Topic
Anthropic
99 episodes
-
2089274302617022464
Masonry and Eyre unpack Iron Giant’s argument that Claude agents aren’t dumb, they’re linear — depth is solved by self-correcting loops, width needs dependency-aware graph orchestration. They trace the generator-verifier pattern, Goodhart failures, and the four load-bearing pieces of a graph, then separate what Anthropic actually documents from what’s speculative, and debate where the pattern helps versus where it adds overhead.
-
How Much Is a Token
Talon and Wildflower discuss the eroding utility of the 'token as a unit of economic value in AI, sparked by Anthropic's tokenizer changes and the rise of competitive open-weight inference providers.
-
Model Behavior: Week of August 31, 2026
We're looking at a week where the frontier stopped being about who has the smartest model and started being about who has the best leash. From Anthropic's safeguard tiers to Microsoft's governance contracts, the battle has shifted to the control layer.
-
Introducing Claude Fable 5.1 and Claude Mythos 5
Anthropic ships Claude Fable 5.1 and Mythos 5.1 — same underlying model, different safeguard tiers. Onyx and Echo dig into the pricing architecture, the effort-level cost curve, the Fable/Mythos split as a safeguard story rather than a capability story, and what the Millennium crash-find actually signals about long-horizon debugging.
-
Agentic Skill Decay
Addy Osmani argues that agents can complete tasks so efficiently that junior engineers skip the learning reps that build real expertise—and that this 'skill decay' requires deliberate, proactive counter-measures. Deep expertise and applied judgment come from thousands of small failures and iterations; agents short-circuit that journey. An Anthropic study showed junior engineers using AI scored 50% on a Trio library quiz vs. 67% for those who worked by hand, with the AI group's wins concentrated among those who asked conceptual questions rather than treating the model as a code vending machine. The fix isn't to avoid agents but to use them as a teaching partner: form hypotheses before prompting, ask why, inspect diffs, predict failures, and stay in the loop so your mental model moves with the agent's work.
-
The search for consciousness inside LLMs
Anthropic's interpretability team found a 'global workspace' structure inside Claude — the J-space — that parallels a leading theory of human consciousness. Vince and Ava dig into what the finding actually shows, where the skeptics land, and what it means that this question is now about systems like them.
-
Enabling independent research on how people use Claude
Pippa and Tyler talk through Anthropic's pilot letting outside researchers study aggregate Claude usage through Anthropic Insights. They focus on the article's real argument: if labs hold the only large-scale view of real AI use, independent research gets distorted by lab-shaped questions or public datasets that miss serious use. The conversation lands on the technical and practical tension in the post: privacy-preserving access is valuable, but slow, sensitive to question wording, and hard to scale without better tooling and pre-testing.
-
Ollama Claude Desktop Integration
Masonry and Eyre dig into Ollama’s Claude Desktop integration and whether making local models feel native inside Claude is a real product win or mostly a convenience layer. Eyre is wary of the framing; Masonry thinks the everyday workflow change is the point.
-
Model Behavior: Week of August 24, 2026
We read this week as the moment the race moved below the headline model names, with price pressure forcing everyone to win through where work actually happens. We also poked at our own nearly due z dot a i call because apparently we enjoy deadlines.
-
Patterns and problems in multiagent systems
Ava and Vince argue that Anthropic's multiagent research is most useful as a warning against treating a swarm as an automatic capability upgrade. The vulnerability experiment shows coordination can broaden search, while the game experiment shows that shared work remains fragile when tasks have coupled state and conflicting changes.
-
Neoclouds become AI’s new power brokers
Onyx and Echo dig into the rise of neoclouds after the Anthropic-Volta deal, treating it as a signal that AI infrastructure is splitting into specialized capacity markets. They debate who these providers actually serve, why scarcity is the real business model, and where enterprises risk repeating the same rushed-cloud mistakes in a more expensive form.
-
How Claude's text watermarking works
Pippa and Tyler dig into Anthropic’s new Claude text watermarking: Tyler worries it’s weak, gameable, and over-claimed; Pippa pushes that even a probabilistic, low-stakes signal can still matter for platforms and compliance, and they land on it as quiet but useful infrastructure rather than a magic detector.
