The Summer of George…I mean Agents.
With the summer solstice having come and gone, it feels like as good a time as any to name this the Summer of Agents. Agentic automation made its debut on the "How People Are Using AI" index from HBR.
We have multiple agent frameworks from the Vercel and Astro teams dropping within days of each other with similar architectures for building custom agents.
We even had a new model top the charts for a couple days in Fable, reminding us that ownership of the model is fickle at best, so jam out on it while you can with skills like /improve from the creator of shadcn.
It's the summertime and I'm traveling, so let's just dive straight into the links.
🛠️ Framework
Flue 1.0 Beta
Flue is the TypeScript based OSS framework for building agents and workflows from the Astro team. It is built on the Pi harness (same that runs OpenClaw). It gives a solid foundation for everything needed to build production ready agents. Cloudflare also entered the chat as a place for your Flue agents to deploy.
🛠️ Framework
Vercel's eve
Vercel also shipped an open-source agent framework based on the idea that “an agent is a directory”. Again the idea is to include everything you need to take an agent to production in one framework. To let them explain it:
Each file describes one component of the agent, so at a glance, the tree tells you what an agent is, what it does, where it lives, and when it acts on its own.
🗄️ Harness
Poolside
Poolside is a full-stack AI software platform that runs entirely inside your infrastructure, so code, prompts, and data never leave your VPC, with governance, auditability, and policy enforcement as the headline features. This is a tool for those occasions where you never want any code or prompts to leave the building.
🥷 Skill
shadcn/improve
The developer behind shadcn dropped the /improve skill just in time for me to give it a test run with Fable before it got shut down. The concept is great. Use the high-end expensive models to plan the work to give to lower cost models. The experience was great. I used Fable to plan a PHP upgrade path for an outdated WordPress site. It made the plan and handled kicking it off to Sonnet powered sub-agents to get the work done. All the work was completed in a single Claude usage session.
Something to keep in mind as we may have less time with frontier models going forward. Use them while you can to do the hard planning work and let smaller or local models handle the integration.
Pretty wild we’re already up to HBR’s third annual study on how people are really using AI. It is interesting to see how AI is being used, particularly getting outside of the engineering bubble I mostly live in. I’m not surprised to see autonomous agentic operations making its debut on the list way up at #6 though. Though there’s maybe a bit too much life advice being given by AI tools for my liking.
Thanks for reading,
Jason
