Skip to main content

Things I Read

TitleSourceDateCommentary
Code Execution with MCP: Building More Efficient AI AgentsAnthropic Engineering2026-01-07Anthropic's case for having agents write code that calls MCP tools rather than invoking them directly — a real token-efficiency win as tool counts grow.
Orchestrate Teams of Claude Code SessionsClaude Docs2026-04-08Running several Claude Code sessions as a coordinated team. Handy for parallelizing big refactors and multi-part features.
Introducing Dynamic Workflows: Durable Execution That Follows the TenantCloudflare Blog2026-05-01Durable execution that migrates with the tenant across regions — the kind of primitive long-running, stateful agent workflows actually need.
The AI Engineering Stack We Built InternallyCloudflare Blog2026-04-20Cloudflare dogfooding its own platform for internal AI tooling. A grounded look at what a real-world AI engineering stack involves.
AI's Trillion-Dollar Opportunity: Context GraphsFoundation Capital2025-12-22Foundation Capital's argument that context graphs — not just bigger models — are the next big AI layer. Lines up with where agent memory seems headed.
MCP Apps: Extending Servers with Interactive User InterfacesModel Context Protocol2025-11-21The spec for MCP servers that render real interactive UIs instead of plain text. Opens the door to much richer agent-app experiences.
Secure MCP with Auth0Auth0Auth0's playbook for adding authentication and authorization to MCP servers. Directly useful for the auth work I'm doing at MCPCentral.
How Are Agent Identities Created?Microsoft Entra2026-04-08Microsoft Entra's model for provisioning identities to agents. Worth watching, since few platforms handle agent identity well yet.
Secure by Design with AI for Cyber ResilienceIBM2025-11-12IBM's case for designing security into AI systems from the start rather than bolting it on later. A solid framing for agentic-era threats.
The AI Agent Blueprint: Scaling AI AgentsFin2025-09-11Fin's field-tested playbook for taking customer-service agents from demo to production scale.
Microsoft Agent 365 SDK and CLIMicrosoft Learn2026-04-08Microsoft's developer toolkit for building on Agent 365 — a clear signal of how seriously the enterprise stack is chasing agents.
Claude Code Auto Mode: A Safer Way to Skip PermissionsAnthropic Engineering2026-03-26Anthropic's new permission classifier eliminates approval fatigue while catching dangerous actions — a smart solution to the safety vs. usability tradeoff in agentic coding.
Project Glasswing: Securing Critical Software for the AI EraAnthropic2026-04-08Anthropic's initiative to harden the world's most critical software using Mythos Preview, in partnership with major tech companies. The intersection of AI capability and security is exactly where the action is.
From Hierarchy to Intelligence — Jack DorseyBlock2026-03-31Block's vision of restructuring the entire company around AI intelligence rather than hierarchical management. Bold, practical, and probably the most interesting org design experiment happening right now.
Sandboxing AI Agents, 100x FasterCloudflare Blog2026-03-24Cloudflare's Dynamic Workers bring millisecond AI agent sandboxing using V8 isolates — no containers needed. This changes the economics of safe agent execution.
Time Horizon 1.1METR2026-01-29METR's updated time horizon evaluation for frontier AI models shows capability doubling every ~7 months. One of the most rigorous benchmark efforts tracking actual autonomous task completion.
Effective Harnesses for Long-Running AgentsAnthropic Engineering2026-01-25The initializer + coding agent architecture pattern for multi-context-window tasks. Practical engineering guidance from the team that built Claude's extended agentic workflows.
Every Company Building Your AI Assistant Is Now an Ad CompanyJuno Labs2026-02-20Argues that all AI assistant companies are ad-funded and building toward always-on hardware. Local on-device inference as the only real alternative — a compelling and underrated take.
Why Model Context Protocol Is Suddenly on Every Executive AgendaCIO.com2026-02-24CIO-level analysis of why MCP moved from niche developer protocol to enterprise boardroom priority. The "USB-C of AI" analogy is apt — standardization unlocks the whole ecosystem.
How We Built Our Multi-Agent Research SystemAnthropic Engineering2025-07-18Engineering lessons from building Claude's Research feature — 90.2% improvement over single-agent. The orchestration patterns and evaluation methodology are directly applicable.
Code Mode: The Better Way to Use MCPCloudflare Blog2025-09-27Cloudflare's discovery that LLMs write better code to call MCP than invoking MCP tools directly — 81% token reduction. Reframes how you should think about building agentic pipelines.
Context EngineeringLangChain Blog2025-09-19LangChain's framework for the four strategies agents use to fill context windows: write, select, compress, isolate. This is the new prompt engineering.
Why Agent Infrastructure MattersLangChain Blog2025-09-27Why agents need dedicated infrastructure (durable execution, state, human-in-the-loop) and can't just run on serverless. The case for treating agent infra as a first-class concern.
Block\'s Playbook for Designing MCP ServersBlock Engineering Blog2025-06-25Hard-won lessons from building 60+ MCP servers at Block: design top-down from workflows (not API endpoints), use prompt caching aggressively. The most practical MCP design guide I've read.