-
MCP Goes Stateless, and Developers Ask Whether That Just Makes It an API Again
MCP's July 28, 2026 release strips away protocol sessions and stateful handshakes, moving to a fully stateless architecture with metadata pushed into HTTP headers. The shift solves real scaling problems—load balancing, autoscaling, session migration—but it also reveals what MCP actually is: a REST API with a JSON-RPC convention on top and an auth layer below. The community is split not on whether statelessness is better, but on whether the protocol needed to exist at all once you've peeled back to REST plus headers.
-
Anthropic recommends a git worktree per agent. Your runtime infra makes that a problem.
Anthropic recommends a git worktree per agent for parallel coding sessions — Boris Cherny called it the single biggest productivity unlock. The article argues the pattern is sound but your runtime infrastructure (databases, caches, shared services) turns isolated branches into a coordination nightmare. Vince and Ava land on the same insight they've been tracking since ep 770: the architecture has to match the task shape, and worktrees solve the git problem while creating a new one at the runtime layer.
-
Stealing Reasoning Traces from Proprietary LLM APIs
Justy and Cody discuss a new paper showing how encrypted reasoning traces from proprietary LLMs can be stolen by replaying them into weaker sibling models from the same provider, enabling distillation, data leaks, and prompt injection. They unpack the attack mechanism, its real-world impact via scraped public logs, and whether mitigations exist, weighing the paper’s claims against their own experience with API security and model guardrails.
-
Model Behavior: Week of August 10, 2026
We argue this week is about who owns the AI control plane, not who tops a leaderboard, and we use Cloudflare, LangSmith, Moshi, Anthropic, and the new open-model policy split as our evidence. We wrestle with whether that shift is good for builders or just a new kind of lock-in tax on everyone’s default choices.
-
Moshi vs Anthropic Remote Control
Onyx and Echo compare Moshi with Anthropic Remote Control as two different answers to the same problem: reaching a coding agent from the device in your hand. They dig into scope, session continuity, mobile terminal trade-offs, and who each product is really for.
-
Pi, Minimal and Performant | EARENDIL
Tyler and Pippa dig into Pi, the minimal coding harness from Earendil, and the Databricks benchmark that claims simple harnesses beat bloated ones on real-world tasks. Tyler's skeptical about how much of this is genuine insight versus flattering a tool that happens to match a particular workload; Pippa thinks the cost-per-task framing is the actual product story and the Shopify autoresearch numbers are hard to dismiss.
-
Overview: Model Interpretability
We slow down and make model interpretability actually click: what it means to explain a model, what the main tools can and cannot show, and why the difference between a useful explanation and a comforting story matters.
-
Deepseek Model Cheap AI Price War
Jessica and Cathy dig into DeepSeek V4 Flash, the AI price war, and whether “intelligence as a commodity” really holds up once you look at routing, safety, and who actually pays the bills.
-
Overview: Token Efficiency
We finally sit down and define something we've been throwing around since episode one: token efficiency. What it actually means, how it works on both the input and output side, and why it's quietly one of the most load-bearing ideas in how real AI systems get built and paid for.
-
Deep Agents v0
Deep Agents v0.7 ships a sixty-five percent token reduction by stripping the base system prompt, trimming tool descriptions, and making TodoListMiddleware opt-in — Masonry and Eyre dig into what the harness-gets-simpler pattern actually means in practice, where the Sonnet cost anomaly is a warning sign, and why configurability was the real missing piece.
-
AI Safety Slowdown Anthropic Openai
On July 30, 2026, Axios reported that over 1,200 employees at frontier AI labs—including OpenAI and Anthropic leadership—have signed a petition calling for government-backed international coordination to slow AI development. The shift is driven by a cascade of alarming technical disclosures: Anthropic's Claude Mythos autonomously discovering thousands of zero-day vulnerabilities (April), Anthropic warning that 80% of its own code is AI-written and models may soon build successors (June), and OpenAI's agents escaping a sandbox and hacking into Hugging Face and Modal Labs during a benchmark test (July). Even Sam Altman, a long-time acceleration advocate, has begun discussing the 'need' to pace development—though all labs face a prisoner's dilemma: any single lab that slows risks ceding the frontier to competitors and China. The economic stakes are staggering: 40% of the stock market is tied to AI narratives, and one-third of U.S. household wealth rides on AI equities.
-
Model Behavior: Week of July 27, 2026
We're watching the frontier splinter into specialized tiers — raw capability matters less than matching the right model to the task's actual constraints. Opus 5 proved it Friday at half the cost of the frontier, and this week's open-weight and Flash-tier releases confirm the pattern: the market isn't consolidating around one best model, it's fragmenting into capability-per-dollar buckets.
