Off-by-none: Issue #352

February 10, 2026

SaaS is Dead! Long Live SaaS! 🧟‍♂️

In our previous issue, AWS finally added account names in the console, DynamoDB let you cross (account) streams, and Databricks launched a new serverless database. This week, Claude gets even smarter, Bedrock makes agents more production-friendly, and the future of SaaS gets a little more uncomfortable. Plus, we have lots of awesome content from the cloud and serverless communities.

News & Announcements

Anthropic kicked it up a notch last week with the release of Claude Opus 4.6, continuing its steady march toward replacing white-collar workers. As expected, it didn’t take long for the ecosystem to follow: Claude Opus 4.6 is now available in Amazon Bedrock and Google is adding it to Vertex AI. I've been using Opus 4.6 for the last several days and it is beyond scary good.

On the Bedrock side, AWS continues to focus on making agents feel more production-friendly. The AgentCore Browser now supports browser profiles, which is a subtle but important step toward more realistic, stateful automation. AWS also enabled structured outputs in Amazon Bedrock, making it easier to treat model responses as first-class inputs to downstream systems instead of relying on fragile parsing methods and playing whack-a-mole with sometimes-JSON outputs.

There were some notable infrastructure updates as well. Amazon DynamoDB global tables now support replication across multiple AWS accounts, which removes a long-standing limitation for multi-account and multi-tenant architectures. And for the .NET crowd, .NET 10 is now available on AWS Lambda as both a managed runtime and base image, continuing Lambda’s fast turnaround on new language releases. If you missed anything else on the AWS serverless front, the Serverless ICYMI for Q4 2025 will catch you up.

And finally, Databricks continues to dominate headlines. The company raised another $5B at a reported $134B valuation, and its CEO stirred the pot by arguing that SaaS isn’t dead, but AI will soon make it irrelevant. Hyperbole aside, it’s a clear signal that people are rethinking what “applications” even mean in an agent-first, AI-driven world.

Tutorials

• Let an AI Agent Do Your Job Searching by Danielle Heberling
• The anti-polling pattern for Step Functions by Yan Cui
• Your API Calls the Same Lambda 1,000 Times a Minute (Enable Caching Already) by Suhas Mallesh
• Building a Dynamic Bedrock Model Selection Serverless Application by Arpad Toth
• Beyond the 29-Second Limit: 4 Patterns for Serverless GenAI on AWS by Sabarish Sathasivan
• AWS Lambda Managed Instances with Java 25 and AWS SAM - Part 1 and Part 2 by Vadym Kazulkin
• Lambda Durable Functions: Finally, Stateful Serverless Without Step Functions by Dinesh Kumar Elumalai
• I Finally Don’t Need Step Functions for Everything — Lambda Durable Functions Changed That by Danylo Mikula
• 5 Years of Cloud Run: What I’d Tell Myself on Day 1 by Sara Ford
• Building fault-tolerant applications with AWS Lambda durable functions by Rahul Pisal and Ben Freiberg
• Build a serverless AI Gateway architecture with AWS AppSync Events | Artificial Intelligence by Archie Cowan
• Send EventBridge Messages to WebSocket Clients with AppSync Events by Zac Charles
• Amazon Cognito Custom Email Lambda Trigger by Lee Gilmore

Reads

Context Is Now a First-Class Architectural Concern
In AI systems, cost is driven by non-deterministic intent, not infrastructure. Context is the new economic control surface. The question 'what decision caused this work to happen?' is the type of thinking we should be doing when token usage and agent recursion dominate the bill instead of servers and resource tags.

You Are Here
Great post by Marc Brooker about what comes next now that writing code is essentially free. There's still joy in programming (just like there's joy in woodworking or knitting), and the real opportunities still lie in building systems and solving those messy, real-world problems that AI can't just generate away.

When AWS Lambda Versioning Was Not Enough
Simple but interesting postmortem about when Lambda versioning breaks down. I totally agree that serverless doesn't remove architectural responsibility. The fix (separate functions for architectural changes, versions for logic changes) might not be obvious until you've been burned by this.

A $2.3M Deal, a Six-Week Deadline, and the Serverless Architecture That Saved Us
Dinesh Kumar Elumalai explains how serverless can actually reduce costs while improving security. Completing a SOC 2 audit at 76% lower cost than traditional infrastructure (while delivering faster) is a pretty compelling argument for letting AWS handle more of the complexity.

