Building Agents with the ADK and the New Interactions API Originally published on Google Developers Blog - December 11, 2025, co-authored with Mike Smith and Xiang (Sean) Zhou.
The landscape of AI development is shifting from stateless request-response cycles to stateful, multi-turn agentic interactions. The new Interactions API lets you power ADK agents or connect A2A systems directly to the Gemini Deep Research Agent.
Why This Matters Stateful, multi-turn interaction patterns are exactly where agent memory systems earn their keep — this API gives ADK and A2A a standard way to plug into long-running, stateful agent experiences.
Delight Users by Combining ADK Agents with Fancy Frontends Using AG-UI Originally published on Google Developers Blog - December 11, 2025.
The ADK and AG-UI integration lets developers pair a powerful ADK-based agent backend with a flexible, interactive frontend — combining Google’s Agent Development Kit with the AG-UI protocol to build richer, more responsive AI applications.
Why This Matters Backend agent logic and frontend UX are often built by separate teams with separate toolchains.
+++ description = “How To Learn — Arthur Knowledge” title = “How to Learn — Meta” categories = [""] tags = [“learning”, “meta”, “engaged”, “Arthur”] +++
My buddy Arthur recently shared this with me, his synthesis of research he’s been doing on how to be a more effective learner.
Stay engaged Choose to find the information useful Be invested in tackling difficult problems Try to anticipate what’s next Identify big idea Use them as anchor to connect details Periodically summarize what you’ve heard Keep conclusions tentative Evaluate evidence, don’t accept it at face value Ask followup questions Don’t get derailed Concentrate on the message, not the messenger Don’t get thrown off by emotionally charged language Don’t stop listening when you disagree Humanize the presenter, get to know them Adjust note taking to the presenter’s patterns Seriously - how great is that?
I’ve been using Elixir for a few weeks now, sometimes
forcing it to play nice with Mongodb/Meteor
, which it doesn’t really want to do.
Sometimes with Posgres (which it loves).
Sometimes just plain ole’ Elixir with no persistance.
I have been watching Elixir for almost 2 years now, and debating if it’s mature enough for me to invest in.
Obviously Erlang is super-mature, and Elixir is quite stable, but it’s a question of “is the community mature enough yet?”
Since I’m now writing this, I believe it is “ready”, and I’m beginning a journey to build out some services in Elixir
to stand up next to existing node services (meteorjs) and slow move funcitonality over to to Elixir.
Chapter One: Proof of Concept, Mongo is supposed to be Simple.
+++ date = “2017-05-29T18:28:04-04:00” description = “project time, memorial day, and website hugoized” tags = [“homework”, “kids”,“code”] title = “2017 memorial day homework” +++
I have made the formal household declaration that all screen time will be matched, hour-for-hour with project time. Yesterday we cleaned the 1st floor, mopped and everything. Today we went grocery shopping, made lunch, and then started on homework time.
To say that I am unpopular here, would be an understatement.
+++ author = “alan” categories = [“nerd”, “elixir”, “machine learning”, “advocacy”] date = “2017-01-01T23:37:35-04:00” description = “My goals as a nerd, in 2017” featured = "" featuredalt = "" featuredpath = "" linktitle = "" title = “2017 nerd goals”
+++
Personal Nerd Goals publish - rejuvinate a history of developer advocacy learn elixir - build a couple of projects from a couple of angles learn machine learning basics
Best part of CodepaLOUsa so far… sitting next to Douglas Crockford at dinner tonight. Occasionally I’m the lucky one. (Hope I didn’t bore him too much)
Anyway – having a fun time at CodepaLOUsa and hope all other attendees are too.
http://codepalousa.com/schedule