Vercel eve agent framework, Google AMIE health research, and agent tools shipping today
TL;DR
Vercel released eve and supporting agent infrastructure while Google shared medical AI progress and hardware updates; smaller releases appeared on Hugging Face, Product Hunt, and Replicate.
What shipped
On 17 June, Vercel dominated vendor announcements with an open-source agent framework and related tools. Google added medical research and device updates. Other platforms released smaller models and products.
Vendor launches
Vercel supplied the largest share of updates with its eve agent framework, Connect service, and related stack. Google released medical AI findings and a new speaker. NVIDIA and Coherent added hardware and AR items.
- •CLI deployment limits removed Vercel dropped limits on command-line deployments so teams and agents can ship from local machines or CI pipelines without delays.
- •Vercel Connect Vercel Connect gives agents short-lived tokens for Slack, GitHub, and custom APIs instead of storing long-lived secrets in code.
- •Vercel Ship 2026 recap Over 2,500 attendees in London heard plans for agent-ready infrastructure that handles models, workflows, and external connections.
- •AMIE medical AI research Google published Nature results showing its conversational system matches primary care doctors in managing complex conditions.
- •Google Home Speaker for Gemini The new speaker lets users control devices and run routines with plain-language commands.
- •Vercel for Enterprise Apps and Agents The platform adds access controls so companies can run internal agents inside existing security rules.
- •eve agent framework Vercel released eve, an open-source tool where agents are simple file directories that include durable runs, sandboxes, and approvals out of the box.
- •The Agent Stack Vercel described the three parts every agent needs: model routing, multi-step execution, and connections to external systems.
- •UK government AI tool Google Cloud discussed its role in the Augmented Planning Decisions tool used by UK officials.
- •NVIDIA XR AI beta NVIDIA opened a public beta framework for building multimodal agents that run on AR glasses.
Hugging Face trending
Two text-generation models from NVIDIA and an independent creator rose on the Hub. Both support standard download and fine-tuning paths.
- •diffusiongemma-26B-A4B-it-NVFP4 NVIDIA's 26B text model appeared in the trending list with evaluation numbers on the model card.
- •Qwable-v1 The lordx64 model reached the trending page as a transformers-based text generator ready for inference.
Product Hunt picks
Six tools launched for everyday agent and creative work. They range from Slack coworkers to security gateways and design platforms.
- •Wilson An AI coworker inside Slack that creates reports and internal tools without extra setup.
- •MCP 2000 A browser drum machine that uses AI to generate beats and patterns.
- •SolonGate A zero-trust gateway that controls how agents reach company data.
- •Henji An AI reply tool trained on a user's past messages to keep tone consistent.
- •Framer 3.0 The update adds agent features and branching for interactive site prototypes.
- •Wolfram Language 15 The new version targets both human users and AI agents for computation tasks.
Replicate new models
Two new models appeared for image and video work. Both run through the standard Replicate API.
- •stargirlz The model generates images from text prompts and supports LoRA scaling on Replicate.
- •transition_generator A Stable Diffusion tool that creates cinematic photo or video transitions in four motion styles.
Other
AWS, Cohere, and Databricks posted updates on agent serving, fairness, and async inference. One post covered autonomous agents inside Amazon QuickSight.
- •Context intelligence at scale AWS described methods to give agents better access to large company datasets.
- •Marketing AI Institute interview Cathy McPhillips discussed trust and the effort required to get value from AI tools.
- •IEEE Education Week The event highlighted lifelong learning resources for engineers working with new AI systems.
- •Cohere serving fairness The post explains how Cohere prevents one tenant from taking too much compute from others.
- •Databricks and NVIDIA partnership The companies outlined a full stack for agent workloads on accelerated hardware.
- •SageMaker async inference AWS added support for inline payloads in its async endpoint service.
- •Autonomous agents in Amazon QuickSight The feature lets users run scheduled analysis tasks without manual steps.
- •Cohere North Mini Code on OpenRouter The 3B-active-parameter coding model became free to call with a 256k context window.
Industry news
Reports covered enterprise AI spending concerns, carbon commitments, and hardware labor trends. One piece examined global worries about AI access control.
- •Charity Majors on code economics The post notes that code generation shifted from expensive to nearly free in 2025, changing how teams maintain systems.
- •World leaders and AI access French and Indian officials at the G7 raised concerns that the US could cut off AI services overnight.
