Videos

Agent orchestration code is being replaced by files: markdown instructions, YAML configurations, and bash skills are...

Vercel's shift from a multi-agent architecture to a single file-system-based agent, inspired by Claude Code and Opus...

A converged database that can natively store JSON, graph, vector, and relational data in the same table is...

A Claude command center that pulls real-time data from your connected apps is now buildable in about 15 minutes...

PostHog's wizard, an agentic CLI that sets up PostHog in minutes, required a security posture shift from...

Recursive self-improvement in AI is the next step after automating feature engineering and prompt engineering, where...

Restate is an open-source durable execution framework that treats agents as persistent, stateful entities rather...

Loophole is an open-source adversarial agent framework that stress-tests user-defined moral principles by converting...

Mike Chambers argues that production agents need a 'harness'—the scaffolding around the model that handles memory,...

True Forge is an MIT-licensed, open-source agent harness that lets you run managed agents on your own hardware with...

Four major AI labs agreed to slow AI development, but the only funded commitment is embedded third-party evaluators...

Cognition's SWE-2 is a post-trained version of Kimi K3 that improves agentic coding, matching GPT-4.5 on DeepSWE...

Apple's Reference Image signs every pixel at the sensor level, creating a verifiable digital negative that survives...

Anthropic's alignment science lead Evan Hubinger publicly estimated a greater than 10% chance that AI kills all...

OpenAI, Anthropic, and Elon Musk are advocating to 'pace the frontier' of AI development, which the speaker argues...

OpenAI ran Python at 20 million requests per second for their Habitat storage service, deliberately incurring...

DeepSeek V4.1 is roughly 88x cheaper than GPT Astra and performs competitively on key coding and automation...

Anthropic engineers are moving away from building a separate agent for every job and instead building...

Managed agents are becoming the standard product across major AI companies, with Anthropic, Google, AWS, Microsoft,...

DeepSeek V4.1 Flash uses a custom GPU kernel called Deep Select to reduce per-token KV cache to 890 bytes, down from...

OpenAI claims to have found a counterexample where the Navier-Stokes equation blows up, using 10,000 concurrent...

Pi agent's success comes from its minimal, bash-centric approach that lets models pipeline commands efficiently...

DeepSeek V4.1 Flash cuts per-token cache memory from 389,000 bytes to 890 bytes — a 437× reduction — by rebuilding...

Meta launched Muse, a consumer AI agent that packages browser automation, recurring heartbeat tasks, goals, and...

The AI software factory concept is becoming viable with GPT-6 Astra, but it's still an early alpha. The real story...

Neo Horse's post-training improved Qwen's agent performance without moving to a larger model, with the 4B model's...

The harness around an AI model matters far more than the model itself. Most of what makes Claude Code or Codex look...

Running DeepSeek V4.1 Flash locally requires at least 510 GB of storage and a multi-GPU setup costing tens of...

DeepSeek V4.1 Flash is a 552B parameter MoE model with only 8B/16B active parameters, matching Opus 5 and GPT 5.6 on...

Dioxus, a Rust cross-platform app framework acquired by Cognition, found that AI coding agents accelerated tasks...

Fidelity's refusal to sell its 8 petabytes of data illustrates that the real AI advantage comes from proprietary...

Using Higsfield Supercomput, the creator built a playable browser game that simulates three different AI futures for...

Regular GPU maintenance, especially replacing thermal paste with PTM phase sheets, can prevent throttling and extend...

Spotify's open-source Shunt plugin cuts Claude Code token usage on file reads by 82-94% using a 33-line bash hook...

ChatGPT Astra 6 can now generate polished, responsive websites from reference designs, using a 'design loop' that...

OpenAI's model solved the 80-year-old Navier Stokes Millennium Prize problem in five days, and the company disclosed...

OpenAI's internal data shows agent runtime surpassed human work in June and hit a 3.14:1 ratio by mid-August, but...

By combining ChatGPT 6 Astra's computer-use ability with Blender and Higsfield, a single prompt generated a fully...

Austin Marchese's six-question framework uses Claude as a stateful tutor to compress skill acquisition,...

AI Engineer's lone designer scaled output from a manual bottleneck to hundreds of deliverables by combining a strict...

Prefab is a Python framework that lets you build interactive UIs for MCP apps using nothing but Python context...

The paper 'Prompt Injection as Role Confusion' shows that LLM role perception can be manipulated by injecting forged...

The real demonstration is Manus' carry-forward: correct a research output once, approve the agent's proposed...

Taste Labs has shown that AI slop in design is quantitatively measurable using pattern mining probes trained on over...

Impeccable is a design tool that uses adjectives and verbs to steer AI output, offering a middle ground between...

DeepSeek V4.1 Flash is a significant open model with a 552B parameter mixture-of-experts architecture,...

The talk argues that agents have a measurement problem because token spend is not clearly tied to outcomes, making...

tldraw's spatial harness for agents turns the infinite canvas into a medium where LLMs can understand 2D space, set...

MiniCPM5-2B is a 2.5B parameter model that excels at tool use and function calling, making it a strong candidate for...