AWS re:Invent 2026: What We're Walking In With
In this blog
Our team's expectations, open questions and bold predictions before the week begins.
From Nov. 30 to Dec. 4, more than 60,000 people will land in Las Vegas for what AWS is calling cloud and AI's biggest week. The theme this year is "Build what's next," and the framing is refreshingly practical: architecture decisions, migration tradeoffs, token spend and security risk. That's the same list our customers bring to us.
Before the keynotes start, our team wrote down what we're hoping to see, what questions remain, and what we think people will be thinking about on the flight home. After the event, we'll circle back and grade ourselves.
Our team brings a range of re:Invent experience. For some of us, the event is still new, and we're approaching it with fresh eyes and a lot of curiosity. For others, it has been more than a decade of re:Invents, and that experience helps us know where to focus. Together, we're set up to cover a lot of ground.
Agentic AI will power everything. The foundation underneath it is the story.
The pilot era is wrapping up. The conversation we expect to dominate the week is how agentic AI moves into production, and we expect it to power nearly everything else on the agenda, from data and security to developer tools and migration. What makes that possible is the unglamorous work underneath: data, governance and cost.
That's also where the business case gets interesting. AI is making it cheaper to leave legacy platforms behind, which means modernization cases that got shelved a few years ago deserve a fresh look. Agents can only work with what they can reach, so a modernized, governed foundation is what turns a pilot into production.
Amazon Quick: Only as good as what it can reach
Amazon Quick is the announcement we'd bet on hearing more about. The problem it is built for is familiar: the answer sits in several systems, and nobody can see across them. But Quick can only work with what it's connected to. A customer on legacy platforms with no clean data layer or identity setup gets a demo, not a result. A customer that has already moved data into a governed cloud foundation gives it something to work with.
That's why we read Quick as a reason to reopen modernization work, not a replacement for it. At the show, we want to hear which customers AWS puts on stage for it, and how it carries source-system permissions through to what each person can see.
The question we want answered: how are customers measuring AI's business value?
Vendor success stories tend to come from one customer under one set of conditions, and they rarely say how the value was measured or what that customer pays every month to keep it running. The harder question isn't only what AI costs. It's how customers decide what measured impact looks like and then balance that against cost. Usage volume, how much data has to move, how long logs and records must be kept, and how much human review stays in the workflow all change the total, and so does what a customer counts as a win. For workloads with sensitive data, the controls add their own cost and effort.
That's where AWS's newer Business Value Realization (BVR) program is worth a closer look. We'd like to hear how it's helping customers tie AI investments to measured outcomes, and what early results look like. We'd also like to see a range of results, along with what drove the low end and the high end. That's the evidence that decides whether a second use case gets funded.
AI is moving into every service, and trust has to move with it
Don't expect one big AI announcement. We expect AI features to show up across a wide range of services: data, security, developer tools, operations, and migration itself. A couple of years ago, AI was its own track. Now it's becoming part of how everything else works. And re:Invent is always a bit of a happy mystery, so we're hoping one of the surprises is agentic e-commerce, along with features to build the data and catalog it depends on. We think that area is ripe for major enhancements in 2027.
The shift we're watching most closely is that trust in agents is turning into an evaluation discipline. Governance, monitoring, and security used to get thrown over the wall to another team once the build was done. That doesn't work with agents, which make decisions, call tools, and touch real data. You build confidence by running repeatable evals before you ship, not by writing a policy after the agent is already in production.
One of us analyzes the session catalog before every re:Invent, because session titles say a lot about what AWS cares about and what its customers are struggling with. This year we're hoping to see plenty on observability and agent governance, and we want to know which patterns are working in practice and where the gaps still are.
Secure cloud and secure migration are two different claims
Every vendor says they've got security covered. But a secure cloud and a secure migration are two different claims, and the second one is where projects tend to stall. We want to see "secure by default" hold up beyond the slide, so we'll be looking for specifics on how security carries through the migration itself, not just the destination.
The partner stage
One of the things we're most looking forward to is seeing strategic partners like OpenAI, Anthropic, and AMD, among many others, co-present with AWS on the future of AI in a hybrid world. That's the conversation we most want to hear.
As for announcements, we're watching for enhancements to Forward Deployed Engineering, updates to AWS's migration tooling, and simpler funding programs. Those are the three we'd bet on.
What customers are asking us
Going in, four questions keep coming up in our conversations: how to modernize without a multi-year slog, how to match workloads to the right infrastructure and keep costs in check, how to turn AI pilots into real outcomes, and how to think through multicloud/hybrid cloud decisions. We'll be listening for who has a real answer to any of them.
For many organizations, it's rarely "can AI do X." It's "can I afford to run two systems at once while I modernize?" Plenty of teams know they need to get off their old platforms, but the real blocker isn't the target. It's the double-run window, paying for old and new at the same time, sometimes for years, and nobody budgeted for it. We're hoping to hear more about actually shrinking that window, not just a nicer landing zone to land in.
We're also hearing three more. First, how fast frontier models are changing, and how to build something today that won't need rebuilding every time a better model ships. Second, how to speed up migration and modernization without adding risk, since many organizations know where they want to go but are stuck on how to get there faster. Third, and probably the most common, how to predict what AI will actually cost. Leadership wants a number they can plan around before they commit, not a surprise on next quarter's bill. We'll be listening for real customer stories in the keynotes, breakouts, and builders' sessions, not just roadmap slides.
Our bold predictions
Three predictions from the team:
- People will leave Las Vegas talking less about individual services and more about the operating model for running, governing, and paying for AI and modernized workloads at scale.
- The phrase on everyone's lips will be AI readiness. With this much pilot activity behind us, the question becomes what has to be in place before AI touches real, sensitive data. It's as much a modernization problem as a governance one. Looking toward 2027, we expect early agent use cases to be bounded tasks with a person approving the result, like drafting a packet for review or answering policy questions for internal staff.
- People will leave talking less about which model is smartest and more about whether they can trust their agents in production. Most organizations aren't blocked by capability anymore. They are blocked by confidence: can they prove an agent does what it should, every time, and can they see what it's doing once it's live? We expect evals, guardrails and observability for agents to stop being side topics and become what you have to have before an agent goes live. The teams that treat agent trust as an engineering problem, not a compliance checkbox, will be the ones moving from pilots to production next year.
What a win looks like
Our week is full: booth sessions, customer meetings and dinners, AWS executive briefings, and partner and competency sessions. But hallway conversations are usually where the best ideas start, and we'll be leaving room for those.
For some of us, the week is less about which keynote and more about which human interactions. It's rare to have so many AWS & enterprise leaders in the same city at once, and we want real time with them. We also have a team with many former AWS folks and people deeply plugged into it, so there's a real chance to bridge connections across everyone's AWS relationships, past and present, rather than each of us working the floor solo. If a couple of email threads turn into real relationships, that's a better week than any session.
For us, a win is customer meetings that turn into next steps, stronger relationships with the AWS team, and a clear plan for 2027. We'd add a few deep architecture conversations to that list, the kind that get past the marketing and into why a team picked one pattern over another, what broke and what they'd do differently. The goal is to come home with ideas we can build, test, and put in front of customers in the weeks after the event, not just notes we file away.
Final thought
Whatever lands on stage, the questions underneath it haven't changed: what to modernize, in what order, on what foundation and at what cost. We will be back after the event with what held up, what didn't and what surprised us. If any of this sounds like the conversation you're already having, we'd love to have it with you.