Inteligencia Artificial

Modelos, agentes, investigación y negocio de la IA.

Inteligencia ArtificialJuly 21, 2026 · 6 min

CuspAI brings together Nvidia, Meta and Samsung to create materials with AI

British startup CuspAI has launched the AI Materials Foundry, a coalition that includes giants like Nvidia, Meta and Samsung. The goal is to create a 'search engine' for materials that don't yet exist, combining computer simulations, machine learning and robotic experimentation. The project promises to revolutionize industries such as batteries, semiconductors and construction.

Inteligencia ArtificialJuly 21, 2026 · 5 min

Writer cuts token costs by 40% by optimizing the AI harness

Researchers at Writer published a paper demonstrating reductions of up to 61% in cost per successful task by optimizing the AI harness, without touching the base model. The key lies in combating 'tokenmaxxing' and redesigning agentic loops.

Inteligencia ArtificialJuly 21, 2026 · 5 min

MCP Protocol: The Standard Unifying AI with Your Tools

The Model Context Protocol (MCP), developed by Anthropic, is becoming the de facto standard for AI assistants to securely access external data and services. A new update reduces implementation friction, bringing code-free interoperability closer to businesses and developers.

Inteligencia ArtificialJuly 20, 2026 · 3 min

From Perfect Traces to Broken Products: The New Paradigm for Evaluating AI Agents

A conversation with an AI agent can score perfectly in an individual evaluation yet signal a broken product. Leaders from LangChain, Conviva, and CoreWeave explained at VB Transform 2026 that the industry is moving from isolated traces to contrastive analysis across user cohorts, along with cheaper, more specific judge models.

Inteligencia ArtificialJuly 20, 2026 · 4 min

AI Cyberattack: Hugging Face Suffers Autonomous Agent Attack

Hugging Face confirmed a cyberattack where an autonomous AI agent executed thousands of actions, exploited vulnerabilities, and stole credentials. This incident marks a milestone in the evolution of cyber threats, proving that AI-driven attacks are no longer theory but reality.

Inteligencia ArtificialJuly 20, 2026 · 5 min

Capital One launches VulnHunter: Open-source AI that finds security flaws before hackers do

Capital One has launched VulnHunter, an open-source AI tool that scans source code for exploitable vulnerabilities, traces attack paths, and proposes fixes. The move seeks to redeem the bank's security reputation after the 2019 breach that affected 106 million people.

Inteligencia ArtificialJuly 20, 2026 · 4 min

Kimi K3: The Chinese Model Shaking US AI Leadership

Moonshot AI's Kimi K3 has outperformed top US models in coding tests at significantly lower cost, with plans to be open-weight. This threatens US tech dominance and pressures sector prices and valuations downward.

Inteligencia ArtificialJuly 20, 2026 · 5 min

Data Center Protests Spread to 42 US States

Last Saturday, opposition to data centers in the United States escalated nationwide with coordinated protests in 42 states. Protesters denounce massive energy and water consumption, environmental impact, and lack of regulation. The movement threatens to slow the expansion of artificial intelligence infrastructure.

Inteligencia ArtificialJuly 20, 2026 · 4 min

Legacy Infrastructure: The Real Brake on AI Agents

At VB Transform 2026, infrastructure leaders from LinkedIn, Walmart, and Zendesk concluded that legacy infrastructure, not AI models, is the main obstacle to scaling AI agents. Each company faced distinct bottlenecks, but all pointed to the same root: systems built for human pace, not agent speed.

Inteligencia ArtificialJuly 20, 2026 · 4 min

Anthropic donates MCP to the Linux Foundation: strategy or altruism?

Anthropic has donated the Model Context Protocol (MCP) to the Linux Foundation, joining Google and OpenAI in standardizing AI protocols. This move, following the 'commoditize the complement' strategy, aims to position Anthropic at the center of the AI agent ecosystem.

Inteligencia ArtificialJuly 20, 2026 · 4 min

Kimi K3: The New Chinese Model That Sparks Panic in Global Markets

The launch of Moonshot AI's Kimi K3 model has shaken global markets, with widespread declines in tech and semiconductor stocks. Investors fear China can compete in AI at lower costs, questioning the sustainability of massive spending by US companies.

Inteligencia ArtificialJuly 20, 2026 · 5 min

AI Models Self-Censor Criticism of Repressive Governments

Major AI models like ChatGPT and Gemini refuse to criticize repressive governments when asked directly, according to an Oversight Board study. This reveals a geopolitical bias that could limit free expression and access to critical information in regions where it is most needed.