AI Shifts Engineering Management Focus, Not Playbook
Should tech managers and leaders run the same engineering playbook with AI in the mix? The ideas might be the same, but the focus is probably different. This @InfoQ piece looks at topics like team metrics, skills development, and guardrails. https://t.co/Tt5YEK3x5J
Keep Coding Actively, Don’t Let AI Dull Your Skills
If you're not manually coding as much anymore, how do you stay sharp as an engineer? This post says we shouldn't passively consume AI output and atrophy our skills. Rather, actively engage and do things like reading diffs and debugging errors. https://t.co/KETBZiEjVO...
Kubernetes 1.36 Launches with 70+ Upgrades, Now on GKE
Kubernetes v1.36 came out a couple of weeks ago and has over 70 enhancements. https://t.co/iMLh6dcHvg Oh, and it's already been available in @googlecloud GKE for a week. https://t.co/HjvEHc0GQh
Tech Hiring Spikes to Three-Year High, 271k New Posts
Tech job postings hit 3-year high https://t.co/NUHjpucj2y < "... more than 271,483 new job postings and more than 575,000 total active job postings." Seems like good news as companies grow their tech investment.
Will AI Ever Produce Trustworthy, Fully Autonomous Code?
What do you think? Will you ever get to "write-only code" that's never skimmed, reviewed, or tweaked by a human? What would have to be true to trust the output? That's the idea here, and it's happening in some places. https://t.co/xVaMIq7HM7
Google Cloud Cuts Cold Starts, Adds Sub‑ms Bigtable Tier
It's apparently "faster performance" Friday at @GoogleCloudTech. With faster node startup for GKE, say goodbye to cold-start latency https://t.co/NU88mzkOPj New Bigtable in-memory tier for sub-millisecond read latency https://t.co/s0GBEMEBQr https://t.co/2JdqpUcLmO
Pinecone Moves Beyond RAG to Upstream Knowledge
The half-life of a "best practice" in this industry is like 3 months. RAG? That's so 2Q 2025. Pinecone who got famous as a vector database serving RAG use cases, now pushes knowledge upstream into artifacts used by agents. https://t.co/yOFcvZmsKR
AI Disrupts SaaS Freemium, Demands Resilient Front‑Ends and AI‑Powered Coding
Seroter Daily Reading List – May 7, 2026 (#779): Today’s links look at why SaaS freemium playbooks don’t work in AI, how to design front-end systems for cloud failure, and why coding with AI agents is a baseline expectations for tech...
Avoid Circular Dependencies: Separate Monitoring From Observed Systems
Your status page or observability system probably shouldn't be using the same system it's supposed to be monitoring. Airbnb found some circular dependencies and made changes to ensure they could do monitoring reliably at scale. https://t.co/rmSAeYTZF2
Documenting Today’s Effective Agentic Coding Lessons
These are solid lessons for agentic coding, from @dbreunig. Will these be the *same* lessons in six months? Who knows, but it's good to document what works today ... https://t.co/DzTFXwI2qm
Gemini 3.1 Flash‑Lite Launches: Fast, Affordable Premium AI
Gemini 3.1 Flash-Lite is GA and ready to use. It's a fast and cost-effective way to access premium AI. https://t.co/ELGpnllkIv
AlphaEvolve Proves AI Can Create Novel, Real‑world Algorithms
"LLMs just parrot back what they've been trained on and don't create anything novel." I hear that, but the @GoogleDeepMind AlphaEvolve agent is designing advanced algorithms that are making an impact in the real world. Powerful update ... https://t.co/EwszAjFzch
GKE Pod Snapshots Eliminate Cold Starts for Massive Models
Facing a cold start when your XX billion parameter model is loading up for inference on Kubernetes? You might love the new @googlecloud GKE pod snapshots. All your app state, along with most file system and networking state, get saved and...
Prefer REST APIs; Reserve MCP/A2A for Complex Reasoning
Choosing between APIs, MCP, and Agent-to-Agent architectures https://t.co/4TnEq1Dbq6 < @lak_luster says to default to function tools that use REST APIs, and only bring in MCP or A2A when there's more reasoning involved, or dynamic use cases. https://t.co/Am9DfWv1Qy
Gate AI Features by Usage, Outcomes, and Compute
Give people a taste of your product and then reveal the paywall to access premium features. SaaS products have used that approach for a while, but it's different with AI. Instead, AI products should gate by usage intensity, outcomes, and compute. https://t.co/V7YSfrBgZo
Managers Must Get Hands‑On to Lead AI Transformation
"Therefore, one cannot drive AI transformation without being an active AI practitioner. But what does being an active AI practitioner look like for a manager?" - @dblockdotorg https://t.co/0qeplV10G6 < go hands on as managers. You might rediscover your passion for...
