AWS Summit 2026: The Agent Announcements That Actually Matter

Dr. Ryan Ries here. I hope everyone made it back home safely from Summit if you were there. I definitely did not stick around for the parade!

The keynote ran through a stack of agent announcements. It's interesting to think about how last year the conversation was about whether agents could do real work. This year the conversation was about what happens once they do.

A lot of the announcements were great, so let’s walk through them and I’ll give you my thoughts. If you have any questions or want to share your opinion on these announcements, reply and let me know what you think.

AWS Context: The Quiet Headliner

AWS introduced Context, a new service that builds a knowledge graph from the data you already have. It reads your databases, your documents, your Slack history, your emails, and it infers how all of it connects. Then it hands that map to every agent in your organization.

Swami said your agent's tenth decision should be better than its first.

With no context, an agent doesn’t get smarter the more you use it. It keeps guessing confidently even when it is wrong.

Give it the purchase history, the shipping status, and the return policy in one connected view, and the guessing stops.

Context keeps its metadata in Iceberg format on S3 Tables, so you build against it with tools you already run. No retrieval pipeline to stand up and governance comes built in, which means agents only reach the data they are cleared to touch. Sounds amazing.

One caveat. This one is coming soon, and AWS shared little on the specifics. I will be watching closely. The problem it targets is the right one. Agents do not fail for lack of intelligence. They fail for lack of context.

Amazon Quick: "Quick Just Does It"

That was the tagline on stage, and I’ll admit I chuckled. If you didn’t know already, Quick is Amazon's AI assistant, and the headline this year was autonomous agents.

You spin up an agent with a job, a personality, and the tools it needs, and it works in the background. One agent could quietly clear incoming orders for your finance team. Another could sit across your CRM, your inbox, and your Slack, ready to draft a reply, raise a flag, or suggest a next move. None of it takes code, so the person who knows the work best can build the helper they want.

The differentiation Amazon leaned on was the knowledge graph underneath. They did a good job explaining what this actually means. Other assistants are either "walled gardens," locked inside one app, or "wild gardens," loose with your data and your security. Quick does neither.

A new activity feed pulls email, messages, calendar, and tasks into one prioritized view. It learns which threads you answer fast and which you skip. Quick also picked up 16 new built-in integrations, with names like Adobe, Moody's, and Snowflake on the list.

Quick started off a bit rough, but it has made leaps and bounds in improvements over the last year and I would say it’s worth experimenting with.

Speaking of that... Mission Cloud launched Quick 5x5x5 at Summit. We can spin up Quick in your environment with 5 data connectors, for 5 users, in 5 days. If you’re curious about Quick, this is the best way to go about testing it out in your environment with minimal effort. Oh, and it’s free!

AWS Continuum: Security at Machine Speed

Continuum is a new AI-native security service for code vulnerabilities, and it works the problem end to end.

It hunts down weak spots across your environment, tests which ones an attacker could actually use, sorts them by what they would cost the business, then proposes a fix and can carry it out inside boundaries you draw.

What I found interesting was the reasoning behind it. Amazon pointed to specialized security models, naming Claude Mythos, as the thing that forced a rethink. Those models have sped up how fast vulnerabilities get found and exploited, on both sides of the fight. Attackers now move at machine speed. So, defenders who move at human speed lose.

Continuum is model agnostic, so it uses different models where each performs best. Every action is explainable and auditable, and you can see what a rollback would do before you commit to it. There is a threat modeling piece too, generating threat models straight from design docs or source code. Security teams have wanted that for a long time.

Amazon Bedrock AgentCore: From Demo to Production

AgentCore is where Amazon wants you to take an agent from a clever prototype to something that runs in production.

A few updates stood out:

• Managed Knowledge Base handles ingestion, parsing, and retrieval for your RAG setup, with native connectors to S3, SharePoint, Confluence, and Google Drive. Fully managed means fewer knobs to turn, which is a trade. Teams that want speed will love it. Teams that want fine-grained control will weigh it carefully. The upside is that it feeds straight into AWS Context.

• New optimization tools turn production traces into improvements, surfacing failure and intent patterns across hundreds of agent sessions, with A/B testing now generally available.

• Bedrock Guardrails are now built into AgentCore, checking every agent action for prompt injection, harmful content, and exposed sensitive data.

• The AgentCore harness is generally available. You declare what your agent does, and it handles the orchestration loop, tool execution, memory, and error recovery. Strands lets you build a custom harness when you need one, with evals to check the results.

Kiro Goes Mobile

Kiro, Amazon's software development agent, is now on iOS. You can start a project, steer an agent, or check progress from your phone. The session runs in an always-on cloud environment, so when you sit back down at your laptop, you pick up exactly where the agent left off.

Amazon wants people to know that team size is no longer the bottleneck. The right tools, available wherever you are, become the new constraint.

I have a couple more closing thoughts on this one at the end of this Matrix.

AWS Transform: Tech Debt That Cleans Up After Itself

Transform is Amazon's agentic modernization service. The new capability is continuous modernization. Instead of running it once at a problem, you let it run constantly. As your agents write new code, Transform works behind them, finding tech debt, fixing it, validating the fix, and learning from each pass. It plugs into CodePipeline, Jenkins, GitHub Actions, and GitLab.

The idea behind Transform is that tech debt never gets the chance to pile up. Your codebase stays current on its own.

My Thoughts

Two things that I'm thinking about as I'm writing this from JFK.

The first: an agent is only as good as what it knows about your business. AWS Context, the AgentCore knowledge base, the Quick semantic store, they all reach for the same prize: a single, trusted view of your data that any agent can draw on. Getting your data in order is crucial for our new AI world. It’s smart that AWS is helping customers prioritize this.

The second thought has nothing to do with any product. It is about us. Kiro on your phone means you can ship code from the subway. Transform means your codebase gets cleaned up around the clock. We now hold tools that let work continue every hour of the day.

Speed is a gift right up until it becomes the only thing we measure. I wish we did not glorify the ability to work 24/7 because honestly, who wants that?! Leverage should buy us better decisions and more room to think, not just more output crammed into the same tired hours.

I want AI to help me work less... not work more.

Let's Talk

If your team is staring at this list and wondering which of these actually fits your environment, that is the exact conversation we have in our Mission Cloud AIM sessions. We help you sort the hype from the fit, map it to your data and your goals, and hand you a plan you can act on. Reach out to our sales team here if that sounds like something your team would find useful.

Until next time,
Ryan

Now, instead of an AI-generated image this week, I figured I’d share some pics from the Summit.