Seroter's Daily Reading — #752 (March 30, 2026)

By Seroter's Daily Reading March 31, 2026

Audio summary of Richard Seroter's daily reading list #752: multi-agent coding orchestration, overburdening engaged employees, ADK for Java, BigQuery semantic matching, Gemini memory imports, blogging, AI literacy, Flutter + Firebase, agentic databases, MedGemma, Gemma-SRE, Waymo growth, and MLB Scout Insights.

Seroter’s Daily Reading #752

🎧 Listen to this episode (12 min)

Welcome to Seroter’s Daily Reading, number 752, for March 30th, 2026. Today’s list has a clear through-line: the world is moving from single-player AI to multi-agent orchestration, and everything from how we write code to how we build databases to how baseball fans watch games is shifting as a result.

We start with Addy Osmani’s write-up from his O’Reilly AI CodeCon talk, called The Code Agent Orchestra. His core thesis is that the era of pairing with a single AI assistant is over. The ceiling on what one agent can do — limited context window, no specialization, no coordination between tasks — means productive developers are now orchestrating teams of agents working asynchronously in parallel. He lays out an eight-level maturity model for AI-assisted coding and argues most developers are stuck at level three or four. The orchestration tier starts at level six, and it requires fundamentally different skills: clear specs, work decomposition, output verification. He walks through concrete patterns — subagents for focused delegation, hierarchical teams of teams, and full agent teams with shared task lists and quality gates. The key insight is that managing AI agents is starting to look a lot like managing a real engineering team.

Speaking of managing teams, HBR has a piece asking Are You Overburdening Your Most Engaged Employees? The research is clear: engaged employees are more productive and less likely to leave, which is exactly why companies invest heavily in engagement programs. But there’s a trap. Managers tend to pile more and more work onto the people who are most willing and capable, precisely because they say yes and deliver. Over time, that enthusiastic team member hits their limits, and you risk burning out the people you can least afford to lose.

Google’s Developer Blog announced ADK for Java 1.0.0 — that’s their Agent Development Kit. The release includes new grounding tools like Google Maps and URL context fetching, a centralized plugin architecture for global execution control, enhanced context engineering with event compaction, human-in-the-loop confirmation workflows, and native support for the Agent-to-Agent (A2A) protocol so agents built in different frameworks can collaborate.

Next, a clever BigQuery trick from Jeff Nelson: How I Finally Killed My Massive BigQuery CASE Statements. BigQuery has made AI.EMBED and AI.SIMILARITY generally available, and these functions let you do semantic matching right in your SQL queries. Instead of thousands of lines of CASE statements and regex for fuzzy string matching, you compare meaning using embeddings and cosine similarity. Great for mid-size queries and ad-hoc data cleaning, though you’d still want pre-computed embeddings for billion-row workloads.

Google announced that Gemini now lets you import your AI memories and full chat history from other AI apps. The memory import works through a prompt-and-paste workflow, and for chat history you can upload a ZIP file. They’re also renaming “past chats” to “memory.” A smart move to lower switching costs.

Nikola Breznjak wrote a love letter to blogging: Why You Should Start Blogging (Even If Nobody Will Read It). The argument: start now. Blogging sharpens your thinking, makes you better at the topics you teach, and creates a searchable external brain. His practical advice: write one hundred words per day.

Google Cloud shared research on how UC Berkeley students use AI as a learning partner. Students call AI a “tutor,” not an assistant. They use it metacognitively — identifying knowledge gaps, clarifying concepts — while building deliberate guardrails against overdependence. One student specifically warned against “vibe coding.”

A comprehensive guide to Using Flutter with Firebase walks through the full backend stack mobile developers need. The author has used Firebase in over fifteen Flutter projects and shares hard-won lessons from setup to production patterns.

Databricks published a piece on how agentic software development will change databases. Their telemetry shows agents creating four times more databases than humans. Three requirements emerge: instant branching, serverless scale-to-zero, and seamless elastic growth. They also argue open-source ecosystems like Postgres have a structural advantage because LLMs were trained on their documentation.

Google announced the winners of the MedGemma Impact Challenge, where over 850 teams built healthcare applications using MedGemma. Winning projects include EpiCast for disease surveillance in West Africa, FieldScreen AI for on-device TB screening, and CaseTwin for accelerating hospital referrals with historical case matching.

Gemma-SRE is a self-hosted vLLM infrastructure agent using Google’s open Gemma model. It’s an MCP server giving an LLM tools to analyze cloud logs, suggest remediation plans, and manage its own inference infrastructure — all without sending sensitive data to external APIs.

Waymo hit 500,000 paid robotaxi rides per week across ten US cities — a tenfold increase from May 2024. They’ve expanded to ten cities while keeping roughly the same 3,000-vehicle fleet, meaning dramatically better utilization per car. Still a tiny fraction of Uber’s 13.5 billion annual trips, but the growth curve is undeniable.

Finally, Major League Baseball is using Gemini to power Scout Insights in their Gameday app. The feature delivers contextual color commentary during live games by scanning all of baseball’s historical data. To solve the latency problem, they pre-generate insights based on lineups and match them in two seconds. They developed a concept called “surprisal” — a mathematical measure of what makes an insight genuinely interesting versus just accurate.

Today’s links paint a picture of a world where AI agents are becoming the default builders — of code, of databases, of healthcare tools, of sports commentary. The humans who thrive will be the ones who learn to orchestrate rather than just operate.

Source: Daily Reading List – March 30, 2026 (#752) by Richard Seroter

Articles covered: