AWS re:Invent 2025 re:cap
First of all, a big shout out to Martin Damovsky for providing re:Watch service where you can rewatch all the sessions from re:Invent 2025.
I still haven't watched all of the sessions I have in my "watch later" playlist but since it will take some time I would like to share what I've learned so far.
My focus was on serverless, architecture, deep dive of individual services. I tried to avoid AI hype.
Compute
Both ECS and Lambda are serverless compute options. I have seen many sessions which were really ineresting. In this section you can read the breakdown.
AWS Lambda durable functions has been introduced in CNS380 by Eric Johnson. If someone asked me what durable function is I would probably say that it is (almost) everything what Step function is but simplified for developer experience. I've used Step functions in the past and it was kind of weird to model them through CDK as their main selling point is that you can compose Step functions within console (and also in VS Code) by dragging and dropping components - I've never used this functionality as I rather describe things via code.
The promise of durable functions is mainly better DX. Basically AWS provides Durable execution SDK by which you can model your workflow within single lambda handler. The "durable" means that if there is some step where you would usually use wait the execution of lambda stops and then it is being re-invoked with the state persisted through SDK. It looks like you can now simply write the code and everything else (how the steps being executed and connected) is being handled behind the scenes.
I haven't tried this yet but I am very keen to try it when I see the fit as it sounds amazing to me.
Of course you should also watch API313 again by Eric - this time he talks about Step functions. After watching these two talks you will know the difference between durable and step functions and hopefully you will be able to chose the right tool for the task you are trying to solve.
Related to Step functions is also talk API323 which is about the simplification of Step functions with JSONata - basically you can do lot of cool transformations with JSONata in order to avoid transforming your data by invoking additional Lambda functions - at scale this might save you some money.
If you still don't have enough of Lambda you can watch CNS376 which sums up all the new things in AWS Lambda. Another exciting introduction is Lambda Managed Instances and tenant isolation. Lambda Managed Instances has been introduced in CNS382. I am not going to use this on my projects any soon (as I don't have the scale needed for this) but I am glad that I have a knowledge about this so I can talk about this option with companies which might want to use it.
If you are not sure if Lambda Managed Instances are for your use-case you can try sedai Amazon partner.
It seems that the Lambda Managed Instances was a hit so AWS introduced also ECS Managed Instances in CNS416. The gist is the same as with Lambda - you can use any ECS family and instance type with ECS Fargate and everything is going to be managed by AWS for you.
Database
DSQL looks very promissing - finally SQL database which is fully serverless. You can see how it works at DAT439. It also mentions that it is dependent on TimeSync which has been explained in CMP409 - basically AWS can sync time at microsecond scale as always very reliable.
You can see the DSQL from dev perspective in DAT401 - there is full repository with various examples on how to connect to DSQL. For instance there is example of Prisma 1 connection with all of its quirks explained.
Another great serverless offering is DynamoDB. It is great database but as with every tool it is not good for any use-case. You kind of need to fully understand how the database works and what use-cases you can cover by using it.
If you are new to DynamoDB or you want to learn more 2, I recommend you DAT311 session. Jason Hunter knows what he is talking about, I've recently come across this great atricle of his. If you want to see some more advanced techniques as always I recommend you talk by Alex DeBrie DAT414.
If you want to understand the DynamoDB fully under the hood you should see this deep dive DAT436 - there are some great architecture choices and principles mentioned here.
Of course you should also see DAT453 if you want to hear more about the DynamoDB outage from October 2025.
There are also interested sessions regarding Aurora for instance DAT456 which talks more about how secure the Aurora actually is agains the 0 day exploits.
Other sessions
Here are few additional interesting sessions which I either watched or I am planning to:
- CNS403 - Best practices for serverless developers
- DAT455 - A tale of two transactions
- NET340 - Advanced VPC design and new capabilities
- DAT441 - Deep dive into Amazon Aurora and its innovations
- API311 - Designing resilient Serverless Applications
- CNS367 - Debug Lambda functions deployed in cloud like never before
- ANT423 - Amazon Kinesis Data Streams under the hood
- API321 - Solving the Observability Mystery with AWS Step Functions
- COP414 - Resilience testing and AWS Lambda actions under the hood
- DAT458 - Better, faster, cheaper: How Valkey is revolutionizing caching
- DAT437 - Optimizing for high performance with Amazon ElastiCache Serverless
Conclusion
It is great that you have a possibility to see all of the sessions even if you don't have time or money to attend re:Invent in person. I have learned lot of cool new stuff and strenghten knowledge thinkgs I knew.
Don't forget that all serverless services are running on top of EC2. We should be very greatful to folks at AWS so they took the burden of managing these instances from us simpler folks and we can enjoy all the serverless options 🙃