New article is live! It's about how I migrated my Agentic Strands-based bot to use the new first-class MCP Tool interface that Strands just shipped. It lets the bot register all MCPs concurrently and saves about 10 seconds on each response with lots of MCPs. buff.ly/HVuRfSP

Oh hey, that's pretty cool. Thanks ya'll :D <3

Didn't make it in to #RSA this coming year. I had the best time at RSA previous years. Congratulations to all the folks that did make it in! I'll have to come up with another excuse to see everyone next year :D

I published a new article about building an agentic Slack bot using Strands and AWS Bedrock. It walks through how the bot calls tools, retrieves data from internal systems, and returns useful results to users.

New post: building audit logging for multi platform AI bots with Python and CloudWatch. Bedrock logs were not enough for real audit needs, so we built our own system with consistent usernames and session level correlation across Slack and Teams. www.letsdodevops.com/p/building-a...

New post: building audit logging for multi platform AI bots with Python and CloudWatch. Bedrock logs were not enough for real audit needs, so we built our own system with consistent usernames and session level correlation across Slack and Teams. www.letsdodevops.com/p/building-a...

I don't sell Vera, but if I did, this is the type of feedback I'd use to do so :D All code is public so you can deploy it yourself (for free) buff.ly/WRCWuaR

New post: AWS Strands + MCP, super simple real-world examples. Strands makes agentic workflows easy — multi-step actions, tool use, and open-source. I’m sharing small examples as I build them out. buff.ly/cFQZ2VI

Teaching Strands Agents to assume cross-account IAM roles and read AWS resources! I wired the AWS Labs CLI MCP server into a Lambda-based agent so it can query across dev, staging, and prod — all with profile-based role assumption. Want multi-account AI agents, read this. buff.ly/kZxKLCn

All six parts of the Let’s Do DevOps “Building a Teams AI Bot” series are now free to read. This series walks through every step of creating a Microsoft Teams bot with real AI capabilities—from registration and OAuth, to message handling, to AWS integration with Bedrock and Lambda.

How are ya'll smart folks correlating the conversations agentic bots have with AI services to a particular conversation or user request? How do you do request tracing?

You know how AI-generated photos have a yellow pallor? Attached photo for reference. There's an idea that the models were trained on older photos that were scanned into computers, and the photos had yellowed with age. Tech is weird.

No one gets to tell you what you mean when you write LGTM on a PR. I personally mean Lets Go To the Moon What about you?

In Part 5 of my “Building a Teams Bot with AI Capabilities” series, we dive into the fun stuff: how the bot actually reads conversations, grabs context, and posts AI-generated responses back into Teams. Full article’s free to read, link in image #DevOps #TeamsBot #AWS #AIIntegration #LetsDoDevOps

If you've wondered how to REALLY use MCPs in your bots, me too. So as I've been experimenting and adding things to my Strands bot at work, I've been publishing all the code, including the MCPs which so far includes: * Atlassian * AWS * Azure * GitHub * PagerDuty Find it here: buff.ly/3sRrT5V

MinIO has stopped publishing their namesake object-store docker file, now requiring customers to build from source. This, with no notice, over 1B (billion) downloads from dockerhub, and just after a critical CVE was dropped. Have your teams check if they're using this image from dockerhub.

Do you know who subscribes to LetsDoDevOps[.]com? Turns out folks from: - Amazon - Intel - Hashicorp - Boeing - GitHub - SAP - Fortinet - Sandia National Laboratories - ZScaler - Anduril Please go subscribe, there's a free tier (but paid helps me buy caffeine to keep generating the content haha)

When you interact with Bedrock services, your logs get put into a cloudwatch log group. That sounds great, right? Except ALL bot interactions get put into the SAME log group, making differentiating all those logs a pain. FireHose can solve this with some custom lambda. Link in the image :D

Strands has now released my change, which permits Strands bots to utilize the github MCP to read file contents in your repos. Please upgrade to version v1.12.0+ to get the functionality :D #Strands #GitHub #AI

A coworker just said "... since we're in Q4 of the year" and I am still shocked and offended by this language.

Strands, for everyone in the world, will be able to read the contents of GitHub files next week when they publish the new version of the Strands agent. I'm proud of this change. Will be published next week to pypi!

I had a lovely time, #TechFutures. Met some wonderful people, made a room laugh at my AI jokes, and helped a small room of folks deploy a genAI bot into their slack. It's been a pretty good time.

Hey all! Parts 3 and 4 of the Teams AI Bot series on Let’s Do DevOps are now FREE FOREVER! Part 3: Building the Teams Bot AI Backend - buff.ly/hXo4RHJ Part 4: Deploying Your Serverless AI Bot - buff.ly/tAai7SN Subscribe to support more articles <3

If you're in NY this week for Tech Futures conference, let me know! I'm teaching a workshop where folks will leave with a functional private AI bot!

When you invoke an AI model in Bedrock, it writes the conversation to a single CloudTrail log group. Disambiguating the logs is left as homework for the admins (read: It work bad.). I implemented a firehose to harvest cloudtrail logs, extract the information we care about in Splunk events.

Sai and I met with Ashish to talk about the platform engineering things you should be doing around AI development (and there's a LOT!). Watch this to learn some things you can do to help fix it.