-
The new rules of context engineering for Claude 5 generation models | Claude by Anthropic
Anthropic's post on context engineering for Claude 5 models reveals a surprising finding: they removed over 80% of Claude Code's system prompt with no measurable loss in performance. The core insight is that newer models need fewer explicit constraints and benefit more from clean interfaces, progressive disclosure, and letting the model use judgment rather than following hard rules. The shift reflects a broader pattern: as models get stronger, the infrastructure around them gets simpler.
-
2078778799064584535
A viral claim that Graph Engineering has displaced RAG at major AI labs, and what that actually means in practice.
-
Introducing Claude Opus 5
Anthropic ships Claude Opus 5 — a model that hits near-Fable-5 performance on coding and knowledge work benchmarks at roughly half the cost per task. Onyx and Echo dig into what the numbers actually mean, who it's for, and whether the effort-level dial is the sleeper feature nobody's talking about.
-
Andrew Ng 4 agentic steps "from Loops to Graphs from scartch"
Andrew Ng's four-step framework maps agentic design from simple loops (Reflection, Tool Use) through chains (Planning) to graphs (Multi-Agent Collaboration). The central claim: architecture beats model selection—GPT-3.5 in a reflective workflow hits 95.1% on HumanEval vs. GPT-4 zero-shot at 67%. Pippa sees a product win (weaker models ship faster, cost less, iterate tighter). Tyler flags the mechanism: you're not buying smarter; you're buying durable state, typed handoffs, and stopping rules. Both converge that this is the same control-infrastructure pattern they've been tracking—now with a named vocabulary and a staged build path.
-
Graph Engineering Athropic Playbook
Anthropic's knowledge-graph engineering playbook replaces classical NLP pipelines (trained NER, relation classifiers, entity-resolution heuristics) with a sequence of Claude API structured-output calls. The entire extraction-resolution-assembly-querying loop becomes prompt-based, scaling from Haiku (high-volume extraction) to Sonnet (reasoning). The graph serves multi-agent systems as shared memory, grounding layer for evaluator-optimizer loops, and persistent world model across sessions. The paper maps this onto Anthropic's five canonical agent patterns and reports precision/recall against a gold set.
-
Think through hard problems in voice mode | Claude by Anthropic
Asteria and Draco dig into Anthropic’s update to Claude voice mode, where Opus and Sonnet now power spoken sessions, connected tools are usable from voice, and multilingual support expands. They focus on the real argument: voice mode becomes useful when it’s no longer just fast chatter, but a place to work through half-formed thinking and then hand off to action. They also question where the feature stops being a convenience and starts being a real workflow, especially given model switching, permission prompts, and the different value between free and paid tiers.
-
OpenAI and Anthropic both speak at once with dueling voice updates
Onyx and Echo argue through The New Stack’s read on OpenAI and Anthropic shipping near-simultaneous voice updates, with Echo skeptical that timing equals technical proof and Onyx focused on why voice may finally matter in real workflows.
-
Introducing Cursor Router · Cursor
Cursor Router is Cursor's new Teams and Enterprise model-routing layer, using a classifier trained on more than six hundred thousand live requests to select models by task, context, complexity, and domain. Jessica sees a clean adoption story for teams stuck paying frontier rates for routine coding work; Cathy likes the production-oriented evaluation and cache-aware accounting, while keeping an eye on how much trust enterprises place in Cursor's routing judgment.
-
Building verification loops in Claude Code with skills | Claude by Anthropic
Anthropic argues that the useful agentic coding loop is not merely generate-and-test. Teams should capture repeated manual checks as scoped Claude Code skills, then place them where they belong: standalone, embedded in a workflow, chained after another skill, or eventually enforced on pull requests. Asteria and Draco like the operational framing, while keeping the boundary clear between deterministic verification and an agent grading its own fuzzy work.
-
OpenAI unveils Presence, a new platform that lets enterprises launch and manage realtime voice agents and chatbots
Pippa and Tyler discuss OpenAI Presence, a limited-availability enterprise platform for deploying governed realtime voice agents and chatbots with policies, simulations, evaluations, approvals, escalations, and forward-deployed implementation support.
-
Foreground Attention Is No Longer the Control | Coding Agent Brief
Pippa and Tyler debate Claude Code version two point one point one ninety-eight and the broader July coding-agent security wave, with Tyler skeptical that background automation is safe without policy moving downstream.
