AI Catchup — Wednesday, August 26, 2026
Wednesday, August 26, 2026Inference
OpenAI's Jalapeño inference chip optimizes low-latency agent workloads
OpenAI reports first results from custom inference accelerator designed for agent tasks, aiming for year-end deployment.
Local-InferencePerplexity launches Portable Computer: fully local AI agents with zero cloud costs
Perplexity's Portable Computer runs on user hardware, keeping data local and eliminating API costs—now rolling out for Pro subscribers.
MemoryAnthropic merges Claude and Cowork memory, now synced by default across platforms
Claude's chat and agentic Cowork memory now share a unified system, enabled by default, letting AI remember work across contexts.
AgentsOpenAI's WebMCP challenge enables websites to connect directly to AI agents
OpenAI's WebMCP protocol lets websites define tool interfaces for direct agent integration, simplifying web automation.
RoboticsFigure uses crowdsourced training to scale humanoid robot capabilities
Figure's Index app pays 44,000 people to label and teach humanoid robots real-world tasks at scale.