🔥 This very popular article is now FREE: AWS AgentCore Agentic Slack Bot - Full Architecture and Code (1/4). Moved Vera from the Lambda “invoke and die” model to a warm, long-lived AgentCore app built for agent workflows buff.ly/glbZD3h

When your mom is into computers you get the jokes early 😆

Post is now free: migrating VeraResearch to first-class MCP tool support in Strands. Running multiple MCP servers caused tool name collisions and a pile of custom wrapper code to manage prefixes and filtering. This post shows how we deleted that complexity using native MCPClient buff.ly/fV2InLi

Claude, I'm gonna take that as a yes

Post is now free on how to build audit logging across AI bots using python and AWS cloudwatch. Agentic bots make many calls per request, so audits get messy fast. This post shows the simple audit layer we added user identity, question, response, sessionID for easier debugging. buff.ly/2TZtqEt

I'm a huge nerd for Andy Weir books, I've read The Martian ~12 times and Hail Mary 3 times. Cannot wait to see this live!! Anybody else going?

I had an unreasonably fun time with Chandana talking about: - How to get started to tech - Programming vs Infra background to DevOps - The rise of AI in DevOps - How to excel at work and life in the age of AI instability It was a great convo, hope you enjoy.

It's not just you, the entire github platform is unstable and crashing right now. So if we need Claude to write code, and GitHub to store and deploy code, and both have nearly daily issues, like... what the heck do we do now. Is this what software engineering is now?

New post is live on how to add AgentCore Memory to an AWS AgentCore Agentic slack bot, (also Strands tools to permit delete memories, something Strands memory manager doesn't natively support) This one dives into how to build persistent user memory. Article is paid but not for you: buff.ly/4VxFHZ7

Just shipped a big update to our open-source AI Slack bot (Vera) on AWS Bedrock AgentCore. New in the repo: • Per-user Atlassian OAuth (no shared svc acct) • Sub-agent delegation • Matplotlib charts → Slack threads • Slack file uploads MIT + Terraform. Deploy it yourself. buff.ly/bDIpfMf

Article is now free! Having an agentic bot read AWS accounts means working with IAM, and assuming roles cross-account. I did all that, and wrote down how it works. You can use boto3 to assume IAM roles on demand. It's awesome. Check it out. www.letsdodevops.com/p/teaching-a...

If you’re building agentic Slack bots with AWS, this one dives deep into the AgentCore MCP Gateway and how I’m wiring Vera into PagerDuty, GitHub, Atlassian, Splunk, Azure, and AWS CLI. I share all code, as well as why. It's paid, but not for you. buff.ly/6yWW6BJ

Oh, so THAT's why my Atlassian/Jira MCP keeps failing. You can track here: buff.ly/FrnVk2i

I've discovered sub-agents, which can work in parallel on tasks coordinated by a manager AI, and I'm never going back. This is super cool.

This was yesterday. I don't remember having seen github runners limited by capacity. Particularly with the ownership by $MSFT, how can they be limited by capacity?

The article Fixing MCP Tool Name Collisions When Building a Unified CLI Tooling Experience is now free to read on the Day Two DevOps blog. buff.ly/MueTGQS

Wrote up how we're combining local and remote MCPs on Strands with the AWS AgentCore gateway. Article is coming 2/17! The code is open source and available now though on github: github/KyMidd/AgentCore_AgenticSlackBot

It's not just you, many github functions are broken right now

I built an entire AWS AgentCore agentic bot for Slack, and you can too. All architecture and code shared. I added memory per user so you can teach the bot how to help you with your job, and how to program. It's all built with Strands toolkit. Absolutely rad stuff here. More coming soon!

🚀 Now free: Solving enterprise scalability and logging around AWS Bedrock. If you're running multiple bots or just want better visibility into who’s using what in Bedrock, this pattern helps you get some insight into your AI usage. Check it out: buff.ly/9rjmMpW

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: