That's the part that's actually hard, and the part I didn't want to own forever. So I'd be wiring three data providers, normalizing their conflicting field names, writing and tuning the LLM prompts, handling the rate limits and the caching, and then maintaining all of it as the upstreams change. For a feature, not a product. What I wanted instead A single endpoint. Ticker in, analysis out — already synthesized, already structured. That's what I ended up building for myself and then put on RapidAPI: Agent Toolbelt — AI Stock Research API . It pulls live fundamentals from Polygon, Finnhub, and Financial Modeling Prep, then returns a Motley-Fool-style read as typed JSON. The numbers are in there too, but the point is the verdict and the reasoning. Here's a real stock-thesis response: { "verdict" : "bullish" , "oneLiner" : "Nvidia owns the essential infrastructure for the AI revolution with a defensible software moat." , "keyStrengths" : [ "~80%+ data center GPU market share" , "CUDA moat creates switching costs" , "42 buy / 5 hold / 1 sell analyst consensus" ], "keyRisks" : [ "36.9x P/E leaves no margin for error" , "Competition from AMD and custom silicon" ], "insiderRead" : "Two executives bought ~47k shares each — meaningful open-market purchases, not routine grants." , "dataSnapshot" : { "currentPrice" : 180.4 , "peRatio" : 36.9 , "marketCapBillions" : 4452.2 } } That's one HTTP call. No data-provider accounts, no prompt engineering, no normalization layer. The

I’ve been a software engineer for more than 12 years. And like many developers, I’ve been watching AI improve at an incredible speed. Every new model seems smarter than the one before it. Tasks that used to take hours can now be done in minutes. Problems that required deep research can often be solved with a simple prompt. A few years ago, we used to say: Think of AI as a junior developer. That made sense at the time. But today, I don’t think that’s true anymore. AI still makes mistakes. Sometimes very obvious ones. But it also comes up with solutions that surprise me. Sometimes it finds an approach I wouldn’t have thought of immediately. Sometimes it helps me solve a problem much faster than I could on my own. And honestly, that’s both exciting and a little scary. But the biggest thing AI changed wasn’t how I write software. It changed how I think about my work. For most of my career, I thought I loved writing code. I spent years doing it. At work, on side projects, and whenever I had free time. Then AI became part of my daily workflow. In the last month, I’ve built more projects than I normally would in an entire year. Ideas that had been sitting in my notes for years suddenly became possible. And that’s when I realized something important: I don’t actually love writing code. I love building things. I love taking an idea and turning it into something real. I love creating products, solving problems, and seeing something that only existed in my head become something people can use. Code was simply the tool I used to do that. And now AI is another tool. That’s why I don’t hate it. In many ways, AI has helped me build more than ever before. It helped me revisit old ideas that I never had time to work on. It helped me experiment faster. It even encouraged me to explore areas outside software development, like animation and content creation. And this isn’t just happening to programmers. AI is changing design. It’s changing writing. It’s changing marketing. It’s changin

AWS is in talks to sell its chips to other data centers. CEO Andy Jassy has said this represents a $50 billion opportunity for the company.