Off-by-none: Issue #353

February 17, 2026

The Claw is the Law 🦞

In our previous issue, Claude got even smarter, Bedrock made agents more production-friendly, and the future of SaaS got a little more uncomfortable. This week, Anthropic levels up with Sonnet 4.6 and a monster raise, AWS sharpens its agent tooling, and serverless databases continue their slow march toward table stakes. Plus, we have plenty of serverless and cloud content from the amazing community.

News & Announcements

If you've been letting Peter Steinberger run agents on your Mac mini, then you probably already know that OpenAI acquired OpenClaw. The speed at which OpenClaw gained adoption is mind-boggling, further proving how quickly the AI ecosystem is evolving. This article from VentureBeat is a good read if you want to get more of the backstory, like why Anthropic didn't want to be his Valentine. 💔

Speaking of Anthropic, they wasted no time keeping their own momentum going with the release of Sonnet 4.6, continuing the steady drumbeat of faster, cheaper, more capable models. And, as we’ve come to expect, it quickly showed up in Amazon Bedrock. The model wars aren’t about exclusivity anymore. They’re about distribution and operationalization. I still think Anthropic has the edge here.

Anthropic also announced improved web search with dynamic filtering, pushing retrieval closer to something that feels intentional rather than bolted on. Plus, they announced a $30B Series G raise at a $380B valuation, despite the missed opportunity with ClawdBot, er, I mean OpenAIClaw. 🙃

On the AWS side, Bedrock keeps filling in the gaps required to run agents like real systems. The AgentCore Browser now supports proxy configuration, critical for enterprises that live behind corporate firewalls. Bedrock also added support for six fully-managed open weights models powered by Project Mantle, a new distributed inference engine for large-scale machine learning models.

In serverless database land, Amazon Aurora DSQL added identity columns and sequence objects. Seems like something that should have existed on day 0, but better late than never. Maybe we'll get foreign key constraints next?

Outside AWS, Cloudflare introduced Markdown for Agents, which is a subtle but meaningful step toward standardizing how agents consume and interpret structured content. Mistral AI made its first acquisition, buying serverless infra startup Koyeb to bolster its cloud ambitions. This is another signal that model companies increasingly want to control more of the stack. And if you’re working in event-driven architectures, EventCatalog rolled out some major visualizer improvements to help teams actually see what they’ve built.

Tutorials

• Integrating Lambda Durable Functions into a Step Functions Workflow by Monica Colangelo
• AWS Lambda Durable Functions - Build a Loan Approval Workflow with Checkpoints, Callbacks, and a React Frontend by Darryl Ruggles
• AWS Lambda Managed Instances with Java 25 and AWS SAM - Part 3 Create Lambda function with LMI compute type by Vadym Kazulkin
• Streaming LLM Responses Through API Gateway REST with Lambda by Pablo Albaladejo
• The Middy After Hook Problem: Why Streaming Lambda Observability Is Broken by Pablo Albaladejo
• Deferred Observability for Streaming Lambda with TransformStream flush() by Pablo Albaladejo
• Testing in Serverless: TDD and Serverless at Scale by Lucas Geovani Castro Brogni
• Content-Based Router Pattern by David Boyne
• Aurora DSQL: The Serverless PostgreSQL That Scales to Zero (Should You Migrate?) by Dinesh Kumar Elumalai
• Nevermore.dev: LLM-as-judge on Lambda Durable Functions by Davide De Sio
• DynamoDB On-Demand is Bleeding You Dry: Switch to Provisioned and Save 60% by Suhas Mallesh
• Things I Wish I Knew Before I Started Using DynamoDB by Sathiesh Veera

Reads

The AWS Cognito Race Condition That SDK Retries Can’t Fix
Basel Issmail walks through a race condition in AWS Cognito where parallel Lambda functions performing read-modify-write operations on a User Pool Client collide hard. Cognito enforces single-writer semantics with no version tokens or ETags, so the solution requires retrying the full operation cycle with fresh reads, not just the failed request. I haven't used Cognito in a while, but I guess there's still some rough edges.

From Kong-on-K8s to AWS Serverless in 24 Hours: Building a Production‑Grade API Platform
Druhin Dhavala shows us a migration from Kong-on-Kubernetes to API Gateway v2 and Lambda, hitting 120ms p95 latency and $0.01/day in operating costs. If you're evaluating serverless for API workloads, the benchmarks and open-source infrastructure code here are worth a look.