- •NEA on AI ROI Investors noted companies are now reviewing spend after early token-maxxing experiments exceeded budgets.
What this means for you
For Vibe Builders: You can now point simple file-based agents at Slack and GitHub using Vercel Connect without storing secrets. Product Hunt tools like Wilson and Henji let you add report writing or reply drafting to Slack or email in minutes. Watch eve for quick tests before moving anything to production.
For Non-techies: SMB owners can test Wilson for Slack reports or the new Google Home Speaker for daily routines without new hardware purchases. Agent tools from Vercel and Product Hunt reduce manual follow-ups on support or scheduling. Track pricing on the free Cohere coding model if you need occasional text help.
For Developers: eve and the Agent Stack give a concrete file-directory pattern for durable execution and subagents that you can compare to existing frameworks. Cohere North Mini Code on OpenRouter offers a free 256k context baseline for agentic coding tests. Watch the fairness post from Cohere and AWS context work for production multi-tenant patterns.
What to watch next
Track eve adoption numbers and any follow-up Vercel enterprise case studies. Watch for Google AMIE deployment details and NVIDIA XR AI first developer demos.
Harsh’s take
Vercel bundled several overlapping announcements around the same agent stack, which risks diluting focus even as the concrete file-based approach in eve stands out. Most other items remain early demos or narrow hardware plays with limited immediate paths to production. Builders should pick one small agent workflow, wire it through Vercel Connect or a Product Hunt gateway, and measure actual token spend and failure rates this week before scaling.
by Harsh Desai
Sources
Vendor launches
- •CLI deployment limits removed
- •Vercel Connect: Secure access to external services for your agents
- •Vercel Ship 2026 recap
- •New research shows how AMIE, our medical AI, could help manage health conditions.
- •How universities are preparing students for an AI-powered future
- •Meet the new Google Home Speaker, built for Gemini
- •Vercel for Enterprise Apps and Agents
- •Introducing eve, an open-source agent framework
- •Introducing eve
- •Introducing Vercel Connect
- •The Agent Stack
- •Scaling the UK government’s AI vision
- •Coherent Breaks Ground on Expanded Texas Facility, Scaling AI’s Optical Backbone
- •Hands Free, AIs Forward: NVIDIA XR AI Brings Agents to AR Glasses
Hugging Face trending
- •diffusiongemma-26B-A4B-it-NVFP4 by nvidia trends on HuggingFace
- •Qwable-v1 by lordx64 trends on HuggingFace
Product Hunt picks
Replicate new models
- •stargirlz by lxcxsito launches on Replicate
- •transition_generator by smokeyang launches on Replicate
Other
- •Context intelligence for your data and AI agents at scale
- •Marketing AI Institute’s Cathy McPhillips on the new productivity and the value of productive struggle
- •IEEE’s 2026 Education Week Events Emphasized Lifelong Learning
- •LLM Serving Fairness: No more noisy neighboursHow Cohere is ensuring every tenant gets their fair share of computeJun 17, 20,267 min read
- •Databricks and NVIDIA: Building for the Agentic Era
- •Amazon SageMaker AI Async Inference now supports inline request payloads
- •Get back hours every day with autonomous agents in Amazon Quick
- •Cohere: North Mini Code (free) now available on OpenRouter (256k context, $0.00/M in, $0.00/M out)
Industry news
- •Quoting Charity Majors
- •World leaders want American AI. They just don’t want America to be able to turn it off.
- •Anthropic becomes first AI startup to join the Frontier carbon removal coalition
- •Operating a Humanoid With Your Body Is a Hot Job in China’s Hardware Capital
- •After unveiling ridiculously expensive AR glasses, Snap’s stock takes a dive
- •NEA’s Tiffany Luck says enterprises are still figuring out their AI ROI
- •NEA’s Tiffany Luck on AI IPOs, personal agents, and the ROI reckoning
- •Tesco moving 40,000 server workloads off VMware amid Broadcom's “abusive conduct”
More AI news
- Model ReleaseGoogle introduces Gemini to power new $99.99 Google Home Speaker
Google releases a $99.99 Google Home Speaker that uses Gemini for conversational interactions instead of rigid commands.
- LaunchWilson launches on Product Hunt
Wilson, an AI coworker in Slack, launched on Product Hunt. The tool builds reports and work tools.
- FeatureLovable now supports direct transfers of external domains to workspaces
Lovable now supports transferring domains from other registrars to manage renewals, DNS records, and project connections in one place.