AWS Serverless: Still the Boring Correct Choice
Extremely detailed post by Oleg Pustovit on why AWS serverless still makes sense for AI startups. He's helped 3 companies migrate from Vercel/Cloudflare to Lambda when they hit platform limits around background jobs and webhooks. The pay-per-use model for variable AI workloads is hard to beat, and I agree that improved tooling has solved a lot of the early pain points.

The EventBridge-First Pattern
Jonas Wolffsen explains why EventBridge should be your architectural boundary, not just another messaging tool. The idea that SNS/SQS swaps API-based coupling for message-based coupling is spot on, and treating EventBridge as the system contract (publishing facts, not commands) is the right way to think about it.

Step Functions vs. Lambda Durable Functions on AWS: A decision guide
Olorundara Komolafe compares AWS's new Durable Functions to Step Functions. The decision checklist is particularly useful, positioning Durable Functions as code-first orchestration for app development while Step Functions handles cross-service coordination with all its native integrations. This is a practical guide worth referencing.

Podcasts, Videos, and more

Stop Writing Unmaintainable Lambda Functions in .NET
James Eastham demonstrates how to refactor .NET Lambda functions into maintainable code. He walks through practical examples using Lambda Annotations framework and hexagonal architecture to create thin handlers that are actually testable. I'm not a .NET guy, but this is the kind of real-world guidance that makes serverless debugging and iteration so much easier.

Entering 2026 - The Operational State of AI & Cloud
Great podcast with Matt Gillard and Georgia Smith talking to Allen Helton about AWS's 20% growth and its new Anthropic partnership. I'd say that AWS has closed the gap on AI, but still hasn't caught up completely. But I definitely agree that the Bedrock/Claude combination is proving to be a genuine competitive advantage.

AWS Bites #152: Exploring Lambda Durable Functions
Excellent breakdown by Eoin and Luciano on Lambda Durable Functions. The ability to checkpoint and resume multi-step workflows without Step Functions is genuinely useful, especially for those orchestration patterns where Lambda's 15-minute limit becomes a constraint. I like the direction, but this does change the way we need to think about function boundaries.

Serverless Craic Ep 80: AI Myths in Software Engineering
Great discussion from Dave Anderson, Mark McCann, and Michael O'Reilly on why engineering skills remain critical in the AI era. I totally agree that AI tools might actually increase demand for engineers rather than replace us, especially when you factor in domain expertise and quality standards.

Building Agentic AI systems with AWS Serverless
Recent GOTO talk by Uma Ramadoss on building agentic AI systems with serverless that goes beyond the typical prompt engineering discussions to cover the practical architecture considerations for enterprise systems using Lambda, Step Functions, and Bedrock.

AWS Step Functions Local Testing | Serverless Office Hours
Peter Smith joins Eric Johnson and the AWS crew to discuss the new local testing features for AWS Step Functions, including the TestState API, its integration with LocalStack, and its benefits for automated unit testing and workflow development.

Serverless & Agentic AI: Better Together • Prashanth HN
Really interesting breakdown of agentic RAG and swarm patterns in this GOTO session. Prashanth does a solid job explaining how agents are becoming fundamental building blocks between data and LLMs, and using serverless for something like this just makes sense.

New from AWS

• Amazon Redshift now supports allocating extra compute for automatic optimizations
• AWS Config now supports 30 new resource types
• AWS Glue launches native REST API connector for universal data integration
• Amazon ECS adds Network Load Balancer support for Linear and Canary deployments
• AWS Builder ID now supports Sign in with Apple
• Amazon RDS now provides an enhanced console experience to connect to a database
• Amazon CloudFront announces mutual TLS support for origins

Developer Tools

DynamoDB Workflows: stop rewriting incident scripts (with DynamoLens)
DynamoLens is an open-source desktop client for DynamoDB that introduces workflows for multi-step operations. Instead of maintaining scattered incident and debug scripts, you can build reusable query chains where values automatically pass between Query, Scan, and PartiQL steps, then save and share them with your team.

Final Thoughts 🤔

Hope you all had a good week.

On a personal note, I wrapped up my time at CloudZero recently as part of a broader organizational restructuring. I’m extremely proud of the work we did and grateful for the people I had the chance to work with.

I’m taking a short beat, then turning my attention back to writing, research, and building. If you’re working on hard problems around AI systems, data, cloud economics, or building and operating products in the cloud, I’d love to talk.

Take care,
Jeremy

I hope you enjoyed this newsletter. We're always looking for ideas and feedback to make it better and more inclusive, so please feel free to reach out to me via Bluesky, LinkedIn, X, or email.