Funding rounds, launches, founder stories and the startup ecosystem.
IBM’s new AI coding platform targets enterprise adoption by blending automated agents with mandatory human checkpoints, saving teams up to 70% time on selected tasks.
AWS Quick’s new desktop agent builds a living knowledge graph from your emails, files, and apps—proactively acting without prompts. But as AI autonomy grows, enterprises face fresh governance and oversight challenges.
Structured output benchmarks often overlook value accuracy in LLM-generated JSON. A new benchmark reveals surprising gaps even in top models like GPT-5 and Claude, with rankings shifting dramatically across text, images, and audio.
Enterprises are paying for expensive GPUs they barely use, trapped in a cycle of over-provisioning and rising cloud costs. A new report reveals shocking underutilization rates and why the FOMO-driven procurement loop is making the problem worse.
By embedding agents directly inside Spark pipelines, Definity’s platform detects and prevents failures in real time, reducing troubleshooting effort by 70% and enabling AI workloads to run reliably. Discover how.
Tindie’s ownership transition has left creators and buyers frustrated by prolonged downtime. The new team details urgent fixes and long-term plans to restore trust in the hardware marketplace.
Developers report that Anthropic’s managed agents repeatedly refuse to execute code, even after confirming it’s safe, because of a stubborn malware check that blocks all modifications. The issue is costing teams time and money with every failed session.
Enterprise teams now have a way to build custom reasoning agents without breaking the bank. A new method called RLSD cuts compute costs while improving model performance, making advanced AI accessible to more businesses.
A small U.S. AI startup just released a lightweight, open-source model that lets developers run coding agents locally without costly cloud fees. Here’s why it matters for privacy and performance.
Mistral AI has unveiled Workflows, a production-grade orchestration engine that bridges the gap between AI prototypes and mission-critical business systems. The launch underscores a growing consensus in the industry: the toughest challenge isn’t building AI models, but reliably deploying them at scale.
A developer spent four months hacking through 400+ council websites to extract 2.6 million UK planning decisions. The fragmented portals revealed surprising approval rates and a worsening trend in decision delays.
A decade-long exclusive deal between Microsoft and OpenAI has been restructured, ending revenue-sharing obligations and allowing OpenAI to serve models on any cloud. The move addresses enterprise demand for flexibility while reshaping the AI infrastructure landscape.
Xiaomi has launched two open-source AI models designed to outperform proprietary rivals on task-completion efficiency, slashing token usage by up to 60%. With a 310B-parameter architecture and a 1M-token context window, they’re reshaping enterprise AI economics for developers worldwide.
Traditional integration methods strain under today's complex, real-time supply chains. Automation-led iPaaS is emerging as the solution, but adoption requires careful planning to balance speed, cost, and disruption.
A groundbreaking AI system from SII-GAIR researchers autonomously refines training data, neural architectures, and learning algorithms—achieving results that surpass human-designed baselines by up to 18% in benchmark tests with minimal oversight.
Fine-tuning retrieval models for compositional sensitivity may seem like a way to improve precision, but new research reveals it can quietly slash retrieval accuracy by up to 40%. Here’s why this tradeoff happens and what teams can do instead.
A newly released open-source AI agent achieved a 65.2% score on TerminalBench, surpassing Google's 47.8% and the proprietary Junie CLI's 64.3%, with no reported cheating mechanisms.
A meticulously curated website annotates the iconic Unix Magic poster, offering insightful commentary and historical context
A new browser-based strategy game merges Checkers and Go mechanics into a turn-based duel. Built with real-time play and AI support, it invites players to test tactics and compete globally.
AI-driven synthetic audiences now simulate human behavior for surveys in minutes at a fraction of traditional costs. But do they replace real insights—or redefine them?
Traditional AI memory systems store everything permanently, leading to bloated models and higher costs. A new approach uses biological forgetting curves to prioritize only the most relevant data, slashing token waste and doubling recall accuracy.
AI models can appear flawless while delivering consistently wrong answers. Discover the hidden infrastructure failures eroding enterprise AI reliability and the monitoring gaps most teams overlook.
Multiple iPhone users have noticed the Headspace meditation app appearing on their devices without manual installation. The issue persists despite disabled automatic downloads and the latest iOS updates, raising concerns about unauthorized installations.
Generative AI’s unpredictability breaks traditional testing methods. Learn how engineers can build enterprise-grade AI by adopting a layered evaluation stack that catches drift, retries, and refusals before they reach users.
A new open-source tool replaces complex databases with a lightweight, Git-backed wiki that lets AI agents share and grow knowledge across sessions without losing context.
VT Code introduces a lightweight Rust-powered terminal interface for AI-assisted coding that supports both proprietary and open-source models. Discover how its semantic search and local inference features redefine development workflows.
A pair of seemingly manageable CVEs—when chained—enabled unauthorized root access across 13,000 Palo Alto devices. Security teams missed the kill chain because scoring systems evaluated vulnerabilities in isolation, not as part of a sequence.
Discover Lightwhale 3, a free, immutable Linux OS designed to streamline Docker container hosting with minimal setup and maintenance. Ideal for beginners and resource-conscious users alike.
Anthropic’s Claude 4.7 update appears to bypass built-in workflow enforcement tools, leaving developers scrambling to ensure critical steps aren’t skipped. Reports highlight persistent issues with stop hooks that were designed to enforce testing and validation rules.
The new DeepSeek-V4 model delivers near state-of-the-art AI performance at a fraction of the cost of premium alternatives like GPT-5.5, reshaping industry pricing and accessibility expectations overnight.
Apple CEO Tim Cook’s retirement in September marks the end of an era, leaving John Ternus to navigate a tech landscape reshaped by regulatory scrutiny, shifting app store policies, and evolving consumer demands.
Most enterprises pilot AI agents but rarely deploy them due to security risks. Cisco’s leadership reveals how trust architecture and zero-trust controls can bridge the gap between experimentation and production at scale.
Praveen Neppalli Naga, Uber's CTO, will join StrictlyVC SF on April 30 to share insights on managing technology at scale amid AI's rapid integration. The event marks a key highlight in TechCrunch's 2024 lineup.
A single HTML file uses Andrej Karpathy’s lecture transcript to visualize the inner mechanics of large language models. No external dependencies, just instant clarity on transformer architecture.
Anthropic reveals changes to Claude's harnesses and operating instructions caused degradation, affecting AI performance and user trust
A new open-source macOS application simplifies organizing and scaling Markdown-based knowledge bases. Discover how Tolaria blends offline-first design with Git integration for writers and researchers.
OpenAI’s latest large language model, GPT-5.5, introduces autonomous task handling and improved coding performance, outperforming rivals on key benchmarks while maintaining efficiency. Enterprise and Pro users gain early access.
