AWS re:Invent 2026 for CTOs

When and where is re:Invent 2026?

Las Vegas, Sunday 29 November to Friday 4 December 2026, with sessions from Monday to Friday lunchtime (the last ones end at 12:30 PM on 4 December). Keynotes are at The Venetian. AWS lists more than 2,000 sessions and labs and more than 420 sponsors in the expo. The full pass is $2,499, with 10% off bulk purchases, and AWS publishes an approval toolkit for making the business case to your own finance team. re:Play, the Thursday night party, is at Allegiant Stadium.

What is the re:Invent 2026 keynote schedule?

Five keynotes, all in Pacific time, all listed by AWS as in person and virtual. AWS had not named the speakers beyond Garman when I checked.

From Europe, the CEO keynote starts at 16:30 in London and 17:30 in Berlin. The Developer keynote on Monday evening is 04:30 Tuesday in Berlin, so watch the recording.

How do I watch the keynotes?

Register for the free virtual pass on the re:Invent site and you can stream every keynote live. Recordings go up on AWS's YouTube channels afterwards. For a CTO, the honest advice is to watch Garman live or the same evening, skim the Data & AI keynote, and let your platform leads cover the Developer keynote.

What is AWS pushing this year?

Agents. Read the session catalog and the same handful of themes repeat at the 300 and 400 level. These are the ones I would expect to show up in your engineers' proposals by Q1.

• Agents in production on Bedrock AgentCore. Runtime, memory, gateway, identity, evaluations and policy, with sessions like "ship an agent to production in 3 hours" (AIM439) and context engineering on AgentCore (AIM450). This is AWS's answer to "where do our agents run".
• Multi-agent orchestration. Step Functions vs Lambda durable functions vs AgentCore, and the MCP, A2A and ACP protocols (ARC312, SVS401, COP402). The decision your architects will bring you is which of these to standardise on.
• Governance and isolation for autonomous agents. Scoped IAM and Cedar guardrails (SEC407), an MCP Gateway and Registry for governing AI assets (OPN301), and Lambda MicroVMs as sandboxes for agent-run code (SVS333). If security has been blocking agent pilots, this is the material to send them.
• Context and memory. Knowledge graphs, S3 Vectors, semantic caching on ElastiCache for Valkey, and a new "AWS Context" for governed agent knowledge (ANT423).
• Agent observability and evals. OpenTelemetry, Strands and Langfuse tracing (OPN304), eval-driven observability (COP401).
• AI-driven development with Kiro. Kiro for serverless development (SVS303), AWS's AI-Driven Development Lifecycle methodology (TNC336) and guarded coding agents on Bedrock (AIM418).

Session codes are from the 2026 catalog via Ran Isenberg's session selection; check the official catalog for times. If you are still deciding how far to let agents into production, my agentic AI security guide covers the controls to ask for before you do.

What did AWS announce last year, and what follows from it?

re:Invent 2025 launched Graviton5 (192 cores), Trainium3 and its UltraServer, Trainium4 (in development) designed to work alongside NVIDIA chips, four new Nova models plus Nova Forge for customising them, AgentCore policy, memory and evaluation features, frontier agents for coding (the Kiro autonomous agent), security reviews and DevOps, on-premises AI Factories, and Database Savings Plans with up to 35% off. The 2026 catalog reads like the follow-through: fewer new primitives, more "run it at scale and govern it". Watch for Trainium4 availability and pricing, and for whether the frontier agents leave preview.

What should a CTO bring back from re:Invent?

1. An agent platform position. Build on AgentCore, stay framework-neutral, or use another cloud's equivalent. Write down which, and why, before your teams pick three.
2. A cost view of agent loops. AWS has a 400-level session on the cost and latency of agent loops (AIM429). Agent workloads bill very differently from chat; get a per-task cost estimate before scaling a pilot.
3. A governance answer for security. Scoped IAM, Cedar policies and sandboxes are the concrete controls. Map them to your current AI governance policy.
4. A view on AI coding tools. If Kiro gets a big launch, your developers will ask how it compares to what they use. Decide whether to run a time-boxed comparison rather than letting it spread informally.
5. A one-page readout within a week. What launched, what it changes for roadmap and budget, what you are deliberately ignoring.

I will update this page with the actual announcements during the week of 30 November. The CTAIO newsletter will carry a CTO recap after the event. For the rest of the year's events, see the AI conferences 2027 calendar.