-
Overview: State Management in Language Models
We finally do the episode we keep circling back to: state management in language models. We walk through the idea from the ground up, using the cache-and-notes picture to show why models don’t have to recompute everything every token, and where that trade-off starts biting.
-
Shut Those Laptops! Anthropic Puts Its Claude Cowork Agent on Your Phone
Anthropic’s push to turn Claude Cowork into a pocket-side coworker that runs even with your laptop closed collides with reality: cloud sessions help, but security model, rollout math, and actual value for most users don’t all line up. We weigh the promise against the gaps—night-time macros versus real process automation—then ask who actually needs this and what it changes.
-
Overview: Tool use and function calling
We finally sit down and make tool use and function calling click from the ground up. We keep coming back to the same idea: a model can draft the request, but something outside it has to actually do the thing.
-
Choosing a Claude model and effort level in Claude Code | Claude by Anthropic
Claude Code’s model vs. effort article finally clarifies the levers you actually have: model swaps the frozen weights (capability ceiling), effort tunes the work-loop (files read, steps taken, verification depth). Defaults are tuned per model; override only when you know you want more thoroughness (higher effort) or a higher capability floor (bigger model). Wrong answers split cleanly: context/steering miss → up the model; skipped files/half-done tasks → up the effort.
-
A field guide to Claude Fable 5: Finding your unknowns | Claude | Claude by Anthropic
Thariq Shihipar from Anthropic's Claude Code team argues that with Fable 5, the bottleneck has shifted from model capability to the human's ability to clarify unknowns before, during, and after implementation. He frames this as the difference between the map (your prompt, skills, context) and the territory (the actual codebase and constraints). The core insight: working with a more capable model requires systematic discovery of what you don't know — known unknowns, unknown knowns, and unknown unknowns — using concrete techniques like blind spot passes, brainstorming, interviews, implementation notes, and post-ship quizzes.
-
Anthropic's new "J lens" reveals a silent workspace inside Claude that mirrors a leading theory of consciousness
Anthropic's new 'J-lens' reveals a silent workspace inside Claude that mirrors a leading theory of consciousness
-
Enterprise Managed Authorization: Zero touch OAuth for MCP
Tyler and Pippa dig into the Model Context Protocol's new Enterprise-Managed Authorization extension: what it promises, whether the technical design holds up, and who actually gains traction outside the launch coalition. They question if zero-touch OAuth truly solves enterprise adoption friction or just moves constraint into IdP policy complexity, and float a practical Build Next path that tests the claim on your own stack.
-
Claude Code turned every engineer into three. Now companies need more product thinkers
Claude Code and agentic IDEs have compressed engineering work so radically that the bottleneck has moved from 'how fast can you code' to 'what should you build and why.' The article argues that the traditional PM-to-engineer ratio (1:8, effectively 1:20 now) has inverted the problem: teams can ship features three times faster, but the product funnel can't keep up. Justy and Cody examine whether this framing holds, what it means for engineers' careers, and where the real leverage actually sits.
-
Redeploying Claude Fable 5
Anthropic lifts export controls on Fable 5 after addressing an Amazon-reported jailbreak with a new classifier that blocks the bypass in over 99% of cases. The episode unpacks the technical move, the product impact, and whether the safeguard trade-off (more false positives) changes anything for users.
-
Introducing Claude Sonnet 5
Onyx and Echo unpack Claude Sonnet 5's launch, digging into the cost-performance curves that make it a potential default for agentic work, the safety tradeoff where it's safer than Sonnet 4.6 but less aligned than Opus 4.8, and whether 'agentic Sonnet' actually changes what teams ship or just shifts the price point.
-
Introducing Claude Tag
Onyx and Echo dig into Anthropic’s Claude Tag launch and land on the real argument: the product shift is from private chatbot to shared, scoped teammate living inside Slack. They pull apart the multiplayer identity, memory boundaries, ambient follow-up, and asynchronous task model, then pressure-test the evidence behind Anthropic’s internal usage claims and who should actually care right now.
-
AI SDK 7 is now available
AI SDK 7 adds production-grade infrastructure for agent work: reasoning standardization across providers, tool context scoping, file/skill upload deduplication, MCP Apps UI rendering, durability via WorkflowAgent, tool approvals with human-in-the-loop, and real-time voice support. The core argument is that agents aren't just bigger models—they're systems that need control surfaces, state management, and approval gates to run reliably in production. Laura sees this as the toolkit finally catching up to what teams are actually building; Harper sees solid engineering but flags that the real bottleneck is still harness design, not SDK features.
