I think I’ve already walked a million steps, and the first two days have been both busy and exciting. Attending the Analyst Summit gives me the opportunity to learn the AWS way in a much more intimate and direct setting.
Here are the 5 main highlights from Days 1 & 2, covering Amazon Web Services (AWS) CEO keynote and yesterday’s “Beyond the Possible” Analyst Summit panel:
1. Agent-building is a new foundational skill
Eric Brandwine emphasized that agent-building is a new skill, distinct from both coding and human delegation. Amazon engineers already build small agents that automate 10-minute tasks, then immediately build more, creating compounding productivity. AWS introduced a concrete example of this direction: the AWS DevOps Agent, designed to accelerate incident response and automate operational tasks.
This reinforces the message that agentic systems are becoming a standard part of how teams build, maintain, and operate services. This is not new, but the push from AWS will drive its customers to move toward more agent-based development.
And let’s not mistake this, building an agent is a skill that not only developers must have, but also marketers, analysts, service reps, and many other professionals and teams. Read further to learn what this means when it comes to complexity vs. simplicity for these users.
2. Project Rainer and 50B! investment
Leading in HPC, the leadership team walked through the long arc of AWS’s supercomputing journey, which began with a foundational doc, followed by five years of learning, partnerships with Lawrence Livermore National Laboratory (LLNL), and ultimately AWS’s commitment of up to $50 billion (!!!) to build AI and supercomputing infrastructure for U.S. government customers.
Project Rainer is the next major milestone, a mission to extend AWS’s relevance and leadership in high-performance computing by combining cloud-scale engineering with national-level scientific workloads.
…and I remember times when people were saying, “AWS is only for startups.” 🙂
3. The “Nova” Family: A New Layer for (Our) Content Engines
AWS introduced the new Nova lineup, and for those building scalable content systems (including us at IOD (iamondemand) ), it marks a practical step forward.
Nova 2 Omni (Preview): A multimodal model that accepts text, video, and speech as input and can generate both text and images. I shared it with our team earlier, and the use cases were immediately clear — for example, taking a raw product demo video and producing a technical blog, social snippets, and a diagram in a single workflow. It has the potential to simplify and shorten content production cycles.
Nova 2 Lite: A cost-efficient reasoning model with a 1M-token context window. Useful for long-form content, research-heavy work, or processing large, complex briefs.
Nova Forge: Allows organisations to train custom models using their proprietary data alongside curated Amazon datasets. This supports a common need we see across enterprises: adopting AI while preserving domain knowledge and IP.
Amazon Nova: AWS’s Strategic Pivot in the Enterprise AI Landscape—Part 1
4. Agentic governance and identity are critical pillars
Byron Cook and Eric Brandwine stressed that agents are a new class of actor, not human and not machine, requiring entirely new identity and permission models. Examples they gave:
- Logs must show “Eric requested it, but the agent executed it.”
- Permissions must be built around workflows, not static agent identities.
- Organizations must explicitly define PII boundaries, sovereignty rules, and regulatory mappings so agents can operate safely and autonomously.
On that note, AWS announced the AWS Security Agent (Preview), which proactively tests applications for vulnerabilities during design and deployment, ensuring these new agentic systems are secure by default. And there are so many new ventures in this space (Check Ariel Dan‘s new one…)
5. Simplicity for users relies on massive complexity underneath
“The user sees simplicity; AWS handles the complexity.”
Rahul Pathak explained that while prompting an AI model feels as simple as typing into a map app, underneath sit layers of sophisticated engineering, the AI equivalent of satellites and sensors. Users can ask natural-language questions and get precise answers powered by advanced reasoning techniques.
This “simple on top, complex underneath” pattern is shaping how AWS approaches every new GenAI and agentic experience. TBH, as we at IOD move forward with building our GenAI frameworks, we as users still experience great complexity in building our new production lines on top of GenAI (just think about what it means to pick the right model for the right task…)
There were dozens more releases across new instance types, services, and capabilities. The pace of innovation isn’t slowing, it’s accelerating!
Looking forward to Days 3 and 4, and as always, to Dr. Werner Vogels ’ keynote on Thursday, which I make sure to attend in person every year in the keynote hall.
