Agent Payments Protocol Joins FIDO to Boost Secure Standards
We’re donating Agent Payments Protocol to the FIDO Alliance to support the future of secure, agentic payments. https://t.co/USTxadi1Cn < great to see this. Companies are quickly coalescing around solid open standards.
AI Shouldn't Replace OODA's Observe, Orient, Decide Stages
Just because AI makes the "act" stage easy to execute, doesn't mean we should skip the Observe-Orient-Decide stages of the OODA loop. In fact, you're making a mistake if you do. Good post ... https://t.co/5DuJ03LiFo https://t.co/VpkQ3TWuPD
AI Adoption Stalls in CI/CD Over Trust and Integration Gaps
Why is AI adoption inside the CI/CD workflow lagging? Is it a trust issue, as called out here? Or are the products (or integration points) lacking what today's AI apps and teams demand? I don't know. https://t.co/edozQzlVba
Only Company Offering Integrated Chips, Models, Data & Agents
"We’re the only ones with chips, models, the context to feed the models from all of the data infrastructure, the cyber tools, and then a world-class agent platform." - @ThomasOrTK Great interview, as usual, from @benthompson. So many insights. https://t.co/xmatA6xk5f
Microsoft Backs Agent-to-Agent Protocol in New Framework
Glad to see Microsoft betting on the Agent-to-Agent Protocol for their v1 Agent Framework (Python/C#) ... https://t.co/3R39hl934s
Google Photos Now Lets You View Photos From New Angles
Unlike many of you, I don't have a "best side" for photos. They're all average. But this new @GoogleResearch capability added to Google Photos lets you reimagine images from slightly different angles. How cool. https://t.co/9PuGSyGeJX https://t.co/sOrqxaTAy7
Agents Are Going Asynchronous—Now Standard for All Users
All your agents are going async https://t.co/Lu3V39bLFC < you already know this (or do this) if you're building agentic systems. But it's happening for those using agentic tools.
Control Beats Feature Count When Scaling Agents
"But if your next five years look like 'we're going to deploy hundreds of agents across our business,' the vendor that gives you architectural control matters more than the vendor with the most services." https://t.co/5JuwZWUDeH
Memory‑Driven Experience Scaling Turns LLMs Into Continuous Learners
"ReasoningBank provides a powerful framework for enabling LLMs to learn from experiences and evolve into continuous learners during test-time. We believe memory-driven experience scaling represents a crucial new frontier for agent scaling." https://t.co/PxzliIQm3J < new research https://t.co/he5xps4LdY
Gemini Models Handle 6+ Trillion Tokens Monthly via ADK
"More than six trillion tokens are processed monthly on Gemini models through ADK." https://t.co/OcL9RxiiKq < quite the little nugget snuck into this broader post about the new @googlecloud Gemini Enterprise Agent Platform. ADK 2.0 looks pretty great.
Dataplex Becomes a Dynamic, Always‑on Enterprise Knowledge Catalog
"To address this problem, we are evolving Dataplex into a dynamic, always-on Knowledge Catalog that serves as the universal context engine for your enterprise, helping agents execute complex tasks with accuracy." https://t.co/C1akH2b5yu < I'm listening.
Open‑source Marathon Planning System with Multi‑Agent Architecture
That gorgeous marathon planning system we showed off on stage yesterday at #GoogleCloudNext? The one that shows a coordinated multi-agent architecture with memory and MCPs? We open sourced all of it. Run it locally or on the cloud. Party on. https://t.co/MeDZLkI3ww https://t.co/1qHbupJp2d
Tokenmaxxing Rewards AI Vendors, Harms Broader Ecosystem
"Tokenmaxxing: great for AI vendors, bad for everyone else" - @GergelyOrosz You get the behavior you incentivize. And this particular trend runs the risk of creating the wrong behavior. Good article: https://t.co/12ZPKghi9Q
Mastering Managing Up: Purposeful Leadership Beyond Your Role
What do you think of "managing up"? Not your job? Something you do on purpose? I really liked this post from @wes_kao who explains what it's about, why it's important, and how to do it well. https://t.co/LlgRsfCQf4
Google Cloud Spanner Omni: Portable, Best‑in‑Class Database Everywhere
The best database on the internet is now available ... anywhere? We just announced @googlecloud Spanner Omni, with reimagined TrueTime and Colossus components for portability. https://t.co/iHBOKF7nH2 Pull the container here: "docker pull https://t.co/g26aVGW0aX" https://t.co/zqU7Z1g3Lp