A Bootstrapper's Diary
David Boyne, EventCatalog's founder, is publicly documenting his path from solo bootstrapper to $1M ARR, now 14 months in and 10 weekly entries deep. Each note covers real experiences like pricing decisions, early sales milestones, and hard lessons. If you're interested in the mechanics of open-source monetization, this is a useful resource to follow.

Claude's new constitution
Anthropic has published the full guiding document used to train its Claude models, covering four core priorities: being broadly safe, broadly ethical, compliant with Anthropic's guidelines, and genuinely helpful. Worth reading if you want to understand how a major AI lab is operationalizing values at the training level, not just as post-hoc policy.

Podcasts, Videos, and more

Observability for Distributed Systems | Serverless Office Hours
Joe Alioto and Eric Johnson use a retail application example to walk through request tracking, performance optimization, and error correlation across AWS serverless services.

AI & Java | Serverless Office Hours
James Ward joins Julian Wood to guide Java developers through AI integration using MCP, agent-based architectures with Bedrock AgentCore, and GraalVM optimizations for AI workloads. If you're a Java developer trying to figure out where your life went wrong, this video will give you hope. 😂

Building Smarter AI Agents: Memory Management with AgentCore
Excellent post by Mike Chambers on building smarter AI agents with persistent memory via AgentCore. The progression here is solid: theory first, then a simple Strands SDK demo, then a full advanced agent with persistent memory.

Will AI replace programmers? OpenClaw creator, Peter Steinberger and Lex Fridman
Including this only because I think Peter Steinberger confirms what most developers already know, but leadership teams probably don't. 😑

Serverless Craic Ep81 AI - differentiator or commodity?
Really interesting take from The Serverless Edge crew on AI as a commodity vs. a differentiator. Their core argument lands well: most organizations should be consuming AI, not building foundational models. The actual differentiator is how clearly you understand your own value chain. That's the framing more teams need to hear before trying to hire Carnegie Mellon PHDs.

Aurora DSQL: Serverless, Scalable, Global OLTP Database System (Marc Brooker)
Speaking of Carnegie Mellon, in the second episode of PostgreSQL vs. The World, Marc Brooker joins to give a full overview of Amazon DSQL.

New from AWS

• Amazon Bedrock expands support for AWS PrivateLink
• Amazon Aurora now supports Server-Side Encryption at Rest
• Amazon MSK now supports dual-stack (IPv4 and IPv6) connectivity for existing clusters
• Amazon S3 Tables add partition and sort order definition in the CreateTable API
• Amazon MSK now supports broker logs on Express Brokers
• Amazon Bedrock reinforcement fine-tuning adds support for open-weight models with OpenAI-compatible APIs
• MSK simplifies Kafka topic management with new APIs and console integration
• Amazon Athena now supports 1-minute reservations and 4 DPU minimum capacity
• Amazon OpenSearch Serverless now supports Collection Groups
• AWS expands Resource Control Policies support to Amazon DynamoDB
• Amazon Aurora DSQL is now available in additional AWS Regions
• Amazon ECR now supports additional metrics for monitoring repositories

Developer Tools

murataslan1/ai-agent-benchmark: AI coding agents comparison by Murat Aslan
Worth a look if you're trying to make sense of the AI coding agent landscape. The repo tracks SWE-Bench scores alongside user-reported real-world performance, which often tells a very different story. Amazon Q Developer is flagged as declining and Devin AI has essentially vanished from user reports, which is useful context if you're evaluating options.

Why I Built My Own AWS Deployment Tool by Alex Dev
Alex explains the architecture behind effortless-aws, a custom deployment tool that uses ts-morph to analyze TypeScript ASTs and extract infrastructure config directly from handler code, with esbuild for bundling and Effect-TS for orchestration. Infrastructure from Code? What a novel idea. 😉

Final Thoughts 🤔

I just got back from the Oakley Icon Alliance Championships in LA where my daughter and her team competed against the top 7 high school girls flag football teams in the country. This has nothing to do with serverless, I'm just a proud dad. The weather was amazing, the guacamole and margaritas at Escuela Taqueria were even better, and most importantly, the girls had the time of their lives (my daughter's the one holding the phone).

Back to thinking about what's next for me.

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.