AI-Driven TensorFlow-to-JAX Migration Mirrors Deterministic Coding Loop
That post today about how Google did an AI-assisted code migration from TensorFlow to JAX? https://t.co/EcyofYfoPB The Planner-Orchestrator-Coder pattern we applied looks a lot like what @CTOAdvisor calls "deterministic coding in a loop." https://t.co/gkGb07wQnq https://t.co/IMxYmPjS5J
Connect Your Agent Gateway to Existing Security Providers
Is your agent gateway connected to your existing security providers? Should be! We're making a big ecosystem bet on the Agent Gateway, which is part of the @googlecloud Gemini Enterprise Agent Platform. https://t.co/8ZDnmwuzmR https://t.co/OlAxPDfC88
AI-Native Teams Prioritize Prototypes, Small Squads, Rapid Cycles
What are AI-native engineering orgs doing? ➡️ Prototypes over PRDs ➡️ Small, high-leverage squads ➡️ Faster release cycles to facilitate learning ➡️ Shortened planning horizons https://t.co/VGEvQU0lRk
Honest Feedback Loops Beat Fancy Eval Dashboards
"The teams that win the next phase of AI engineering won’t be the ones with the most elaborate eval dashboards; they’ll be the ones with the most honest feedback loops." https://t.co/yx5BRTSHCV < invest in evals, says @mjasay
Google Cloud IAM Gets Major New Features and Guardrails
All of a sudden, identity management is a vibrant and exciting space. We made a ton of @googlecloud IAM improvements lately across agent identities, gateways, guardrails, and more. Check out this recap: https://t.co/ShLXPI78Md https://t.co/C1IBJ1ylrN
Add Human‑in‑the‑Loop to Pause AI Code Generation
[blog] How to force your custom agent to stop and seek human approval https://t.co/fPu5tjkMQ4 < in which I use the new human-in-the-loop feature of the Agent Development Kit to approve AI-generated code tutorials. https://t.co/0xkrl1CKmj
Empathy Enhances Direct Communication, Not Weakness
Can you be direct without being a jerk? Is being "strategic" just code for being soft or manipulative? No. Having empathy doesn't make you weak. It gives you a better chance of actually landing your (direct) message with the recipient. Good...
Netflix Routes Millions of Requests to Thousands of Models
Wow. The Netflix ML serving platform serves hundreds of model types/versions, and around 1 million requests per second. How do they route traffic to the right model in such a large-scale service system? Deep dive post: https://t.co/1sU4Jtt5l8 https://t.co/ANPlcxqpSA
Start Small, Personalize Fast, Empower Others to Build
Some cool lessons learned from Stripe in this podcast in the @lennysan network. ✅ Don't engineer the platform right away. You can start with what you have around ✅ Quickly personalize with your context ✅ Let more people build to unblock themselves https://t.co/ful1vfbMFG
Local LLMs Offer Cost‑Effective Compute and Data Sovereignty
Run LLMs on your own hardware? This scenario didn't get me fired up for a while, but now I get it. Whether offloading token-hungry tasks to self-hosted LLMs where you only pay for compute, or satisfying unique sovereignty needs, local AI...
Newsletter as Personal Discipline, Not Just Audience Reach
I have a confession. My daily newsletter of links and commentary (https://t.co/okv3uDONrs) is mostly for me. Oh, I love that plenty of people read it. But my main goal is forcing a reading and publishing discipline. @allen_hutchison is doing it too...
Flip the Ratio: Cheap Assessment, Board‑
"The answer probably starts with flipping the ratio: making assessment as cheap as generation, paying for fixes instead of just finds, and treating supply chain security as the board-level priority it has been pretending to be."
Uber Solved AI Scaling with Centralized Gateways and Registry
What problems did Uber face when scaling AI? There wasn't a shared way of building, security was inconsistent, opaque visibility into call patterns, and discovery was unmanaged. They added centralized MCP gateways and a registry. More ... https://t.co/pVTbFSPRR9
Webhooks Deliver Real-Time Gemini API Operation Completions
"Webhooks allow the Gemini API to push real-time notifications to your server when asynchronous or Long-Running Operations (LROs) complete." That's a BFD.
