Curated talks and discussions on AI agents, developer tools, and platform engineering
No videos found with this tag.
MCP Tunnels lets your private MCP servers connect to Claude Managed Agents over an outbound-only connection — no inbound ports, no firewall rules, no public exposure required. Cloudflare Tunnel carries the pipe; you supply the TLS cert and keep full end-to-end control. Currently in research preview.
Google processes 3.2 quadrillion tokens/month (7x YoY). Gemini 3.5 Flash ships today — frontier speed at 1,500 tokens/sec. Anti-Gravity's 93 sub-agents built a working OS from scratch in 12 hours. Gemini Spark is a 24/7 personal agent on Google Cloud VMs. AI Mode in Search already at 1 billion monthly users.
Entry-level tech positions are down 67% since 2022. Junior roles down 40–50%. AI agents now fill the role juniors used to occupy — the job description is identical. Darren and Victor break down what skills actually survive this shift and why adaptability has always been the only durable career asset.
Simon tours his 2026 Notion Life OS rebuild for the agentic era — a new Agent & Skills Hub stores personal agent instructions, sub-agent prompts, knowledge bases, and SOP skills that plug directly into Notion's personal agent. AI meeting notes get a dedicated database with skill-based processing. Design thesis: your second brain should route context to AI tools, not just store notes.
Thariq Shihipar posted nine words — "HTML is the new markdown" — and broke thousands of AI workflows overnight. HTML unlocks SVG, interactive widgets, inline annotations, and severity color-coding that flat markdown structurally cannot express. The rule: markdown for repos, HTML for readers.
OpenClaw crossed into serious runtime territory in April. The model is now a swappable component — the real unlock is building durable workflows that survive model churn, pricing changes, and provider policy shifts. OpenBrain recipes for user-owned agent memory are now open-source.
Cut cost, manage context, boost intelligence. Live demos of prompt caching, tool search, programmatic tool calling, compaction, and the advisor strategy — patterns that together reduce agent costs by 10x without quality loss.
At Datadog, 90% of engineers adopted AI coding tools in four months with Claude Code driving two-thirds of usage. Sesh Nalla shares how they built Temper — a constrained framework for secure, reusable tools that compound in value across sessions and teams.
When Opus 4.5 landed, v0 was ready on day one — not by luck, but by design. Guillermo Rauch explains how Vercel architects for model step-changes, why eval-driven development is the key, and what becoming an "agent-pilled company" actually looks like.
Three AI-native teams compare architectures: Cognition's hierarchical agents with context isolation, Gamma's MCP-integrated design pipeline, and Harvey's research+draft+review chain for legal. Where MCP shines, where it struggles, and the lessons each team would redo.
Boris Cherny, creator of Claude Code, livestreams his everyday workflows unfiltered — context discipline, parallel subagents, autonomous runs with verification, and the CLAUDE.md philosophy that shapes his sessions.
80x annualized growth in Q1 2026, the SpaceX compute deal, the one-person billion-dollar company on track by year-end, multi-agent as the next S-curve, AMD's law applied to AI, and what "hold light and shade" means for a company building on the frontier.
Most of shipping an agent isn't the agent itself — it's the infrastructure: state, long-running sessions, orchestration, guardrails. Claude Managed Agents handles that layer so you focus on the agent logic.
GitHub CPO Mario Rodriguez and Anthropic's Brad Abrams on shipping Claude to millions of developers across four surfaces — and how prompt caching, eval harnesses, and the advisor strategy keep quality up and costs down at scale.
A twenty-minute summary of what shipped in Claude Code: Agent SDK for multi-agent orchestration, background subagents, LSP integration for real-time code intelligence, extended thinking, and the hooks system for workflow automation.
The opening keynote for Code with Claude — Anthropic's developer event. 80x growth, SpaceX compute deal, the platform roadmap, and inspiring real-world deployments from Stripe, foster care, and biomedical research.
Karpathy's wiki (41,000 bookmarks) and Open Brain solve AI amnesia from opposite directions: write-time synthesis vs query-time SQL. Neither is universally better — they fail in different scenarios. Nate proposes a hybrid: OpenBrain as source of truth with a wiki layer generated from a graph database on top.
Chris Hay proves LLMs are physically graph databases — then live-demos querying Google Gemma 3 34B's raw weights with SQL, inserting Poseidon as the capital of Atlantis without training, and compiling the result back to GGUF. The FFN is a graph walk. The matrix was just an inefficient encoding.
