Axoniq

AI pilots are everywhere, but you can't trust 𝘢𝘭𝘭 AI agents with real payments. When an agent takes action on a payment, someone will eventually ask why. To answer, you need a record of what the agent saw at that moment: the data, the rules, and the inputs it acted on. If your system didn’t keep that record, you can’t rebuild it later. On November 12, Timothy L. and Fu Sing Voon from MoneyLion, who run AI agents against real payment operations, join me for a fireside chat moderated by Preben Thorö of Trifork. We'll talk about what separates a working demo from a production system, what happens when an agent gets it wrong, and how teams earn their way toward autonomy. If you're building agents anywhere near core financial systems, I'd love to have you there. 𝗡𝗼𝘃𝗲𝗺𝗯𝗲𝗿 𝟭𝟮, 𝟭𝟰:𝟬𝟬–𝟭𝟱:𝟬𝟬 𝗖𝗘𝗧 / 𝟴:𝟬𝟬–𝟵:𝟬𝟬 𝗔𝗠 𝗘𝗦𝗧 The webinar is free and online, and there will be a recording afterward: https://lnkd.in/euS2w_nD #AIAgents #FinancialServices #Payments #EventSourcing

For 16 years, I've been working on making event sourcing easier to adopt. I just realized I've only been looking at half of the problem. When I started experimenting with CQRS and event sourcing back in 2008, I quickly saw the value it brought. The cost of implementing it, however, was high. Lots of plumbing, lots of new concepts to learn. So I set myself a goal: bring that cost down until the return on investment made sense. Axon Framework was built around that idea, and so was Axoniq. This summer, I gave the keynote at the Event Modeling Conference in Munich, titled "Event Sourcing for the other 95%". While preparing it, something started to bother me. Event sourcing is pretty much the summum of event-driven architecture. For greenfield applications, building on it from day one has proven to be a clear win, time and time again. But most applications aren't greenfield, and most of them don't need the full 100%. For those, 80% is more than enough. Restructuring an existing application around event sourcing is a major undertaking, and that last 20% rarely brings enough value to justify it. Especially when the system has been running for a decade and nobody really dares to touch it anymore. So we asked ourselves a different question. What if we took everything we've learned about what good events look like, combined it with the storage and distribution technology we already have, and applied that to legacy systems? And what if that didn't require you to change a single line of code? Regardless of the language the system was written in, and regardless of whether the people who built it are still around. That sounded too good to be true, which is exactly why we wanted to try it. We ran a few experiments to find out whether it can be done. It can. So that's what we're building right now. Here's how it works. We look at how a system interacts with the outside world: its APIs, its database, the messaging middleware it talks to. We combine that with the internal code flow and, if it's available, the source code. From that, we map out which actions carry which business meaning. When such an action occurs, we emit an event that represents what happened in business terms. Live, as it happens. This allows any legacy system to take part in an event-driven architecture. And unlike Change Data Capture, which tells you that a row in table ORD_HDR was updated, these events tell you that an order was placed. That makes them far more useful for long-term storage and for analysis later on. Should I have thought of this 16 years ago? Maybe. Perhaps I even did, at some point. But it wouldn't have helped much. One of the ingredients that turned out to be absolutely necessary wasn't reliable enough back then: AI. How many systems in your landscape will never be rewritten, but would tell a very interesting story if they could?

Hand a dev agent a fuzzy request and it won't ask what you meant. It fills the gaps with guesses, and ships them as code. 😱 Agents generate fast, but speed only counts if you can verify what came back. The fix isn't a better prompt. It's moving the rigor upstream: write a spec precise enough to check against, then let the agent build to it. Our new guide, "Working with Dev Agents," walks through it: model deliberately, specify precisely, slice cleanly, review formally. Specify first, generate second, review formally. 👇 https://lnkd.in/dqfWrM27/?utm_campaign=BoostPost-EngagedAccounts-WebVisits&utm_source=linkedin&utm_medium=paid&utm_id=1216220776&hsa_acc=516475204&hsa_cam=1216220776&hsa_grp=904865686&hsa_ad=1628660396&hsa_net=linkedin&hsa_ver=3 #EventSourcing #AIagents #SoftwareArchitecture #DevTools #EventModeling

When a server crashes mid-transaction, most systems lose their place. And when a regulator asks why a decision was made, the answer is often already gone. Axoniq workflows turn long-running business processes (payments, fulfillment, onboarding, claims) into durable execution with a permanent, queryable record. ✅ Crash recovery and retries handled automatically ✅ Every step captured as a human-readable business event ✅ Engineers ship business logic instead of failure-handling plumbing – making audits shrink from weeks to hours, and agents get durable context to actually reason over Plus, it's all inside the application you already run, no separate cluster required. Learn how it can help your engineering team. 👉 https://lnkd.in/gaTNWv76/?utm_campaign=BoostPost-EngagedAccounts-WebVisits&utm_source=linkedin&utm_medium=paid&utm_id=1216220776&hsa_acc=516475204&hsa_cam=1216220776&hsa_grp=904865686&hsa_ad=1628563096&hsa_net=linkedin&hsa_ver=3 #AxoniqWorkflows #DurableExecution #WorkflowWednesday

Most agent frameworks give you a clean way to push session events into a vector store and run similarity search, which may be good enough for a lot of use cases. But as Marc Klefter put it: most memory implementations take you far, then miss the sweet spot in the middle. That sweet spot is episodic memory as a first-class artifact, modeled in domain terms by humans, not summarized from technical logs by an LLM after the fact. Which is exactly what event sourcing already produces. Marc walks through the event-sourced memory layer on the Axoniq stack: Axon Server as the canonical event store, projections (analytical queries, application views, context graphs) surfaced to the agent as MCP tools, plus four freshness guarantees (least latency, max age, read-your-writes, snapshot-at) so the agent can reason about how stale its data is instead of guessing. Worth a read if you're building agents into production systems: https://lnkd.in/g66HD3r7 #SoftwareArchitecture #EventSourcing #AxonServer #Axoniq #LLM

Curious how Kafka and Axoniq work together? Kafka moves messages fast, but it doesn't keep them. Most topics roll off after a retention window, by design. Axoniq isn't here to replace that; it's just the other half of the equation. Keep Kafka as your real-time distribution bus. Let Axon Server hold the permanent, replayable history: every business decision, any system state reconstructable from any point in time. The Kafka Extension wires the two together, no duct tape required. It gives your team real-time speed, plus a memory that holds. 🤝 Read how they fit: https://lnkd.in/etq8tseR/?utm_campaign=August-2026-WV-Finance-Tier1-LinkedInEngaged&utm_source=linkedin&utm_medium=paid&utm_id=1188901356&hsa_acc=516475204&hsa_cam=1188901356&hsa_grp=864571366&hsa_ad=1584168486&hsa_net=linkedin&hsa_ver=3 #Kafka #AxonServer #AxoniqFramework #EventStore