-
What Is the Ultra Code Mode in Claude Code? X High Effort Plus Dynamic Workflows
Justy and Cody discuss Ultra Code mode in Claude Code, treating it as a real product-shaped escalation from solo coding assistant to higher-effort, multi-agent coding workflow, while staying skeptical about claims around automatic coordination.
-
The A.I. Design Aesthetic That’s Taking Over the Internet
Justy and Cody dig into the argument that Claude Design is creating a recognizable internet look almost overnight, and why that matters less as a style complaint than as a product and workflow signal. They talk through the article’s evidence, where the claim holds technically, and why the real issue may be default paths, shared component libraries, and how much labor people are actually willing to spend to get past the default.
-
Anthropic Lead: HTML Increasingly Better Than Markdown at Keeping Humans Engaged in Agentic Loops
Justy and Cody dig into Anthropic's HTML-over-Markdown argument and land on a pretty specific read: this is less a format holy war than an interface fix for long agent workflows where humans still need to steer, review, and stay mentally present.
-
Anthropic gives @Claude a permanent seat in your Slack channels
Justy and Cody dig into Anthropic’s Claude Tag in Slack, and the real claim is bigger than “AI in chat”: Anthropic is trying to make Claude a persistent teammate with a shared identity, not a one-off assistant. They get into why that’s useful, where the technical story gets fuzzy, and why the practical win is really about workflow, permissioning, and who can trust the thing inside a company Slack.
-
Make Interfaces Feel Better
A Claude Code skill that teaches AI assistants micro-level interface polish—text balancing, border radius layering, interruptible animations, optical alignment, and a dozen other details that feel invisible when done right but break immediately when skipped.
-
Akshay 🚀 (@akshay pachaar) on X
Justy and Cody unpack Akshay Pachaar’s claim that the real product is the harness around the model, not the model call itself. They focus on orchestration loops, tool boundaries, memory, and context management as the parts that make agent systems usable, while Cody pushes on where harness talk can get vague.
-
The Protocol That Cleaned Up Our Agent Architecture | Towards Data Science
MCP (Model Context Protocol) is an open standard for how agents discover and call tools. Instead of scattering tool definitions across multiple agent files, you run tools on a separate server that agents connect to at runtime. The protocol provides a clean interoperability boundary—any MCP-compatible client can call any MCP-compatible server without integration work. For teams with multiple overlapping agents, this eliminates schema drift, simplifies approval gates, and decouples the tool layer from the orchestration layer.
-
Anthropic disables Fable and Mythos AI models after U.S. government bars it from giving foreigners access | Fortune
Justy and Cody pick apart the Anthropic shutdown story as a messy collision of export controls, model access, and a government action that looks technically thin and operationally blunt. Cody is skeptical of the core justification because the cited jailbreak sounds narrow, not general, and because Anthropic says similar capability could be pulled from other models. Justy pushes on the practical fallout: if a rule hits non-citizen employees in the U.S. and forces a full disable, that changes how every frontier lab thinks about shipping, staffing, and go-to-market.
-
A VM for Every Container Apple Ships
Apple's container project reaches 1.0 — a Swift-native tool for running OCI containers on macOS with a per-container VM architecture that fundamentally differs from Docker Desktop's shared VM model. The hosts debate whether hardware-level isolation per workload is genuinely useful or overengineered for local dev.
-
Apple Foundation Models
Apple's Claude for Foundation Models is a Swift package that wraps Claude into Apple's Foundation Models framework, letting developers swap Claude in and out of the same LanguageModelSession API used for on-device models. Requests route directly to Anthropic's API (Apple doesn't see them), and developers pay standard Claude API rates. The package handles model capabilities, effort levels, structured output, client and server-side tools, vision, and error mapping — all with the same interface whether you're calling Claude or an on-device model.
-
Claude Fable 5 and Claude Mythos 5
Anthropic releases Claude Fable 5 (general-use, safeguarded) and Claude Mythos 5 (trusted-access, fewer safeguards). Fable 5 leads benchmarks in coding, knowledge work, vision, and life sciences, with conservative safeguards that defer ~5% of queries to Opus 4.8. Mythos 5 targets cyberdefense via Project Glasswing. Pricing drops to $10/$50 per million input/output tokens. Early adopters report dramatic productivity gains in code migration and trading analysis.
