Off-by-none: Issue #354

February 24, 2026

Context Engineering is the New Moat 🛡️

In our previous issue, Anthropic leveled up with Sonnet 4.6 and a monster fundraise, AWS sharpened its agent tooling, and serverless databases continued their slow march toward maturity. This week, DSQL inches toward "new default" status, Claude keeps launching startup killers, and FinOps consultants start turning their advice into software. Plus, we have plenty of awesome content from the serverless and cloud community.

News & Announcements

Aurora DSQL dropped some new Go, Python, and Node.js connectors that simplify IAM authentication, which is a pretty big deal if you’ve wasted time with token generation and connection management in serverless environments. Cleaner auth flows plus first-party connectors go a long way toward making DSQL feel less like an experiment and more like the new default.

Aurora DSQL also now integrates with Kiro powers and AI agent skills, so you can build Aurora DSQL-backed applications faster with AI agent-assisted development. If you’re fully bought into Kiro, this is great for you! But if you're still skeptical like me, the Aurora DSQL skill also extends the same capabilities to pretty much any of the other AI coding agents as well.

Anthropic keeps shipping startup killers. They introduced Cowork and finance plugins, pushing Claude further into highly complex domain-specific workflows. And with Preview, Review, and Merge in Claude Code, they’re inching closer to making AI a native part of the pull request lifecycle. Review code and monitor PRs without leaving the app.

On the FinOps front, long-time AWS cost consultant, and perpetual thorn in AWS's side, Corey Quinn, announced that his company Duckbill is expanding into software, raising $7.75M for its new Skyway platform. As cloud costs get more intertwined with AI workloads, expect more advisory shops to turn their playbooks into products. There aren't many moats left.

Tutorials

• From Zero to Agentic Coding: Running Claude Code with Amazon Bedrock by Gunnar Grosch
• Implementing protected Lambda function URLs in user-facing applications by Arpad Toth
• Automatically decompress files in Amazon S3 using AWS Step Functions by Sandeep Mishra, Ali Syed, and Santosh Jade
• Building long-running usage reports on AWS by Daniele Frasca
• Scaling ECS Fargate like Lambda by Rehan van der Merwe
• A Serverless Recipe Assistant with Bedrock AgentCore, Knowledge Bases, and S3 Vectors by Darryl Ruggles
• DynamoDB E-Commerce Orders Schema Pattern
• Building Reliable Distributed Systems with AWS Serverless by Walter Fernández
• Building the AWS Serverless Power for Kiro by Gunnar Grosch
• Understanding Lambda Tenant Isolation by Pubudu Jayawardana
• How to Build a Serverless AI Agent with Amazon Bedrock and Lambda by Sachin M
• AWS Lambda Durable Functions vs Step Functions: a real-world comparison by Davide De Sio

Reads

Claude Built My Wix Website in 3 Hours - Is SaaS Dead?
Ran Isenberg rebuilt his website using Claude Code in only 3 hours, applying AWS's AI-SDLC methodology for structured development. His conclusion is that while building software has become dramatically easier, the real challenge remains in maintaining, operating, and supporting it over time. This seems to be the conclusion most people (sans executive leadership) are coming to.

Vibe Coding Got Me a POC in Days. Production Broke Every Assumption I Had.
Great post by Luca Bianchi on the reality of AI-assisted development. His distinction between vibe coding and real engineering should resonate. AI reduces the cost of change but doesn't eliminate the need for disciplined practices. If anything, you need deeper expertise to orchestrate these tools effectively.

The Software Development Lifecycle Is Dead
Boris Tane explores how AI agents are collapsing traditional software development stages (requirements, design, implementation, testing, review, deployment) into a continuous loop where context engineering matters more than process management. Worth reading if you're thinking about how AI will reshape your development workflows.

How we rebuilt Next.js with AI in one week
Steve Faulkner shares how Cloudflare rebuilt Next.js from scratch in a week using AI, creating vinext with 4x faster builds and 94% API compatibility. The fact that one engineer and Claude Code pulled something like this off for $1,100 in tokens seems to be becoming a very common story.

