Kai Wähner

Technology Evangelist with Focus on Integration, Event Streaming, Big Data, Analytics, Machine Learning, and Cloud-Native Microservices

[Kai Wähner @KaiWaehner]

#StreamProcessing is still misunderstood. Not new, not niche. The core idea is continuous processing.

My 10-min explainer: stateless vs. stateful, #KafkaStreams vs. #ApacheFlink, through real-time payment #FraudPrevention.

https://t.co/oXMIXBznwa

[Kai Wähner @KaiWaehner]

#Databricks and #Snowflake both shipped Kafka-compatible ingestion. The #ApacheKafka API is the de facto standard now.

But: analytics platforms use it as an on-ramp. #DataStreaming vendors run the live data backbone around it. Different problems, same protocol.

[Kai Wähner @KaiWaehner]

In manufacturing #ERP selection, bigger is not always better. Three German manufacturers, three different ERPs, three correct choices. Fit beats default.

My new post on the vendor #landscape and how #AgenticAI is reshaping ERP selection:

https://t.co/esfdz2K7ei

[Kai Wähner @KaiWaehner]

#EventDriven systems fail. Events get dropped or block the pipeline without a proper error strategy. A #DeadLetterQueue (DLQ) solves this for any messaging or #Orchestration engine.

DLQ deep dive with #ApacheKafka showing Uber, CrowdStrike, Robinhood:

https://t.co/Twb78iWKIX

[Kai Wähner @KaiWaehner]

#Swiggy runs Learning-to-Rank inside #OpenSearch for autocomplete: sub-10ms #ML inference with no external service. Fast retrieval + model rescoring in one request.

Key principle: put inference where the data is. Not every #AI workload needs a dedicated serving layer.

[Kai Wähner @KaiWaehner]

#CoreWeave acquired #Bufstream, the #ApacheKafka compatible streaming platform, from Buf.

More consolidation in #DataStreaming, which has quietly become core #AI infrastructure. Meanwhile #Buf bets on schemas as the source of truth. Both reshape #EnterpriseArchitecture.

[Kai Wähner @KaiWaehner]

From #BPM to #ProcessIntelligence. It is three things, not one: #ProcessMining, #ProcessOrchestration, and a decision gate that keeps automation and #AgenticAI inside auditable boundaries.

New post: https://t.co/kj7H1dlxCQ

[Kai Wähner @KaiWaehner]

Most AI agents work in the demo and break in production. The reason is usually not the model.

It is the hidden debt around it: runtime, harness, and orchestration.

#AI #AgenticAI #AIAgents

[Kai Wähner @KaiWaehner]

Batch pipelines break #AgenticAI. Stale data means bad decisions.

The answer is #EventDriven architecture.

#ApacheKafka, Kafka Streams, and #ApacheFlink deliver fresh, consistent data the moment it happens.

Read more:
https://t.co/cAMe33Aa6O

[Kai Wähner @KaiWaehner]

#ERP migration is not just a technology decision. It is a process decision.

Process first, platform second.

No big-bang cutover.

Three German manufacturers, three different vendors, same lessons.

https://t.co/a3bDIqsdFQ

#SAP #DataIntegration #AgenticAI #Manufacturing

[Kai Wähner @KaiWaehner]

Knowing when NOT to use #ApacheKafka matters as much as knowing when to use it.

Not hard real-time. Not a proxy for thousands of clients. Not a blockchain. Not a database replacement.

Still as relevant as ever:
https://t.co/qpcvsYoi4h

[Kai Wähner @KaiWaehner]

#Netflix at Data & AI Warsaw Tech Summit 2026: #ApacheIceberg at exabyte scale. 6M tables, 1+ EB, 12K table loads/sec.

They centralized all metadata into one IRC catalog.

#ApacheSpark, #Trino, #ApacheFlink, #Snowflake all hit one endpoint. Thin client, #EventDriven maintenance.

[Kai Wähner @KaiWaehner]

Apache #Kafka 4.3.0 is out.

An operator's release: broker cordoning, retention headroom metrics, share group tuning, tiered storage ISR fix, classic protocol deprecation phase 1.

The 4.x series keeps improving the operational story.

#ApacheKafka #OpenSource

[Kai Wähner @KaiWaehner]

#ApacheCamel = integration framework.

#ApacheKafka = event streaming platform.

When to choose which one?

Full comparison + decision tree:
https://t.co/uoTaeySlZ3

#opensource

[Kai Wähner @KaiWaehner]

#PointOfSale is no longer just a cash register. It is a real time platform for payments, inventory, loyalty, and analytics.

#Square, #SumUp, #Shopify brought it to small merchants. #ApacheKafka and #ApacheFlink power it underneath.

https://t.co/5nR3I9AVcD

[Kai Wähner @KaiWaehner]

Vendor-specific #DataLineage stops at the platform boundary. Partial lineage creates false confidence.

The answer is a platform-independent #DataCatalog owned by none of the platforms it connects.

https://t.co/zoW1iAYO1Z

#DataGovernance #AgenticAI #Datahub #OpenMetadata

[Kai Wähner @KaiWaehner]

Most enterprises ask "which #AI model is best?" Wrong question.

#PredictiveAI answers. #GenerativeAI reasons. #AgenticAI acts.

Each step demands more from the architecture: current data, process boundaries, audit trails.

Not a model problem. An infrastructure problem.

[Kai Wähner @KaiWaehner]

#Siemens uses #DataStreaming with #ApacheKafka and #ApacheFlink to shift processing left in manufacturing and logistics. Faster decisions, better data quality, real-time orchestration. #ShiftLeft in action.

Details + video: https://t.co/XRMja5YGdm

[Kai Wähner @KaiWaehner]

#RequestResponse works at the #API layer. It breaks down for continuous, multi-consumer flows: fan-out, chain failures, polling.

#EventDriven and #EventStreaming handle what synchronous patterns cannot.

Knowing the boundary is key for a good #EnterpriseArchitecture.

[Kai Wähner @KaiWaehner]

#Gartner renamed the #ProcessMining #MagicQuadrant to #ProcessIntelligence.

Good evolution, but #orchestration is still missing.

Vendors are merging mining and orchestration into complete platforms anyway. Does the separation still make sense?

Here is my perspective: