Aker Systems is a data modernisation company, and for data engineers like me, it means thinking a lot about what AI-ready data actually looks like.
From 30 November to 5 December 2025, Aker Systems attended AWS re:Invent in Las Vegas, focusing on data engineering and data platform topics, including Amazon Redshift, Amazon S3, Amazon Athena, and Amazon Sage Maker. The week highlighted a key trend: modern data platforms on AWS are being redesigned around open table formats, increased computing power, and an AI-first approach.

‍^{What I learnt about Modern Data Architecture through chalk talks}

The strongest signal was that the traditional split between data warehouses and data lakes is disappearing, giving rise to Data Lakehouse Architecture. Across the Redshift,S3, and Athena sessions, Apache Iceberg repeatedly appeared as the standard that ties them together. Real-time data and streaming were recurring themes throughout the chalk talks. Many of the most interesting use cases from clickstream analytics to operational monitoring and AI-driven personalisation assumed that events were arriving continuously, being processed in flight, and then landing quickly into those same open tables.Rather than treating streaming as a separate pipeline, the pattern was to use it to keep the Data Lakehouse fresh. That aligns closely with how we build platforms at Aker Systems, where streaming ingestion and low-latency processing are important parts of the architecture rather than an afterthought.

Redshift can now create and write Iceberg tables directly with familiar SQL, storing the data inS3. S3 Tables adds an analytics layer on top, handling layout and optimisation in the background. Athena then queries those same Iceberg tables where they live.In practice, you get a single open table format with multiple engines that all see the same data.

As a data engineer,that changes the job. Instead of duplicating data across systems, you design clean schemas, enforce quality, and build trustworthy pipelines once. For Aker and our clients, where one governed platform must serve many teams and security boundaries, that shared Iceberg pattern is a good fit.

 Another theme was how much more computationally intensive everything has become. Improvements in Redshift and Athena, along with new instance families, all point to AWS expecting complex analytical workloads and near real-time processing as normal.That is good for demanding queries, but it means observability and cost control must be designed in from the start.‍

From Data Platforms to AI ready Data Platforms

 Sage Maker now makes it far easier to customise foundation models without building bespoke infrastructure. Combined with the managed model offerings on AWS, you have realistic options for bringing your own data to models, from light tuning to deeper domain-specific work. The link back to the data platform is simple: these models expect high-quality, well-described input. If your data platform estate is messy or hard to govern, you feel it quickly when you try retrieval-augmented generation, recommendations, or natural language interfaces. AI exposes the strengths and weaknesses of your data platform very quickly.

‍Agents Everywhere, Including Operations

Another big theme for me was agents, especially the AWS DevOps Agent. DevOps Agent is presented as an experienced engineer who never sleeps. It learns the shape of your environment, plugs into observability tools, and uses that context to investigate incidents and suggest fixes. In many ways, it behaves like an SRE, very good at the basics of triage, correlation, and suggested runbook steps, but itis not about to replace human SREs any time soon.
Architecture choices and the design of safeguard rails still need skilled DevOps and SRE teams. That is where Aker’s DevOps capability comes in, using experience of complex, highly regulated platforms to design systems that an agent like this can operate in safely.
If there is now an agent for DevOps, it is easy to imagine agents for data quality, cost, and access governance.

That creates two clear requirements. Your data platform needs to expose enough metadata,lineage, and context for agents to act safely. Your people need to learn how to work with agents, which is where AI literacy and prompt engineering start to look like core skills, not side interests.

‍What organisations should do to modernise their Data Platform and become AI ready?

When speaking with several vendors and architects alike, I thought less about individual product names and more about the pattern that is emerging. For organisations, especially in the regulated sectors Aker serves, the message was simple: use the new features to tidy and centralise your data, then layer AI and agents on top in a controlled way.

1.      Modernise the Core Data Platform

Organisations can use a combination of AWS data services to modernise their platforms. Amazon S3provides durable object storage for analytical data, while AWS Glue Data Catalog and AWS Lake Formation offer central metadata, access control, and governance. For query and analytics, Amazon Redshift and Amazon Athena are steadily converging in capability as AWS adds similar features to both for warehousing and interactive querying, so teams can choose the right tool based on workload pattern and cost, not hard limitations.

Around this core,services such as AWS Glue for ETL, AWS Step Functions for orchestration, and streaming options like Amazon Kinesis, Amazon MSK, and Amazon Managed Service for Apache Flink help organisations keep data flows current without building everything from scratch. Together, this toolkit lets companies modernise their data platform incrementally, using managed services rather than bespoke infrastructure.

2.      Add AI and Agents on Top of the Platform

If you’re ready to add AI and ML features into your organisation, use Amazon Sage Maker, Amazon Bedrock, and the new model offerings to build an AI layer that reads from those trusted tables instead of shadow copies.

Services such as S3Vectors and the growing family of agents, including AWS DevOps Agent, work best when they can rely on accurate, well-described data in one place. Decide early how embeddings, features, and model outputs will be stored and monitored so agents behave predictably.

3.      Invest in People, Skills, and Prompt Engineering

Finally, invest in the people who will make all this work. Create AI and ML learning paths for engineers and architects and give teams practical training in how to design tasks for agents and how to write prompts that express intent, constraints, and risk clearly.

Prompt engineering is not magic, but it is a real craft. As more of the AWS portfolio and its services gain an agentic side, the organisations that combine strong data foundations with people who are confident working with agents will move faster and more safely than those who cannot.

For me, that was the real lesson of re:Invent. Modern data platforms and AI readiness are now the same journey, and the organisations that start that work now will be the ones who benefit most from what comes next.