We Ralph Wiggumed WebStreams to make them 10x faster
Vercel's engineering team built a WebStreams library that achieves 10x+ speedups by eliminating Promise-per-chunk overhead, and they're contributing it upstream to Node.js. I love seeing performance work like this, especially when teams share the optimizations with the broader ecosystem instead of keeping them proprietary.

Deep Dive into AWS Lambda: Official Node.js 24.x Runtime Analysis
Great post that analyzes the Lambda Node.js 24.x runtime source code. This kind of deep dive is super useful because understanding how Lambda actually bootstraps and manages the execution environment helps you write better functions (and debug the weird edge cases).

Context Engineering for Commercial Agent Systems
Context isn’t just what you pass into a model. It’s infrastructure. It defines your isolation boundary, your cost surface, your audit trail, and your upgrade path. This guide isn’t meant to be canon. It’s a distillation of the patterns I’ve seen actually hold up in production.

Podcasts, Videos, and more

AI for content creators | Serverless Office Hours
AWS Serverless Hero Allen Helton and Community Builder Andres Moreno walk through their AI-powered content repurposing platform. The architecture uses Step Functions to coordinate MediaConvert, FFmpeg, and Lambda for extracting quotes and generating shareable content from videos. Very cool.

Cut .NET Lambda Response Times to 300ms with These Steps
James Eastham explains three optimization strategies for .NET Lambda functions: memory allocation, initialization patterns, and native AOT compilation, including before-and-after benchmarks showing how to achieve 300ms cold start times.

Head of Claude Code: What happens after coding is solved | Boris Cherny
Fascinating interview with Boris Cherny on how Claude Code evolved from a terminal prototype to generating 4% of public GitHub commits. I mostly agree that "coding has been solved", because I can't remember the last time I actually wrote a line of code, yet I've produced hundreds of thousands of lines. As he says, though, great engineers are always going to be necessary because coding is just a small part of the software engineering craft.

New from AWS

• Amazon announces generative AI-based artifacts in Amazon Q Developer for visualizing resource and cost data
• Amazon Redshift Serverless introduces 3-year Serverless Reservations
• Amazon Managed Grafana now supports AWS KMS customer managed keys
• Amazon SNS now supports sending SMS in additional AWS Regions

Thoughts from Social

On two separate occasions Amazon’s Kiro AI assistant caused an AWS outage, one that was 13 hours long. Amazon blames this on “user error not AI error,” which is one of the most embarrassing things you could ever say as a human being https://t.co/SlEJzqYDhH pic.twitter.com/LHFRFPQdPN

— Ed Zitron (@edzitron) February 20, 2026

According to Ed, Amazon's Kiro AI assistant caused two major AWS outages, including one lasting 13 hours. Responding with "this was user error, not AI error," sounds exactly like what an AI bot would say. 🤔

Developer Tools

Code Mode: give agents an entire API in 1,000 tokens by Matt Carey
Matt Carey shares Cloudflare's new Code Mode approach. It is genuinely clever: instead of exposing 2,500+ API endpoints as individual MCP tools (consuming 1.17 million tokens), they give agents just two tools to write JavaScript that discovers and interacts with the API, reducing token usage by 99.9%.

Turning AWS Serverless Experience into a Claude Code Plugin by Gunnar Grosch
Gunnar Grosch built an AWS Serverless Plugin for Claude Code that includes skill guides, the AWS Serverless MCP Server for tool access, and validation hooks. It covers Lambda, DynamoDB, Kinesis, EventBridge, Step Functions, and provides guidance on configurations, error handling, and observability patterns for production use. This should come in very handy.

Final Thoughts 🤔

I spent most of last week and this past weekend writing my guide on Context Engineering for Commercial Agent Systems. It is truly fascinating to see all the work going into responsibly steering AI agents. The OpenClaw moment is also mind-blowing. I do worry a bit about the chaos that unrestricted agents could bring, but there is some real solid engineering going on right now that will deliver a tremendous amount of value.

See you next week,
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.