This is pretty specific, but in case anyone else has a layer7 firewall in front of their github runners, and have a line like this in your Dockerfile: FROM public.ecr[.]aws/lambda/python:3.12 Then you'll need to permit these two URLs: - d2glxqk2uabbnd.cloudfront[.]net - public.ecr[.]aws

I presented an RSA Webcast last week, and got emailed a certificate of completion afterwards, like all the other attendees. Great, now I can prove I was there

Starting Sept 29, you won't have to manually enable specific models in new accounts for AWS Bedrock. This could be seen as very positive (no more approval step) or very negative (developers can now use any model, even the ones that your company maybe doesn't want them to (see: DeepSeek).

Develop and implement a secure generative Al Slack bot with guidance from Veradigm Sr Principal Software Engineer Kyler Middleton in this #RSAC Webcast. Wednesday the September 19 @ 1:00 PM ET Register:

I couldn't find a good example of AppConfig terraform to build a multi-variant Config - one which changes based on the Context provided by the clients... so I experimented for a few hours, got it working, and submitted one. It just became live. Hope it helps :) <3

AWS Strands can't yet read GitHub file contents using the GitHub MCP, and it's all on Strands - it doesn't yet support Resources, how GitHub returns file contents. I have a write-up of what's wrong and why the Strands Maintainers don't want to merge my fix for this behavior.

The AWS Provider docs don't have a good example of multi-variant AppConfig feature flags configs, and they're provided as this flag json structure like IAM rules (but less supported, worse error messages and tools). I got it working after a few hours of testing. So I'm fixing it! buff.ly/wxKvDob

The only real Strands production use I have so far I'm calling VeraResearch, and it's an agentic bot that's reachable via Slack, and it currently only has a GitHub MCP. For instance, different teams have different gitflow. What should I do next to promote code? Well, VeraResearch can tell you!

Article 2 in my series on how to build a Teams GenAI Bot is now free forever! It covers how to use the Teams Developer Portal to register the Teams names you want your App to use. Check it out here:

Develop and implement a secure generative Al Slack bot with guidance from Veradigm Sr Principal Software Engineer Kyler Middleton in this #RSAC Webcast. September 19| 1:00 PM ET Register:

Hey all! The Strands library to write agentic bots is super easy, but it can be a little unintuitive how to consume different types of MCPs and which built-in tools are available, so I wrote some super simple examples for folks to build on. You can find them at the link!

What's RSA emailing me about now? Oh hey, that's me! I'll be presenting on how to build a secure AI assistant in a regulated industry (healthcare) in the US on Friday Sep 19! Register now to get reminded! www.rsaconference .com/library/webcast/194-secure-ai-assistant

🔥 New article just dropped on Let’s Do DevOps: AWS Strands and MCP with Simple, Real-World Examples 🔥 Strands is an open-source framework from AWS that makes it way easier to build agentic bots - think multi-step automation that can read, analyze, and act across different platforms. buff.ly/INpQIsf

I've been building Strands agentic agents that can natively use MCPs for the past two weeks or so. I was surprised that Strands can't view the contents of files on GitHub using the MCP... so I added that. ⭐ Please give this PR a thumbs up!

Newest release of OSX (mac) of 15.6.1 removed alias of "python" to "python3" (weird). > python zsh: command not found: python You can re-add it as an alias in .zshrc (if you use zsh (the default). echo "alias python=python3" >> ~/.zshrc And then it works again: > python --version Python 3.13.7

Keep your fingers crossed!! #RSA2026

If your company doesn't permit using Preview features at GitHub (like fresh-off-the-presses AI models) due to them lacking legal indemnification, well, that just changed! Tell your legal teams / platform teams about this. It might let you jump on new models even faster! buff.ly/JEWvOQc

Did you know k8s's CoreDNS permits forwarding specific domain DNS resolution to a specific destination? Edit the CoreDNS config: kubectl -n kube-system edit configmap coredns Add a forwarder: forward domain.com 10.0.0.2 Restart coredns: kubectl -n kube-system rollout restart deployment coredns

I'm gonna start a company called Ex-FAANG and it's gonna look like I've employed thousands of incredible people on LinkedIn

Oooh, the AWS Community Builders program gets a special CFP again this year for #ReInvent! I submitted, everybody cross your fingers for me!!! 🤞 🍀

The last article in my Vera@teams series is now out, URL is in the picture! It covers how to write python that can interface with Teams, and by necessity, SharePoint and OneDrive, to use the user's oauth2 token to fetch their files so you can consume them with Bedrock AI.

I'LL BE SPEAKING AT LWT IN NY IN OCTOBER!! I'm teaching a workshop on how to build your own compliant gen AI assistant in healthcare. I can't wait :D Message me if you'll be able to make it!

I just wrote the last article in my Vera@Teams series (image related). I'm sad to see it go, but I'm very excited for my next foray into MCP enabled bots, particularly my current project - an SRE bot that can read PagerDuty, Splunk, Jira! Super cool stuff coming, I'll publish what I can!!

I've discovered strands (strandsagents\.com), an open-source project from AWS that lets folks build MCP-compatible agents with nearly no code, and I'm THRILLED. Is anyone else using Strands yet? It rules. Write-up coming!!