OpenRouter launches @openrouter/agent, a model-agnostic TypeScript SDK that simplifies building agentic loops with tool execution, multi-turn loops, stop conditions, streaming, cost tracking, and tool approval.

OpenRouter introduces two skills for building agent harnesses: create-agent-tui for terminal UIs and create-headless-agent for headless agents. Both generate complete TypeScript projects using the Agent SDK, offering customizable features and integration with any OpenRouter model. The skills provide fine-grained control, minimal deployment, and educational value. Key features include interactive checklists, customizable UI, session persistence, safe retry, and structured output.

Anthropic's new tokenizer for Claude Opus 4.7 increases token consumption, resulting in 12-27% cost increases for prompts over 2K tokens, but short prompts get cheaper due to shorter completions. OpenRouter analyzed millions of requests to provide real-world data.

OpenRouter's April releases include video generation, workspaces for multi-project isolation, a TypeScript SDK that turns any model into an agent, reranker models, model fusion, prompt history, benchmarks, knowledge cutoff dates, and new frontier models like GPT-5.5 and DeepSeek V4 Pro.

OpenRouter introduces response caching that allows developers to cache identical API requests, returning responses in milliseconds with zero token billing. The cache sits in front of providers, hashing request details. It supports streaming and non-streaming, works across multiple endpoints, and includes features like TTL control and cache busting. Use cases include agent retries, test suites, and repeated prompts.

OpenRouter introduces two dedicated audio endpoints for text-to-speech and speech-to-text, offering faster and more cost-efficient models from providers like OpenAI, Google, and Mistral.

The Agent SDK now supports human-in-the-loop (HITL) tools, allowing agents to automatically handle routine calls and pause for human input when stakes are high, all controlled by a single hook.

Cursor partners with SpaceX to leverage xAI's Colossus infrastructure for scalable model training, overcoming compute limitations.

Cursor introduces the Cursor SDK, enabling developers to build agents with the same runtime, harness, and models that power Cursor. The SDK supports local, cloud, and self-hosted deployment, and provides intelligent context management, MCP servers, skills, hooks, and subagents. Available in public beta.

Cursor shares how they treat their AI coding agent harness as a software product, iterating through vision, hypotheses, experiments, and instrumentation. The post covers the evolution from static to dynamic context, two assessment methods (benchmarks and online A/B testing with keep rate and LLM-based satisfaction), a comprehensive system for tracking and fixing degradations, deep customization for different models (including handling model quirks), and the challenges of mid-chat model switching. It concludes with a vision for multi-agent software engineering.

Cursor introduces autoinstall, a system that uses earlier Composer models to automatically set up RL training environments, improving efficiency. The two-stage process involves goal setting and environment configuration, and has been successfully applied to real-world projects like Celo. Composer 2 shows significant improvement on Terminal-Bench.

Bugbot is switching from a $40 per seat per month subscription to usage-based billing for Teams and Individual plans. For existing customers, this change will start at your next billing renewal after June 8th, 2026. Average run costs $1.00-$1.50. New effort levels allow deeper reviews.

Cursor launches new tools for configuring cloud agent development environments, including multi-repo support, Dockerfile improvements, enhanced agent-led setup, and governance controls, enabling teams to run parallelized agents that handle tasks end-to-end.

Cursor uses a hybrid online-offline eval process to measure coding agent model quality. Its internal eval suite, CursorBench, based on real developer sessions, better reflects developer experience. Public benchmarks suffer from alignment, grading, and contamination issues, while CursorBench shows greater model separation and aligns with online metrics.

Cursor launches Composer 2, a frontier-level coding model with competitive pricing at $0.50/M input and $2.50/M output tokens. It achieves significant benchmark improvements across Terminal-Bench 2.0 and SWE-bench Multilingual, enabled by first-time continued pretraining and reinforcement learning. A technical report is also released.

Cursor releases its third major version, a unified agent workspace for building software, featuring multi-repo layout, seamless local-cloud agent handoff, and faster review workflows.

Cursor describes the evolution of AI-assisted coding from tab completion to synchronous agents to the current era of autonomous cloud agents, which are now handling a significant portion of development work.

PwC and Anthropic deepen alliance to deploy Claude across enterprise functions, including a new Office of the CFO, with plans to train 30,000 professionals and expand to hundreds of thousands of users, achieving up to 70% faster delivery times in areas like insurance underwriting and cybersecurity.

Cerebras partners with Armis to leverage Armis Centrix™ for Application Security and Cerebras' ultra-fast AI capabilities, enabling teams to identify and remediate vulnerabilities faster, reduce noise, and focus on critical risks throughout the software development lifecycle.

Perplexity's move from MCP to APIs and CLIs sparked a debate about protocol overhead. While MCP's token consumption and latency are real issues, faster inference hardware (e.g., Cerebras Wafer-Scale Engine) and secure execution environments (e.g., Monty interpreter) can mitigate these, benefiting both MCP and CLI approaches.

Lessons learned from building multi-agent workflows, covering the shift from single-agent ceiling to multi-agent architecture with five practical patterns.

This article describes the author's experience using Codex and Figma MCP to automatically replicate website designs into Figma. Through multi-agent orchestration, they overcame context limits, long run times, and other issues, achieving perfect replication of 5 pages in under 5 minutes.

Cerebras is scaling access to ultra-low-latency inference, turning speed into a broadly accessible platform. With its wafer-scale chip delivering up to 15x faster inference than GPUs, the company is expanding model support, cloud availability, and developer integrations. The ecosystem now covers major open models, agent frameworks, coding tools, and observability platforms, making fast inference a practical infrastructure layer for production AI applications.

Cerebras Inference powers Cognition's SWE-1.6 and SWE-grep agents, delivering up to ~5x faster coding performance than GPU, enabling real-time code generation and smoother developer experience.

Cerebras announces the private preview of Multi-LoRA (multi-adapter Low-Rank Adaptation) on Cerebras Inference, allowing teams to deploy multiple LoRA adapters with a single shared base model, enabling specialization for different domains, tasks, customers, and workflows without maintaining separate full models.

AI-generated UI suffers from predictable patterns like dashboard mimicry and card nesting. Faster generation speeds (1200 tok/s) and vision models now enable rapid iteration. Practical methods include using shadcn/ui with MCP, defining design tokens upfront, and small-change iteration.

In early 2026, the AI race shifted from model intelligence to inference speed. Major labs like Google, Anthropic, and OpenAI released faster models for coding. Fast inference accelerates model development and product iteration, making it a critical factor for AI progress and business revenue.

Mortgage document automation leverages intelligent document processing to transform document-heavy workflows into structured, machine-driven processes, improving efficiency and reducing errors. This article examines the complexities of mortgage document handling, the automation workflow (ingestion, classification, extraction, validation, human review, and system integration), challenges, and best practices for implementation with LlamaParse.

This article examines the shortcomings of standard OCR in KYC workflows, including its inability to handle real-world identity documents with security features, varied layouts, and non-Latin scripts. It introduces agentic OCR (e.g., LlamaParse) that uses layout-aware segmentation, model orchestration, and self-correction loops to achieve over 90-95% straight-through processing, and discusses implications across banking, insurance, and crypto industries.

This week's LlamaIndex newsletter highlights intelligent table extraction, LiteSearch for local document retrieval, improved Word document processing, and integrations with Gemini Live API, along with guides for legal discovery and community projects.