Diffusion-Style Decoding Triples LLM Speed on TPUs
Supercharging LLM inference on Google TPUs: Achieving 3X speedups with diffusion-style speculative decoding https://t.co/j1QPASxHKP < some pretty amazing increases in tokens per second
Treat Skills as Programs: Architecture Drives Runtime Loading
"Skills are programs, not prompts. How the skills runtime actually loads, and why the architecture is everything." https://t.co/oojCUzACLQ < I liked this post which promoted a thoughtful approach to designing (and testing) skills
Startups Favor Flutter Over Native for Faster Scaling
Why Startups Are Choosing Flutter Over Native in 2026: A CTO’s Perspective https://t.co/RCOd4dDYln < fairly strong case here for when to use a cross-platform framework like @FlutterDev and when to go native
Git Worktrees: Essential Tool for Modern Development
Do you know what git worktrees are? Do you use them? Does your agent? @kweinmeister explains them well here, and why they're valuable to modern developers ... https://t.co/vbJXFxSgQ2
Stop Redundant Map Lookups to Save CPU
You might not notice, but this way you're accessing maps in code is wasting CPU cycles. They add up! Here's quick advice on avoiding redundant lookups ... https://t.co/uOpmIOwM5r
Agents Disrupt Database Contracts; Build Defensive Data Layers
Have agents broken the unspoken contract we had with our databases? You know, human-authored apps running deterministic code and predictable queries? @arpit_bhayani wrote a terrific post for how to create a defensively designed data layer ... https://t.co/8OOXrmkD6g
Human Infrastructure Matters as Much as AI Tools
"It is impossible to predict at this point just how AI will transform the workplace. However, one thing is certain: understanding and building the right human infrastructure will be as important as picking the right AI tools." https://t.co/sm48Z9f6lH
AI Investment Requires Systems, Not Just Tool Licenses
"The most important takeaway is that AI investment is a systems decision, not a tooling decision. Buying licenses without investing in the surrounding platform, data, and governance simply accelerates the rate at which existing dysfunction shows up in production." https://t.co/dng1pr7LZr
Fully‑managed Remote MCP Servers Accelerate Google Cloud Integration
0-to-50 in record time. We've now got dozens of fully-managed remote MCP servers that let your agents easily interact with your favorite @googlecloud services. Infra, AI, databases, ops, security, docs, Workspace, you name it. https://t.co/u7LZhEnT1B https://t.co/wt2CQODKq4
AI Productivity Gains Modest Now, Will Surge with New Models
AI productivity gains: More modest than expected https://t.co/xW6Scx6ATH < checks out, so far. But once the agentic operating model really takes hold, team structures change, and better platforms improve the build-to-prod stages, these numbers will jump.
Rapid Storage Integration with PyTorch Keeps GPUs Busy
"By integrating Rapid Storage, powered by Google’s Colossus storage architecture, directly with PyTorch via the industry-standard fsspec interface, we are enabling researchers and developers to keep their GPUs busier than ever before." https://t.co/T0oMWbNKkc https://t.co/xS9wXRcmyi
Evaluation Costs Outpace Training for Modern ML Agents
"For neural operators, ML research agents, and replication benchmarks, the ratio has flipped: a credible evaluation can cost more than training the candidate model." https://t.co/BNE4tmQS7q < @huggingface post says static eval costs were tolerable, but agent evals aren't cheap.
Managing State and Coordination for Long‑Running Agents
Long-running agents pose fresh challenges. Where and how do you persist state? Who does compute coordination and completion verification? @addyosmani wrote a fantastic deep dive that lists the patterns, solutions, and limits. A must read: https://t.co/Gg7apYKJuG https://t.co/XEBSbYnzTe
Agents Build Agents, Generative UI, 1,000 Real AI Use Cases
Seroter Daily Reading List – April 29, 2026 (#773): Today’s links look at using agents to create agents, what generative UI is actually about, and a thousand real-world AI use cases from actual companies. https://t.co/xbjkX63V0c
Workday’s Service Empire May Expose Vulnerability
"A $30B company with 10,000+ customers and a services cartel that’s arguably bigger than the product itself will not roll over quietly." https://t.co/eBzsJxBGUV < is Workday finally vulnerable? They've got an enviable footprint and new focus. All platforms need to...
Putting Payroll in the Field for Lean Operations
"We are going to become an organization that puts its payroll in the field." https://t.co/69FZ6RHusY < The future involves a lean back office. Pay people who do the work, and let agents do the work about work.
A2UI Shows How Simple Component Schemas Enable Generative UI
I'm somewhat embarrassed to admit that the whole "generative UI" thing hasn't clicked with me. But after reading this post, the lightbulb went off. For something like A2UI, you give the agent the components and a schema, and it does the...
Winning Agentic AI Requires Clean Data, Rigor, Not Flashy Bots
"If you’re trying to pick winners in agentic AI, don’t look for those with the cleverest agents. Instead, look to the companies with the cleanest data contracts, the best evaluation discipline, the most coherent identity model, and the least tolerance...
On‑Device AI Thrives with LiteRT and NPU
Building real-world on-device AI with LiteRT and NPU https://t.co/8PbZfxIKiW < you can do some pretty amazing things with on-device AI right now. Here are some real-world examples.
Clarifying AI Agent Memory: Files, Blocks, or Services?
Where can you store "memory" for an AI agent? This post calls out files, memory blocks, and skill as core options. Would memory services (like the one in our Agent Platform) be categorized as a memory block? Or is that another type...