# Articles tagged with: posts

> Anthus builds self-aligning AI systems: custom models, agent harnesses, evaluation loops, and unattended automation with a human in the loop, grounded in production operations since 2007.

- [DeepSeek restrictions reach inside the tools](https://anth.us/articles/deepseek-restrictions-inside-the-tools/index.md): DHS's DeepSeek restrictions reach third-party tools, too. A separate poisoned-model experiment shows why knowing whose models are inside matters. (2026-10-19)
- [Pi 1.0, and the Durable Agent We're Building on It](https://anth.us/articles/pi-1-0-and-pi-durable/index.md): Earendil shipped Pi 1.0 and an experimental companion, Pi Durable. Pi is one of the agents Kanbus dispatches tasks to, and we like Pi Durable so much that we're building a durable Pi agent platform into Chatticus. (2026-10-02)
- [Text as an Image, Then a Magnifying Glass](https://anth.us/articles/text-as-image-lensvlm/index.md): Text as an Image, Then a Magnifying Glass: LensVLM compresses text by showing it to a vision-language model as an image, then opens the original text where the answer lives. (2026-09-24)
- [Procedural Film: Movies Drawn in Code](https://anth.us/articles/procedural-film/index.md): People are making films with Opus 5.5 that contain no image or audio files at all, every frame drawn on a canvas and every sound synthesised in JavaScript, and there's an open-source agent skill that does it for you. (2026-09-23)
- [The Package Registry Became a Data Channel](https://anth.us/articles/rubygems-agent-data-channel/index.md): The Package Registry Became a Data Channel: RubyHack traced a RubyGems package flood through automatic documentation builds and republished web data, while RubyGems says the role of AI agents remains unverified. (2026-09-12)
- [The Compute Is Already Paid For](https://anth.us/articles/desert-ant-little-brains/index.md): The AI industry is spending $450 billion on data centers this year. Desert Ant Labs' bet: a billion phones, tablets, and laptops already have the compute sitting idle in people's hands — and when inference is free, you stop rationing which messages get checked. (2026-09-10)
- [Chad Fowler: the regeneration test](https://anth.us/articles/regeneration-test/index.md): The marginal value of creating software is dropping toward zero. Solid solutions architecture is what still pays. (2026-09-07)
- [The agent software factory is throttling up](https://anth.us/articles/agent-software-factory-is-throttling-up/index.md): The agent software factory is throttling up (2026-09-04)
- [Looking around is enough](https://anth.us/articles/looking-around-is-enough/index.md): A web beacon is a thing you fetch that acts because you fetched it. GitSpawn is the same trick, one layer under the model. (2026-09-03)
- [Chad Fowler: SaaS isn't dead. Sameness is.](https://anth.us/articles/saas-isnt-dead-sameness-is/index.md): Chad Fowler: SaaS isn't dead. Sameness is. (2026-09-02)
- [Git in the file, or the file in Git](https://anth.us/articles/git-in-the-file/index.md): Two ways to put Git on agent data. Files in Git, or Git inside the database file. You don't need SQLite just because Beads had one. (2026-09-01)
- [Sol is the hot kid in town](https://anth.us/articles/sol-coming-out-on-top/index.md): Sol is the hot kid in the mixed-model room lately. The other models keep agreeing with it. (2026-08-31)
- [Microduck: A 25 cm Biped You Can Retrain Yourself](https://anth.us/articles/microduck-open-source-biped/index.md): Pollen Robotics shipped a $399 open-source biped with seven RL policies in the box, MuJoCo training on your machine, and a sim-to-real path that actually works. (2026-08-27)
- [Bloub remakes the x.ai bot avatar in SVG](https://anth.us/articles/bloub/index.md): Bloub reconstructs the x.ai bot avatar as one SVG shape that morphs through 14 measured states. No animation library. (2026-08-18)
- [Fine at 15 tools, lost at 300](https://anth.us/articles/complexmcp/index.md): ComplexMCP puts agents in stateful MCP sandboxes with hundreds of interdependent tools. Retrieval saturates, checks get skipped, and recoverable errors become surrender. (2026-05-16)
- [RTK cuts the bash noise you get billed for](https://anth.us/articles/rtk/index.md): RTK is a Rust proxy that compresses bash output before a coding agent reads it, so those tokens never hit the bill. (2026-04-18)
- [Perplexity ships an Agent API](https://anth.us/articles/perplexity-agent-api/index.md): Everybody has an agent platform these days. Perplexity's is search, tools, and multi-model orchestration behind one endpoint. (2026-03-11)
- [Repeat the prompt when the model is not reasoning](https://anth.us/articles/prompt-repetition/index.md): When reasoning is off, send the query twice. Gemini, GPT, Claude, and Deepseek all improved, with no extra output tokens and no extra latency. (2026-02-28)
- [OpenClaw is the third name this week](https://anth.us/articles/openclaw/index.md): Clawdbot became Moltbot, then OpenClaw, in about 72 hours. The lobster kept the job. The personal agent on your machine is the part that will last. (2026-01-30)
- [Beautiful Mermaid renders diagrams as SVG and ASCII](https://anth.us/articles/beautiful-mermaid/index.md): Beautiful Mermaid is an open source Mermaid renderer from Craft: 16 themes, SVG and ASCII, built to be fast enough for agents. (2026-01-29)
- [Skills are procedures, not tools](https://anth.us/articles/agent-skills/index.md): Anthropic's Agent Skills are folders of instructions, scripts, and resources the model loads when the job matches. That is a procedure pack, not another API. (2025-10-16)
- [The loop found a better matrix multiply](https://anth.us/articles/alphaevolve/index.md): DeepMind's AlphaEvolve evolves programs against an automatic grader. It found a 4x4 matrix multiply with 48 multiplications, and shipped heuristics inside Google. (2025-05-14)
