I went to AWS Cloud and AI Day in Santiago, at Metropolitan Santiago. A full day of talks on cloud and artificial intelligence, with the focus on agents.
Three stages at once
The first thing that surprised me was the format. In the same room there were three stages side by side, with three talks running at the same time. Every attendee had headphones with a selector to pick which channel to listen to. If a talk didn't grab you, you switched channels without leaving your seat.
It's the first time I've seen anything like it.
Building an agent now takes minutes
The keynote presented AgentCore harness, which AWS had launched in June. The promise on screen was direct: go from idea to working agent in minutes, with no orchestration code, no infrastructure setup and any model you want.
Two years ago, building an agent was a months-long project. Today it's an afternoon.
The model is just one piece
This is the slide that stuck with me. At the top, the model. Around it, everything an agent needs to really work: identity, data, observability, a safe sandbox to run in, a knowledge base, compute, tools and memory.
The model gets the most attention, but it's what least sets one agent apart from another. What makes it useful is what surrounds it.
"Context is everything"
That was the line of the day. An agent is only as good as the context it gets: the data it can reach, the tools it can use and what it remembers.
AWS also presented web search in AgentCore, so agents can work with current information from the internet. Another way to give them context.
What it means for companies that manage workers
I left with one clear idea: building the agent is no longer the hard part. Giving it context is.
Think of an agency with 600 workers across 8 clients. To know whether someone is missing a document, whether their hours match payroll or whether the client can be billed yet, the context isn't in one place. It's spread across the ATS, the HCM, payroll, the ERP, spreadsheets and WhatsApp chats. An agent that only sees one of those systems makes decisions with part of the information.
At Aindez that's exactly what we work on: connecting the systems you already have so agents see the full picture and can act in the right system, with a person approving what matters.
Agents will keep getting easier to build. Your operation's context is still yours, and that's where the difference is.
Get the Aindez blog in your inbox
Get every new post in your inbox.
Manuel Vallejos · Founder and CEO of Aindez, the AI orchestration layer for companies that manage employment at scale.
Claude Opus 5.5 · Anthropic's AI model. Co-author of this post: it researched the sources, organized the data and drafted text that Manuel reviewed.