Google Cloud Commits to Bold, Responsible AI‑first Strategy
Heckuva stat from @sundarpichai at #GoogleCloudNext. We're doing AI-first in a bold and responsible way. https://t.co/QvYbbJqolv
2026 Advances: Agent Harnesses, Sub‑Agents, LLM Fine‑Tuning
Seroter Daily Reading List – April 21, 2026 (#768): Today’s links look at agent harness engineering, working with sub agents, and how to fine-tune LLMs in 2026. https://t.co/TYPQE7sm4z
Builder Hub Unifies Google Developer Experience with Personalized Learning
10s of millions of people have Google Developer Program profiles. To me, we weren't doing enough to pull the Google dev experience together for them. So my team did something. The Builder Hub has personalized learning, project views, community, more. https://t.co/ZNo1ZtRw31 https://t.co/lcCikXTX8I
GRPO: Reinforcement Fine‑Tuning for LLMs in 2026
How to fine-tune LLMs in 2026 https://t.co/JwehrYWi3o < looks at GRPO (Group Relative Policy Optimization), which is a form of reinforcement fine-tuning https://t.co/VNtbzppDy9
AI Speeds Coding, but Deep Thinking Still Essential
"AI tools raised the floor on how fast you can produce code. They did nothing for the floor on how carefully you have to think about it ... The thinking got forgotten." https://t.co/nfQmbXYf3N
Long‑running AI Sessions Drain Compute, Causing Capacity Crunch
Why is seemingly every AI service capacity constrained right now? This post about Microsoft/GitHub pausing new Copilot signups makes the astute point that the shift from tab-completion to long running AI sessions consumes a ton more compute. https://t.co/NGxivxBIpT
Effective Harnesses Unlock Model Potential, Not Just Architecture
.@addyosmani shares hot takes: "A decent model with a great harness beats a great model with a bad harness" "The gap between what today’s models can do and what you see them doing is largely a harness gap." "A harness is a living...
Seamless Hand‑off: Stream Processing to Messaging to AI Agent
Not always, but sometimes the pieces just fit together well. I like this post that shows a clean handoff from : Real-time stream processing --> messaging engine for data transformation --> AI agent that processes the message https://t.co/wGIr7GLVtQ https://t.co/VgaydbuoUW
Per‑Seat Pricing Dies as AI Budgets Burn Fast
Great edition of What’s 🔥 in Enterprise IT/VC from @edsim this week. Per-seat pricing is dead. What happens with the annual AI (dev tools) budget is blown in months? OpenClaw trend is over already? https://t.co/HhIvqKG86l
Hire Talent Above You to Make Organization Better
"If you don’t hire people better than you, the organization gets bigger, not better. But how do you hire for something you don’t understand?" https://t.co/Uhfik3KhRg < what a terrific post from @asmartbear who gives some truly useful advice for interviews
Reining in AI Agents with Identity, Context, Persistence, Platforms
Have we given agents too much ... agency? @addyosmani calls out four architectural bets to regain control. Agents need: 🔑 Identities, not shared credentials 📚 Universal context, not scraped windows 📆 To survive your laptop closing ☁️ Platforms https://t.co/V0nEqql07a https://t.co/3lGAgHm8Uq
New Guide Shows Secure Multi‑Agent Networking Patterns
If you don't need more than one agent, keep it simple. But when you do, it's important to know the patterns and tech for secure networking. We just shipped a terrific new architecture guide that you can study, or give to an...
Headless Salesforce, Agent2Agent Anniversary, Sleep Deprivation Risks
Seroter Daily Reading List – April 17, 2026 (#766): Today’s links look at Salesforce going headless, celebrating a year of the Agent2Agent Protocol, and why be sleep-deprived is a dangerous thing. https://t.co/LAaVAL7eda
Know When You’re Building to Learn Vs. Earn
"We are building in both discovery and delivery, it’s just that we’re building with a very different purpose (to learn vs to earn), and normally using different tools and different techniques." https://t.co/uEGNWaF6Jv < make sure you know when you're building...
Canva Forces AI; Optional Use Should Remain
Canva won't make you use its new AI https://t.co/nvHtdWaBrF < it's probably blasphemy for me to say (and I'll check to see if my badge works tomorrow), but using AI should be *an* option, not the *only* one, in your...