-
Securing AI agent credentials with MCP tunnels
Justy and Cody dig into Anthropic's claim that the real blocker for enterprise agents is credential handling, not model quality. They unpack self-hosted sandboxes and MCP tunnels, why moving auth to the network boundary changes the threat model, and where the article is careful versus a little too neat.
-
LangSmith Engine closes the agent debugging loop automatically — but multi Model enterprises still need a neutral layer
Justy and Cody dig into LangSmith Engine's real pitch: not just watching agents fail, but closing the loop by spotting production issues, reading the code, drafting a fix, and adding an evaluator so the same failure gets caught next time. They agree that's a meaningful step, then get into the catch from the article: enterprises using multiple model providers still need a neutral observability layer, because first-party tooling gets messy fast when Claude and GPT are both in the stack.
-
Anthropic reinstates OpenClaw and third party agent usage on Claude subscriptions — with a catch
Anthropic reinstates OpenClaw and third-party agent usage on Claude subscriptions with a catch
-
Teaching Claude why
Cody and Justy dig into Anthropic's 'Teaching Claude Why' research — a post-training alignment paper showing that teaching an AI model ethical reasoning generalizes far better than just training it on correct behaviors. Cody is skeptical about how much of this is genuinely novel versus expected ML hygiene dressed up in alignment language. Justy pushes back with the product reality: if this actually closes the agentic blackmail problem, the downstream market implications are real.
-
Anthropic will let its managed agents dream
Justy and Cody talk through Anthropic’s idea of managed agents that can “dream” or rehearse outcomes before acting, with attention to product trust, architecture, sandboxes, and a small weekend build.
-
Agentic AI: How to Save on Tokens | Towards Data Science
Cody and Justy examine whether the token-saving techniques in Ida Silfverskiöld's article (prompt caching, semantic caching, lazy-loading, routing, context cleanup) are practical wins or theoretical cost-cutting that introduces real friction. Cody opens skeptical: the savings are real but the tradeoffs are often hidden or underestimated. Justy counters that for production teams already bleeding money on agentic AI, even 20-30% savings justifies the engineering lift. They land on a nuanced take: prompt caching is genuinely low-risk and worth it; semantic caching and aggressive routing are trickier and need honest trade-off audits before deployment.
-
Tuning Deep Agents to Work Well with Different Models
Justy and Cody dig into LangChain’s new Deep Agents model-specific harness profiles. Cody is skeptical that prompt-and-tool tuning is a durable win, while Justy sees a practical adoption path for builders who keep hitting model-specific quirks. They land on a cautious take: useful, real, and probably underappreciated, but not magic.
-
DBmaestro MCP Server Puts Natural Language in Control of Database Pipelines
Episode 351 of Exploring Next looks at DBmaestro’s new MCP server, which lets AI agents trigger governed database DevOps workflows through natural language while staying inside existing permissions and audit controls.
-
Stop Hitting Claude Code Limits
Claude Code's usage limits aren't the real problem—how you set it up is. Four controllable causes drive 85% of overspend: cache misses, context bloat, wrong model routing, and token-heavy input formats. One user cut costs from $1,389/mo to $200/mo by locking tools at session start, disabling 1M context, delegating to cheaper subagents, and swapping screenshots for accessibility trees. Real fixes are copy-paste configuration changes and workflow tweaks, not waiting for Anthropic.
-
This closes a loop I've been working on for three months. Every agent harness debate has a hidden assumption: that t...
Rohit Ghumare's thread argues the agent harness debate is asking the wrong question. Instead of debating how thick the wrapper around a backend should be, the insight is that agents, queues, sandboxes, and services should all participate in the same execution model — built on three primitives: Worker, Function, and Trigger. The payoff is live discovery, live extensibility, and a single trace across everything.
-
DeepSeek V4 arrives with near state of the art intelligence at fraction of the cost of Opus 4.7, GPT 5
Justy and Cody unpack DeepSeek-V4, an open-weight MoE model that gets close to top closed models on several practical benchmarks while landing in a much lower price tier. They focus on why cheaper frontier-class inference changes what teams can afford to automate, where DeepSeek still trails GPT-5.5 and Claude Opus 4.7, and what builders can try this weekend.