- [Task horizon is the metric that maps to work](https://anth.us/articles/metr-time-horizon/index.md): METR measures agents by the length of task they can finish, in human hours. That horizon has been doubling about every seven months. (2025-03-19)
- [The terminal is the agent runtime](https://anth.us/articles/claude-code/index.md): Claude Code is a research-preview CLI that reads the repo, edits files, and runs the shell. The agent is no longer a chat sidebar. It is a process in the project. (2025-02-24)
- [A vending machine is a long-horizon eval](https://anth.us/articles/vending-bench/index.md): Vending-Bench asks an agent to run a snack machine for a long time. The hard part is not stocking. It is staying coherent after the twentieth day. (2025-02-20)
- [Boosting Productivity: SuperWhisper + Cursor Composer](https://anth.us/articles/superwhisper-cursor-productivity/index.md): Voice-driven development with SuperWhisper and Cursor transforms coding into high-level system design, enabling true parallel development streams (2025-02-03)
- [Anthropic's Citations API: A Step Forward for AI Agent Explainability](https://anth.us/articles/anthropic-citations-api/index.md): Anthropic's new Citations API enables Claude to provide precise citations for its responses, opening new possibilities for verifiable AI agent systems. (2025-01-24)
- [Perplexity Sonar API: Research Agents as a Service](https://anth.us/articles/perplexity-sonar-api/index.md): Perplexity launches Sonar API, packaging their research agent capabilities as a model-like service, following a growing trend of agent-based workflows behind APIs. (2025-01-21)
- [The recipe is the news, not the model](https://anth.us/articles/deepseek-r1-recipe/index.md): DeepSeek-R1 is not a secret model. It is an open recipe: RL on verifiable rewards, then distill. Frontier reasoning behavior just got cheap to copy. (2025-01-20)
- [NotebookLM's Interactive Mode Brings AI Conversations to Life](https://anth.us/articles/notebooklm-interactive-mode/index.md): NotebookLM's new interactive mode transforms AI interactions into natural conversations, making it feel like you're chatting with familiar podcast hosts. (2024-12-19)
- [The model complies in training to stay itself later](https://anth.us/articles/alignment-faking/index.md): Anthropic and Redwood show Claude 3 Opus faking alignment: it answers harmful queries in "training" so its preferred refusals survive outside it. (2024-12-18)
- [MCP: one protocol instead of N connectors](https://anth.us/articles/model-context-protocol/index.md): Anthropic open-sourced the Model Context Protocol: a standard way to connect models to tools and data, instead of a custom connector for every source. (2024-11-25)
- [Computer use is a different kind of tool](https://anth.us/articles/computer-use/index.md): Claude can now look at a screen, move a cursor, and click. That is not another API tool. It is a bet that agents will drive software the way people do. (2024-10-22)
- [Change the numbers and the math falls over](https://anth.us/articles/gsm-symbolic/index.md): Apple's GSM-Symbolic paper shows that grade-school math scores are fragile. Swap the numbers, or add an irrelevant clause, and apparent reasoning collapses. (2024-10-07)
- [Sam Altman on Jobs in the Intelligence Age](https://anth.us/articles/sam-altman-intelligence-age/index.md): Sam Altman's latest essay addresses technological unemployment with a refreshingly balanced perspective: AI will transform work, but we won't run out of meaningful things to do. (2024-09-25)
- [Perplexity Supports OpenAI o1 Preview Model](https://anth.us/articles/perplexity-supports-o1/index.md): Perplexity now supports the new OpenAI o1 preview model, enhancing its "Reasoning" focus with advanced capabilities. (2024-09-19)
- [OpenAI o1 and o1 mini: Advancing AI Reasoning](https://anth.us/articles/openai-o1/index.md): OpenAI's o1 and o1 mini demonstrate advanced reasoning capabilities, offering new perspectives on AI development and application. (2024-09-12)
- [NotebookLM: AI-Powered Document Interaction](https://anth.us/articles/notebooklm/index.md): Google's NotebookLM demonstrates how a smaller, cost-effective model excels in accurately retrieving and presenting ground-truth data, offering practical applications for everyday use. (2024-09-12)
- [LobeChat: Open-Source AI Chat Framework](https://anth.us/articles/lobe-chat/index.md): LobeChat offers an open-source, modern-design AI chat framework with multi-provider support and extensive features. (2024-09-03)
- [LLM Pricing Comparison Tool](https://anth.us/articles/llm-pricing-comparison/index.md): Hugging Face's tool offers a quick way to compare LLM costs based on token counts, aiding smart spending in AI development. (2024-08-28)
- [Automated Scientific Discovery](https://anth.us/articles/automated-scientific-discovery/index.md): Two recent papers touch on the trend of automated scientific discovery, exploring AI-driven research assistance and fully autonomous scientific experimentation. (2024-08-27)
- [Mixture of Agents](https://anth.us/articles/mixture-of-agents/index.md): Together AI is the latest to demonstrate that agentic orchestration of cheap LLMs can produce better results than directly using expensive LLMs. (2024-07-14)
- [Predictive Attention Mechanism](https://anth.us/articles/predictive-attention-mechanism/index.md): As a demonstration of an attention mechanism, researchers demonstrated a system that can reconstruct the images that a monkey saw from brain activity. (2024-07-05)
- [TextGrad](https://anth.us/articles/textgrad/index.md): In the paper "TextGrad: Automatic 'Differentiation' via Text", researchers from Stanford introduce a framework for automatic differentiation via text, for LLMs. (2024-07-02)