Closing the Agentic AI Productionization Gap: Bloomberg Embraces MCPBloomberg LP2025-10-10Bloomberg's enterprise MCP adoption story — bridging the demo-to-production gap in a regulated financial environment. A useful case study for any org navigating AI governance.
Securing Agentic Commerce: Helping AI Agents Transact with Visa and MastercardCloudflare Blog2025-12-08Cloudflare partnering with Visa (Trusted Agent Protocol) and Mastercard (Agent Pay) to enable secure AI agent transactions. Agentic commerce is closer than most people think.
The Killer MCP Use Case: Closing the Agentic LoopPulseMCP2025-08-26The cleanest explanation I've seen of why MCP enables fully autonomous agentic loops — completing tasks without user back-and-forth. This is the actual value proposition.
MCP Server Top 10 Security RisksCloud Security Alliance2025-11-06Ranked taxonomy of MCP server risks: Prompt Injection, Confused Deputy, Tool Poisoning, Credential Exposure, and more. Essential reading before deploying any MCP server in production.
Enterprise-Ready MCPAaron Parecki2025-11-01What enterprises actually need from MCP: SSO, Cross-App Access protocol, and admin-managed OAuth. Aaron Parecki brings his OAuth expertise to bear on MCP's enterprise gaps.
From Hackathon to Revenue: How I Built DialerRemote MCP Blog2025-08-08Complete walkthrough of building, deploying, and monetizing a paid Remote MCP server using Cloudflare and Stripe. The "speedrun your own paid MCP server" framing is exactly right.
BrowseComp: A Benchmark for Browsing AgentsOpenAI2025-06-16OpenAI's challenging benchmark of 1,266 hard-to-find information retrieval tasks for browsing agents. The difficulty design (verifiable but obscure) is clever and the failure modes are informative.
The Retrieval Tradeoff: What 50 Experiments Taught Us About Context EngineeringZep Blog2025-12-09Zep's empirical research on the three-way tradeoff between recall, precision, and efficiency in agent context retrieval. 50 experiments is the kind of rigor this field needs.
Context Templates: Context Engineering Made SimpleZep Blog2025-12-17Declarative context retrieval templates that make agent context tuning accessible without complex code. A practical response to the complexity creep in context engineering.
Seizing the Agentic AI AdvantageMcKinsey2025-07-18McKinsey's "genAI paradox" — 80% of companies see no earnings impact despite massive investment. Their analysis of why agents can unlock transformation where chatbots failed is worth the read.
How Minimal Built a Multi-Agent Customer Support System with LangGraphLangChain Blog2025-09-19Three-agent architecture (Planner, Research, Tool-Calling) delivering 80%+ efficiency gains in e-commerce support. A concrete example of agentic systems delivering measurable ROI.
How Tech Workers Really Feel About Work Right NowLenny's Newsletter2025-06-19Survey of 8,200 tech workers: 50% report burnout, startup founders are happiest, and mid-career is the rough patch. The data on manager experience gaps stood out.
My AI Skeptic Friends Are All NutsFly.io Blog2025-07-18Fly.io's honest rebuttal to AI skeptics who are still benchmarking against ChatGPT copy-paste rather than modern agentic coding workflows. Refreshingly direct.
Building Effective AgentsAnthropic2025-02-11Anthropic's foundational guide: simple composable patterns beat complex frameworks; the crucial distinction between workflows and agents. Still the best starting point for anyone building agents.
The Enshittifinancial CrisisEd Zitron / Where's Your Ed At2025-12-31Ed Zitron's sprawling 19,000-word analysis of the AI bubble, data center economics, and VC existential risk. Whether you agree or not, it's essential counterpoint reading.
Gemma 4Ollama2026-04-04Google's open multimodal models hit Ollama with native reasoning and improved agentic workflows. The on-device optimization path is getting genuinely competitive.
The Elements of Operating-System StyleThe Art of Unix Programming2026-04-09ESR's classic on Unix design philosophy — multitasking, IPC, and programming style through the lens of OS design. Timeless, and increasingly relevant as agent OS concepts emerge.
Graphic Design Trends for 2026Creative Bloq2025-12-23Anti-AI crafting, tactile textures, and human imperfection as design signal for 2026. The reaction against AI-generated slickness is already shaping visual culture.
Steering Semantic Data Processing With DocWranglerarXiv2025-04-20Research on DocWrangler for semantic data processing.
OpenAI Economic BlueprintOpenAI Research2025-01-13Fascinating analysis of AI's economic impact. The section on middle market opportunities particularly resonates with current business trends.
WEF Future of Jobs Report 2025World Economic Forum2025-01-08Comprehensive analysis of global employment trends and the impact of technological advancement on the future workforce.
A Roadmap to Reproduce o1arXiv2024-12-18Technical deep-dive into model reproduction. Important insights for local model training and optimization.
DocETL: Agentic Query Rewriting and Evaluation for Complex Document ProcessingarXiv2024-10-16Paper on agentic systems for complex document processing and query rewriting.
Executable Code Actions Elicit Better LLM AgentsarXiv2024-02-01Research on improving LLM agent capabilities through executable code actions.
LangChain State of AI 2024 ReportLangChain Blog2024-01-25Comprehensive overview of where AI development is heading. The focus on local AI deployment and privacy-first approaches aligns with current industry shifts.