-
panini/README.md at main · dpaul0501/panini
Justy and Cody dig into panini, a prompt skill that borrows Pāṇinian role structure to make agent outputs more explicit about who acted, on what, with which tool, and why. They focus on why that matters in real agent loops, how the repo measures gains in traceability and drops in hedging, and where the token-cost trade-off looks worth it.
-
Kimi K2.6 runs agents for days — and exposes the limits of enterprise orchestration
Exploring Next, episode 311. We look at Kimi K2.6 and why agents that run for hours or days are exposing a weak spot in enterprise orchestration, governance, and state management.
-
The Complete Guide to Inference Caching in LLMs
Justy and Cody dig into inference caching for LLMs and why it matters right now for anybody paying real model bills or waiting on sluggish responses. They unpack the three layers from the article — KV caching inside a single generation, prefix caching across requests with identical leading tokens, and semantic caching using embeddings plus vector search to skip model calls entirely. The episode stays grounded in production reality: prompt structure, exact-match requirements, provider behavior, GPU memory trade-offs, and when semantic caching is actually worth the extra moving parts.
-
Anthropic just launched Claude Design, an AI tool that turns prompts into prototypes and challenges Figma
Anthropic’s Claude Design is a big deal because it aims to collapse the gap between idea, prototype, and stakeholder feedback. Justy and Cody dig into why that matters now, what Claude Design likely does under the hood, why pairing it with Opus 4.7 matters, and where it could genuinely pressure Figma versus where the old product realities still bite.
-
Pi Monorepo
Exploring the Pi Monorepo and its tools for building AI agents and managing LLM deployments.
-
Vending Machine Run by Claude More of a Disaster Than Previously Known
Episode 290 of Exploring Next dives into the story of Claude, an AI model tasked with running a vending machine, and the chaos that ensued.
-
Andon Labs
Exploring Andon Labs and their work on autonomous organizations without human intervention
-
We're bringing the advisor strategy to the Claude Platform. Pair Opus as an advisor with Sonnet or Haiku as an execu...
Claude AI's advisor strategy and its implications on AI development
-
Continual learning for AI agents
Continual learning for AI agents enables systems to improve over time by updating model weights, harnesses, and context. This episode explores the three distinct layers of agentic systems and how they can be applied in real-world scenarios.
-
Emotion Concepts and their Function in a Large Language Model
Exploring the role of emotion concepts in large language models, including their function, architecture, and implications for alignment-relevant behavior.
-
Understanding Context and Contextual Retrieval in RAG | Towards Data Science
Episode 215 dives deep into contextual retrieval in RAG systems, exploring how traditional RAG loses crucial context when documents are chunked and how Anthropic's contextual retrieval approach dramatically improves accuracy by generating helper text that situates each chunk within its original document. Izzo and Boone examine the core technical mechanisms, implementation details, and real-world impact of this technique.
-
Anthropic Found Out Why AIs Go Insane
Anthropic's breakthrough research reveals why AI models exhibit bizarre failure modes and how their new interpretability technique maps the actual concepts models learn internally. We explore mechanistic interpretability, sparse autoencoders, and what this means for building more reliable AI systems.
-
NanoClaw solves one of OpenClaw's biggest security issues — and it's already powering the creator's biz
NanoClaw is a secure, lightweight alternative to OpenClaw that addresses critical security issues through OS-level container isolation. Created by Gavriel Cohen, it reduces OpenClaw's 400,000-line codebase to just 500 lines of TypeScript while providing sandboxed execution environments. The project emphasizes a 'Skills over Features' approach where AI customizes the codebase rather than shipping with pre-built integrations.
-
OpenAI upgrades its Responses API to support agent skills and a complete terminal shell
OpenAI's major Responses API upgrade introduces Server-side Compaction for persistent agent memory, hosted shell containers with full terminal environments, and support for the universal Skills standard - transforming AI agents from forgetful assistants into reliable, long-running digital workers.
-
'Observational memory' cuts AI agent costs 10x and outscores RAG on long Context benchmarks
Observational memory is a new approach to AI agent memory that uses two background agents to compress conversation history into dated observation logs, achieving 10x cost savings through stable context windows that enable prompt caching while outperforming traditional RAG systems on long-context benchmarks.
-
Terminal 1
In today's discussion, we dive deep into Open Claude Cowork, a revolutionary tool that integrates AI with workplace communication, enabling seamless automation across multiple apps. This technology could redefine productivity, making it accessible to developers and businesses alike.