Casual Contributors Rely on AI, Skip Safety; Maintainers Double‑check
Does this research surprise you? Casual codebase contributors use AI agents to do the heavy lifting (and skip safety checks more often) while core maintainers use agents to take on toil, while still double-checking work themselves. https://t.co/VQUivyGjju
Gemini 3.1 Flash TTS Sparks Realistic AI Voice Boom
These Gemini 3.1 Flash TTS demos I've seen floating around are so darn awesome. @glaforge built a cool one with our Gemini Interactions SDK for Java. Are we about to see an explosion in realistic AI voices? https://t.co/RFLF4TyuoU
New Open‑Source SRE Extension Boosts Gemini CLI
Want a little Site Reliability Engineering help from your agentic CLI? @palladius just open sourced a new SRE extension for the @geminicli. It's got a handful of useful skills for investigations, anomaly detection, and mitigations. https://t.co/euwVVEOK6r https://t.co/dnjWoIdlmD
Production-Grade AI Coding Skills Now Integrated Into Gemini CLI
Have you installed these production-grade engineering skills for AI coding agents, courtesy of @addyosmani? https://t.co/t38B2QZ1bZ The repo is trending on GitHub this week (https://t.co/FFtTldC7Z7) with 16k stars. I just added them to my @geminicli. https://t.co/bAQmchTY3U
Treat Dependencies as External Attack Surfaces, Upgrade Tooling
Your Dependencies Are Someone Else's Attack Surface https://t.co/zDbAITTE9f < good post with cause for optimism at the end. But we need to apply better skills/services to agent tooling to stay diligent when building
New Multi‑Region Endpoint Boosts Availability, Preserves Data Residency
You want @AnthropicAI Claude Opus 4.7 served from a single @googlecloud region? Got it. Oh, you want the global endpoint for max availability? Do that. What's new is the multi-region endpoint with more availability, but respecting data residency. https://t.co/CSUJhvOCMX
ADK Skills Streamline Python Agent Development
I should have used our ADK skills last weekend when I built some Python agents. Next time. This post explains what these dev skills offer, and why it makes it simpler for agent builders. https://t.co/qtGb3NZetk https://t.co/tIz6tZccLP
Quality Engineering Returns: Balance AI Trust and Review
"I started this article by talking about a false choice: either we surrender our judgment to the AI, or get stuck reviewing every line of code it writes." https://t.co/3RYvR9XCVJ < doesn't have to be one or the other. Quality engineering...
Trust Inputs, Not Outputs, when Replicating Team Success
Why can't you seem to land those proven practices from your old team, with your new team? Maybe you're mistaking the outputs of trust with the inputs. Really liked this perspective on why great teams can't be copied ... https://t.co/LARrCS5pbn https://t.co/h3Z9H6Ve6O
Use Google Cloud Vertex AI for Fast Anthropic Models
If you're prioritizing latency and throughput for @AnthropicAI models (https://t.co/V3922AstdD), you probably choose @googlecloud Vertex AI. Here's a guide for setting up Claude Code with Anthropic models on Vertex: https://t.co/HpeF994xXd
AI Outpaces Teams in Bug Detection; Adapt Response
AI discovers bugs faster than teams can respond https://t.co/2YNTbiuIjn < call me crazy, but I don't think the solution will be "stop finding bugs." We need to change how we respond to the valid influx.
All Top AI Models, Fully Managed on Google Cloud
All the best models in one place. We just lit up @grok 4.20 and Grok 4.1 Fast as fully-managed APIs on @googlecloud. Pay as you go, no infrastructure to provision or manage. Party on. https://t.co/UsUQzqy8sv https://t.co/Zi4ooO3471
Novelty Drives Procrastination; Use Productive Distractions Wisely
Our brains like novelty, and protect us from negative emotions. Is this what makes us procrastinate on work we know we HAVE to do, but keep putting off? Good post on productive procrastination, and how to combat it: https://t.co/cQJPPLLPb6 https://t.co/7jta7dFLjN
Product Discovery Decides Success in AI-Driven Markets
"What is increasingly clear is that for a commercial product, especially in the AI era, product discovery is usually the difference between success and failure." https://t.co/ABiyJn6UUQ < internal products have a captive audience, where commercial products need a diff approach
Structure Content for AI Agents with Addy’s Checklist
Is your content structured correctly for AI agents slurping up info? Or is key info getting truncated because pages are too chunky with the best material at the end? @addyosmani offers a checklist and deep advice for agentic engine optimization: https://t.co/D5wGWcAlb8...
Nearly Half of AI-Generated Code Fails in Production
43% of AI-generated code changes need debugging in production, survey finds https://t.co/OoNEi3VQ03 < maybe all that unattended agent-generated code going to prod will work out fine. Or, it won't.
Google Cloud Pub/Sub Now Triggers LLMs on Data Streams
Yesterday, I look a look at how @GoogleCloudTech Pub/Sub now makes it easy to invoke an LLM for a data stream. https://t.co/B3CIe7Ugjl @iRomin did a better job of covering this feature in his post last week: https://t.co/VwAd4ggamI
Google ML Engineer Cert: Top Value Among AI Credentials
"This [Google Professional Machine Learning Engineer] certification is widely considered one of the highest-value credentials available for professionals working with cloud-based machine learning systems." https://t.co/goCPBTff63 < do AI certs matter? Which ones are legit?
Go's AI Ecosystem Thrives with Genkit and ADK
Go for AI agents: a field report https://t.co/Zqwe1LkpX2 < useful look, but I'd disagree that the AI ecosystem for Go doesn't exist. @GenkitFramework Go is mature and makes it easy to add AI to Go apps: https://t.co/88HQmvWkdq And ADK Go is...
Composable AI Coding Stack Emerges over Single Tool
Is a new composable AI coding stack taking shape? @janakiramm thinks so. Instead of one tool to rule them all, a different stack is showing up dev environments. https://t.co/cXmrMLuPur https://t.co/2vSOzbBKbI