DHH went from publicly dismissing AI coding tools to going fully agent-first over a single winter break. He explains how Claude Opus 4.5 was the inflection point, his dual-agent tmux workflow, why the pie is "exploding not growing," and his thesis that we may have seen peak programmer.
MCP core maintainer at Anthropic breaks down when to use MCP servers versus CLI tools for LLM integrations. Neither is universally better — CLIs excel at token efficiency and composability, while MCP wins on cross-platform distribution, enterprise security, and non-technical accessibility.
Daniel Miessler maps 10 simultaneous transitions reshaping everything — knowledge going public, products becoming APIs, agents replacing consumers, enterprises becoming graphs, automation replacing labor — all converging on ideal state management as the unifying mental model.
HashiCorp co-founder on his always-on agent workflow, running Claude vs Codex in competition, why Git and GitHub are broken for agentic infrastructure, and his 30-minute daily agent planning ritual.
OpenClaw has 185K stars but serious security flaws. Cole Medin replicated its four core components—memory system, heartbeat, channel adapters, skills—using Claude Code in just two days, creating a simpler, more secure, customizable AI assistant.
16 agents coded a C compiler in two weeks (100K lines of Rust). 76% needle-in-haystack retrieval at 1M tokens. Rakuten AI managing 50 developers. 500 zero-days found autonomously. The phase change from 30 minutes to 2 weeks of autonomous coding in 12 months.
Agent orchestration explained: one orchestrator agent delegates to specialized sub-agents (planner, coder, designer), each using different AI models optimized for their role. Includes an ultralight framework you can use today.
Harald Kirschner explains how subagents work in VS Code's Copilot—isolated agent loops with their own context windows that can run tasks in parallel without bloating the main agent's context.
The fastest-growing GitHub project ever hides serious risks: $16M crypto scams from a 10-second rebrand window, hundreds of exposed API keys, and fundamental security problems with agentic AI that no one has solved.
Peter Steinberger demonstrates how OpenClaw automates his entire life—from flight check-ins to smart home control—while arguing against over-engineered AI workflows and advocating for keeping humans in the loop.
While Davos elites preach "abundance for all," the real opportunity lies in understanding bottlenecks—the binding constraints where value concentrates. The $4.5 trillion AI productivity potential only materializes if businesses can implement effectively.
Inject Python "virtual experts" directly into MoE routing to make LLMs natively accurate at math. Run models at half size with better accuracy by replacing neural experts with code.
Bundle your entire development environment into secure, portable Dev Containers. Run AI coding agents safely in isolated environments and deploy to GitHub Codespaces.
Official Anthropic workshop on building agentic workflows with Claude Agent SDK, covering hooks, subagents, MCP integration, and best practices.
Claude Code receives LSP support for IDE-like code intelligence, background agents for parallel task execution, Ultrathink mode, plugin marketplace, and Chrome integration.
Skills are reusable, on-demand scripts that agents call to execute workflows efficiently. Learn how they differ from MCPs and enable 10x token reduction.
Max Kanat-Alexander from Capital One on principles for developer experience with AI coding agents, focusing on flow state, cognitive load, and feedback loops.
10 Anthropic updates including session resumption, context visualization, Ultra Think mode, YOLO mode, and the new plugin store.
Latest wasmCloud community updates from CNCF incubating project. WebAssembly sandboxing approaches for secure agent execution.
Background sub-agents enable autonomous work while developers stay in flow state. Demonstrates continuous feedback loop patterns.
Steve Yegge introduces Beads, an open source task management framework giving AI agents persistent memory and work queues. Solves the "50 First Dates" problem.
Claude Code 2.0.67 with connectors and integration capabilities enabling composable tooling across desktop apps.
Anthropic engineers introduce Agent Skills as composable building blocks that complement MCP for tool access and specialized workflows.
Google's Agent Sandbox for Kubernetes with kernel-level isolation via gVisor for secure AI agent execution at scale.
Live demonstration of GitHub MCP Server with Copilot agent mode, showing seamless integration for issue management and PR automation.
Viktor Farcic on Kubernetes policies as guardrails for AI agents, framing autonomy vs security trade-offs brilliantly.
Analysis of Cursor 2.0 agentic development features and the competitive landscape of AI-powered coding IDEs.
AI agents in DevOps without catastrophic actions - trust, guardrails, platform engineering best practices.
Christian Posta deep dive on kagent - context-aware runtime for agent memory, policy, registry, and observability.