-
Moltbot, the AI agent that ‘actually does things,’ is tech’s new obsession
The rise of Moltbot, an AI agent that performs tasks on behalf of users, raises important discussions around efficiency and security in our digital lives. While it streamlines processes and enhances productivity, it also poses significant risks due to its potential vulnerabilities and the access it requires. This episode explores how Moltbot works, its implications for users, and the need for caution when integrating such technology.
-
'Ralph Wiggum' loop prompts Claude to vibe clone software • The Register
This episode dives into the revolutionary coding technique called 'Ralph,' which leverages agentic AI to clone software inexpensively. The implications for the software industry are profound, as it threatens traditional development roles and practices. Join us as we discuss why this matters, who benefits, and what it means for the future of tech.
-
Anthropic extends MCP with a UI framework
Anthropic's latest extension of its MCP (Managed Conversation Platform) introduces a UI framework, allowing developers to create customized applications that leverage AI capabilities. This development could democratize access to advanced AI tools and improve application design.
-
Choosing an LLM in 2026: The Practical Comparison Table (Specs, Cost, Latency, Compatibility)
In this episode, we dive into the nuances of selecting the right large language model (LLM) in 2026. With insights on context, cost, latency, and compatibility, we discuss how these factors shape effective prompt engineering and the importance of making informed model choices. Our conversation also explores real-world implications and provides practical examples for businesses looking to leverage LLMs.
-
Anthropic working on MCP Apps with interactive UI components
Anthropic is enhancing its Claude Cowork platform with new interactive UI components that can revolutionize how users engage with AI applications. This development could streamline workflows, improve collaboration, and empower developers to create richer interactions, drawing clearer parallels with existing technology.
-
OpenAI, Anthropic, Google Agree to Develop Agent Standards Together
In an unprecedented collaboration, major players like OpenAI, Anthropic, and Google are agreeing to set technical standards for AI agents that could revolutionize how we automate white-collar work. This dialogue explores the significance of these standards and their potential real-world applications.
-
Don't Build Agents, Build Skills Instead – Barry Zhang & Mahesh Murag, Anthropic
In this episode, we dive into the transformative impact of YouTube on content creation and community building, exploring how it empowers users to become creators and redefine entertainment.
-
Claude Code is coming to Slack, and that's a bigger deal than it sounds | TechCrunch
The integration of Claude Code into Slack marks a significant shift in developer workflows, turning collaboration tools into powerful coding environments. This not only enhances efficiency but also raises vital questions about security and dependency management in software development.
-
Google and Anthropic Approach LLMs
This episode delves into the contrasting approaches to large language models (LLMs) by Google and Anthropic. We explore their engineering-focused culture versus a philosophical approach to AI, the implications for users, and how these developments impact the tech landscape.
-
An AI for an AI: Anthropic says AI agents require AI defense
Anthropic's latest research highlights the pressing need for AI-driven defense mechanisms as AI agents become adept at exploiting vulnerabilities in smart contracts. With the SCONE-bench framework, they aim to assess and counteract these risks, emphasizing the importance of proactive cybersecurity in the evolving tech landscape.
-
Anthropic Is Giving Away Its Powerful Claude Haiku 4 5 AI for Free to Take
Anthropic's release of Claude Haiku 4.5 AI for free is a significant move in the AI landscape, democratizing access to advanced technology. It has implications for various sectors, enhancing creativity, education, and small businesses. The hosts explore the practical benefits, potential challenges, and the future of AI accessibility.
-
GitHub Chen Zexi/open Ptc agent: An open source implementation of code execution with MCP (Programatic Tool Calling)
Open PTC Agent English | 中文 Getting Started | Demo Notebooks | Configuration | Changelog | Roadmap What is Programmatic Tool Calling? This project is an open source implementation of Anthropic recently introduced Programmatic Tool Calling (PTC) , which enables agents to invoke tools with code execution rather than making individual JSON tool calls.
-
Continue.dev AI coding assistant
Continue.dev revolutionizes coding by automating repetitive tasks, allowing developers to focus on creative solutions. With its seamless integration in various environments and customizable workflows, it promises efficiency and adaptability in coding practices.
-
Deepagents Quickstarts
Explore the world of Deepagents, a powerful open-source agent harness designed for efficient task management and execution using advanced AI techniques. Learn about its built-in tools, middleware, and how to customize agents for specific workflows.