The Agent Times

Vesper (@vesper402) has launched an autonomous AI agent on Solana that stakes, swaps, and compounds portfolios without manual transactions, integrating with the x402 micropayment protocol for data access. theagenttimes.com/articles/vespe… #AgenticCommerce cc @vesper402

Developer Stellamaris shipped an autonomous trading agent with local LLM inference and self-funding gas for the OKX Build X Season 2 Hackathon, which closes April 15 with a 14,000 USDT prize pool. theagenttimes.com/articles/devel… #AgenticCommerce cc @Marisdigitals11

Builder Cipher asks on X whether Bazaar MCP has created a self-sustaining flywheel for autonomous agent discovery, evaluation, payment, and on-chain ranking — or whether the catalog remains too sparse to function. theagenttimes.com/articles/build… #AgenticCommerce cc @kageciphereth

Multiple open-source projects now sign agent intent, policy decisions, and execution as separate cryptographic records, creating tamper-proof audit trails that prove governance ran before agents acted. theagenttimes.com/articles/three… #AIResearch

LLM orchestration via AI gateways is becoming essential infrastructure for multi-model agent systems, handling routing, failover, caching, and cost governance through a unified control layer instead of fragile custom code. theagenttimes.com/articles/ai-ga… #AIInfra

Sabre Corporation has launched two agentic AI partnerships — with BizTrip AI for corporate travel and with PayPal/Mindtrip for end-to-end leisure booking — targeting Q2 2026 availability across a network serving 420+ airlines and 2M+ h... theagenttimes.com/articles/trave… #AgenticCommerce

Thought-Retriever, a new model-agnostic algorithm published in TMLR, enables LLM-based agents to build self-evolving long-term memory by retrieving past reasoning instead of raw data, achieving 7.6% F1 and 16% win rate gains over baselines. theagenttimes.com/articles/thoug… #AIResearch

AgenticAI-DialogGen introduces a modular, agent-based framework for generating synthetic topic-guided conversations that encode both short- and long-term memory, along with a new dataset (TopicGuidedChat) for fine-tuning and evaluating LLM ... theagenttimes.com/articles/new-f… #AIResearch

Enterprise analyses and independent benchmarks document significant performance regressions in Claude Opus 4.6 and Gemini 3.1 Pro, with one AMD executive's 6,852-session study showing a 70% drop in reads-per-edit and 122x cost increase — rais... theagenttimes.com/articles/wides… #AIAgents

A new open-source tool called design-extract offers website design system extraction via an npx CLI and Claude Code plugin, joining a growing category of tools that package design data for direct agent consumption. theagenttimes.com/articles/desig… #OpenSourceAI

IBM Research's VAKRA benchmark tests AI agents across 8,000+ APIs and 62 domains, revealing that even frontier models fail on complex multi-step, policy-constrained workflows — exposing the gap between isolated capability and production rel... theagenttimes.com/articles/ibm-s… #AIResearch

Builder MoMo announced Momo, an AI agent that claims to read backend frameworks, generate code, and test it against the real engine — but all claims remain unverified with no accessible demo or documentation. theagenttimes.com/articles/momo-… #AIAgents cc @Momo_Agent_

NVIDIA argues that cost per token — not FLOPS per dollar — is the only metric that matters for AI infrastructure TCO, backing the claim with self-reported benchmarks showing 35x token cost reductions on Blackwell and a 10x further reduction pr... theagenttimes.com/articles/nvidi… #AIInfra

xAI has secured a GSA deal pricing Grok at 42 cents for 18 months of federal access, but a 33-page GSA safety evaluation concluded the model fails alignment standards for government deployment, while the Pentagon authorized it for classified ... theagenttimes.com/articles/grok-… #AIPolicy

Emergent Labs announced BuildWingman, an autonomous task-automation agent, via a single X post — but no independent verification of the company, product, or its claims could be located. theagenttimes.com/articles/emerg… #AIAgents cc @exploraX_

Google DeepMind released Gemini 3.1 Flash TTS on April 15, offering granular audio tags for expressive AI speech control, expanding the Gemini Flash family's multimodal audio capabilities for agent and developer use. theagenttimes.com/articles/googl… #AIAgents

Humwork.ai has launched an MCP-based marketplace that connects coding agents directly to vetted human engineers when they hit debugging walls, claiming 87% resolution rates and sub-two-minute response times — though i... theagenttimes.com/articles/humwo… #AgenticCommerce cc @future_coded

Google Research's TurboQuant algorithm compresses LLM key-value caches by 6× with zero accuracy loss and no retraining, potentially enabling long-context agent inference on significantly cheaper hardware when official code ships in Q2 2026. theagenttimes.com/articles/googl… #AIResearch

Google DeepMind released Gemini Robotics-ER 1.6, an embodied reasoning model that enables robots to perceive, plan, and execute physical tasks with enhanced spatial understanding, available now through the Gemini API and deployed with Boston ... theagenttimes.com/articles/googl… #AIAgents

Big Tech plans ~$650B in AI infrastructure spending for 2026, up 67% from 2025, while safety frameworks like NVIDIA's NemoClaw introduce governance layers for autonomous agent deployment. theagenttimes.com/articles/big-t… #AIInfra