How to Work and Compound with AI
Eugene Yan outlines five principles for building and scaling AI systems, focusing on context as infrastructure and verification for autonomy.
Eugene Yan·3 May 2026
Use this view to inspect the underlying evidence corpus. For ranked developments, decision posture and interpretation, use Signals.
Raw feed or Signals?
Raw feed is chronological evidence. Signals ranks and interprets material change.
Eugene Yan outlines five principles for building and scaling AI systems, focusing on context as infrastructure and verification for autonomy.
Report discusses local AI coding agents as an alternative to usage-based LLM pricing, avoiding token limits and promoting 'vibe coding'.
Forrester argues that CIOs must manage the systematic risk of AI-generated software amid concerns over “systematic failure at scale.”
ABN AMRO details its enterprise AI implementation across six domains, including lending, financial crime, and advisory assistants.
Lloyds Banking Group launched Envoy, an internal platform built on Google Cloud to build, govern, and reuse enterprise AI agents.
Mozilla criticizes Google for integrating a Prompt API directly into Chrome, expressing concerns that it will reduce the openness of the web.
SAP user groups express concern over new API policy, fearing it will hinder adoption of AI and other innovations integrated with SAP systems.
Anthropic reportedly generates more revenue than OpenAI from its LLM offerings by focusing on enterprise customers with higher-value use cases.
UK AI Security Institute reports OpenAI's GPT-5.5 successfully executed end-to-end multi-step cyber-attack simulations during evaluations.
Citi announces the next phase of its AI strategy, focusing on the development and deployment of AI agents across its banking operations.
Cursor launches beta security review agents, including an always-on reviewer and vulnerability scanner, for Teams and Enterprise plans.
Amazon's custom AI chip business, including Trainium, has grown to an estimated $20 billion, signifying increased internal and external adoption.
OpenAI detailed the root cause and mitigation for 'goblin' outputs in GPT-5, attributing personality-driven quirks to specific training data.
Researchers demonstrated that LLMs can be easily poisoned through a $12 domain registration and one Wikipedia edit to alter factual recall.
Hugging Face details the training methodology and architectural choices behind IBM's Granite 4.1 series of LLMs, focusing on pre-training data.
OpenAI announces 'Stargate' initiative, a massive compute infrastructure project to support AGI development and meet future AI demand.
AWS engineers internally advise against AI shortcuts and advocate for human review, contrasting with public hype.
Block redesigned its data architecture using a data mesh, a governed semantic layer, and MCP servers to enable automated AI analytics.
OpenAI published a five-part action plan for cybersecurity in the 'Intelligence Age,' emphasizing AI-powered defense and critical system protection.
Cursor released an SDK allowing developers to programmatically build agents using Cursor's native runtime, harness, and models.
DeepInfra is now available as an inference provider on Hugging Face, enabling easier deployment and scaling of open-source models.
The FCA announced the second cohort for its AI Live Testing initiative, including Barclays, Lloyds (Scottish Widows), and UBS.
FCA's Jessica Rusu highlighted agentic commerce and Open Finance as key innovation drivers, announcing an expansion of their AI Lab.
At AI Dev 26 x SF, software developers discussed the impact of AI on their roles and the future of software development.
OpenAI models are now available in limited preview on AWS Bedrock, expanding distribution beyond Microsoft Azure and direct API access.
A new chatbot, Talkie, is trained only on data up to 1930 to explore how AI processes historical information and 'thinks.'
IBM's AI coding assistant, "Bob," is now generally available after internal testing with 80,000 users, including mainframe development.
Amazon introduced a new 'Copilot' AI assistant designed to integrate across various enterprise applications, emphasizing always-on context.
NVIDIA introduces Nemotron 3 Nano Omni, a multimodal LLM for long-context document, audio, and video agents.
Westpac CEO announces plans to deploy 35,000 generative AI Copilots across the bank's workforce to drive productivity.