Andy Kroll

Agent-ready is not the same as modern. My re:Invent session, TNC339, shows how to get AI agents working on legacy apps. In our session, you scan a sample booking service. It comes back with 43 findings, and only one is a BLOCKER: no way for an agent to sign in. You rank the fixes into a roadmap with Kiro, then apply one automated fix, with no changes to the business logic. In this example, that moves the service from Remediation Required to Pilot-Ready. Unblock first, perfect later. The other 42 go on the roadmap. It is a 2-hour hands-on bootcamp for AWS Partners (Level 300). Bring a laptop. First session: Tue Dec 1, 3:30 to 5:30 PM. Repeat session: Wed Dec 2, 1:00 to 3:00 PM. Reserve a seat: registration.awsevents.com/f…

🚀 We're LIVE! The AWS AI FastTrack kickoff is happening right now — and it's not too late to join. This free 4-week program is open to all AWS Partners worldwide and gives you everything you need to build and sell AI solutions: ✅ Business & Technical learning tracks ✅ Live sessions + self-paced Skill Builder courses ✅ Free AWS Foundational Certification exam voucher opportunity Register now and catch today's Program Overview before it wraps. 👉 events.ringcentral.com/event…

If you've ever gone through the AWS Partner Certification Readiness (APCR) program, there's a decent chance you've run into something I built. I've spent the last couple of years building most of that content on the AI and ML side, and for the past several months I've been using Kiro Crew, previously known internally as MeshClaw, to help check accuracy before anything ships, enforce design guidelines, keep content current, and more. Everything still comes back to me before we ship it. What I didn't expect is how much of this improved just by correcting it. I keep giving it feedback on what it generates, how something should be formatted, how I actually want a thing designed, and instead of that feedback disappearing, it turns into a rule the agent applies on its own. Six months ago I was still writing formatting guidelines from scratch every time. Now I correct it once. This year I connected it to nearly everything else I use, to the point where it functions less like a tool and more like someone I brief once and delegate to. For complex decisions with multiple valid paths, like how to frame a concept for a mixed audience or where to draw the scope line, I run it past a multi-agent review panel first. My Obsidian vault backs all of it, every project in one place, so when a new tool shows up like Amazon Quick Desktop, I point it at the vault and it already knows what I'm working on. I've been running this internally for most of the year. Glad to see it go public. It's open source now, so anyone can build on it. kiro.dev/crew #MyKiroCrew #KiroCrew @kirodotdev

I've spent a big part of this year helping build AWS AI FastTrack 2026 and am proud of where it landed. Registration is open; the program runs throughout September. This isn't last year's program with a new coat of paint. We rebuilt it around agentic AI - 75%+ net-new content across Amazon Bedrock, Amazon Nova, and Amazon Kiro, new gamified SimuLearn exercises, and expanded badge and certification pathways. It's a free 4-week program for AWS Partners, built for your whole team: Business track – Sales, GTM, BD Technical track – SAs, Developers, Engineers Looking forward to seeing everyone in a few weeks!

Waco will always have a piece of my heart after 15 years, but I’m excited to share that we’ve officially relocated to the north Austin area (Liberty Hill)! It’s a big change leaving such a tight-knit community, but the timing felt right for a new adventure. I’m still at AWS doing AI enablement and exited to connect with the Austin tech scene! It’s been a fun week of volunteering and exploring the area but if you’re in the area I’d love to hear your food and kid activity recommendations 🤤

View my verified achievement from @awscloud. credly.com/badges/3eb740cc-0… via @credly

This is nice because now I can pull in this power to validate templates only when I need to do that action which frees up my context window!

Having post #AWSreInvent blues, so I decided to reminisce about my visit to the AWS Certified Lounge. Generated this video using @grok Imagine. The opening image is a real photo I took, then had Grok generate the video from it. It looks remarkably like my desk when I'm studying for certs with all the notes I take! Here’s the prompt I used: Late night study session aftermath, MacBook Pro screen glowing with AWS Solutions Architect practice exam results, camera crane movement rising from keyboard level upward and backward, revealing scattered study materials, celebratory CERTIFIED AWESOME coffee cup still steaming, half-eaten victory brownie on plate, disposable coffee cup knocked over empty, purple-blue neon lighting from off-screen monitor creating dramatic rim light, exhausted but triumphant atmosphere, desk cluttered with handwritten flashcards and crumpled notes, golden hour window light mixing with cool LED glow, emotional payoff moment after months of preparation, shallow focus pulling from laptop screen to celebration spread, 8 seconds, documentary-style authenticity

A great concise way to see the big announcements at #AWSreInvent 👇

Love seeing all the Kiro updates! #AWSreInvent

That’s a wrap for #AWSreInvent day 3! Ended the night at the AAI party and had a great time catching up with everyone. Had the pleasure of running into @DigitalColmer again! Took some selfies at the party but after reviewing them they were very blurry so enjoy this Wizard of Oz pic 😂

Learning how to migrate the well-architected way in our workshop this morning! #AWSreInvent

Love the analogy of Kiro powers being like Neo instantly downloaded martial arts expertise in The Matrix! #AWSreInvent

Had a great discussion in today’s workshop about agent memory in AI systems, and thought I’d share what makes short-term vs. long-term memory in AgentCore Memory different and when to use each. Short-term memory captures turn-by-turn interactions within a single session, letting agents maintain immediate context so users don’t repeat themselves. Think of it like working memory—when you ask “What’s the weather in Seattle?” and follow up with “What about tomorrow?”, the agent remembers you’re still talking about Seattle. Long-term memory automatically extracts and stores key insights across multiple sessions—things like user preferences, important facts, and session summaries. This is persistent knowledge that creates personalized experiences over time. For example, if you mention preferring window seats during a flight booking, that preference gets stored and the agent can proactively offer window seats in future conversations. This ties directly into context engineering—the practice of designing systems that provide the right information, in the right format, at the right time to help AI models accomplish tasks. While prompt engineering focuses on crafting individual instructions, context engineering is about building dynamic systems that manage information flow across multiple interactions. Memory systems are a core component of context engineering, determining what information gets included when the model generates responses. By combining short-term and long-term memory strategically, you’re engineering the context that makes AI agents feel less like stateless question-answerers and more like intelligent assistants that actually know you. #AWSreInvent