Artificial Intelligence is being positioned as the next architectural leap for enterprise platforms, but not much has had a significant impact on ETRM and TRMS. Until recently.
Change Happens Slowly, Until It Happens Fast
It’s been loosely quoted that a lot of major change happens two ways: “gradually and then suddenly.” Something has shifted. No matter what technology you have now — a legacy monolith, spreadsheets, or a semi-‘new tech’ platform still in progress — your next technology will be built by AI. Not mid-process, not theoretically, but as a shift in the day-to-day reality of how AI tools can serve financial and energy trading firms. This paper is our attempt to put ‘where we are now’ into writing, mostly because things move quickly and build on the present foundation.
There have been a number of articles and interviews going around, but it’s worth re-posting one shared by a colleague in the space, as it’s written by Matt Shumer and he speaks candidly about how significant the change has been and how immediately usable the technology is. We could not agree more — this IS the leap forward people have been anticipating. Not just Copilot taking notes and ChatGPT cleaning them up (or hallucinating), but very good code being written by a tool you previously would not have trusted with more than a line or two. This has led others to boldly predict that the price of software is going to zero. To be clear, we are not the type of firm to make bold predictions, especially as the tool’s success will depend on many variables, but it is a good indication of where things are headed and how companies will provide value to their clients.
Does What It Says On The Tin
We will go deeper on some of these subjects in subsequent papers. The below list are the capabilities we are actively deploying right now — not piloting, not evaluating, but deploying:
- Agentic LLM and Automation. AI agents that can catch changes, describe them, catalogue them and make them accessible with minimal human intervention. In our space, this means a lot as monolithic systems lend themselves to customizations but rarely to keeping humans interested in creating, updating, and storing documentation in an effective manner.
- Coding Acceleration. The speed at which we can build and iterate on software has fundamentally changed. What used to take a senior developer several days now takes hours. Working without supervision. This is not about replacing developers, it’s about dramatically compressing the distance between “here’s what we need” and “here it is, working.”
- Test Automation and GUI Development. One of the most underappreciated gains. AI can now write comprehensive test suites, run them, interpret results, and feed fixes back into the codebase — autonomously. UI work becomes a matter of giving instructions and following up with changes. Much of the work is done while you are away from your desk. For both of these, we are talking faster releases, higher confidence, less manual QA burden.
Especially in our space, all of these need to be led by someone with considerable expertise. It is still a big ‘human’ ask. And while ‘vibe coding’ sounds very entertaining, it is not a valid strategy for vendor-provided systems that solve real business problems. Replace your internal CRM? Sure, why not?! But to serve a client as a vendor — running a real business — you not only need to deliver an enjoyable, compelling product, it must have real knowledge behind it.
Pounding At The Monolith Door
When you look at traditional monolith platforms, one thing that used to be a barrier was the ability to write code that would apply to their internal, custom systems. That is no longer the case. Give it a shot — drop your desired script description into Claude Code (we have not seen the same level of success with others) and see what it comes back with. This may not be as helpful for large implementations that are already up and running and only need minor code tweaks for upgrades. Small bits of code written by people with a lot of experience, after a lot of thinking and strategizing — that is not where AI will make much difference, or even be usable.
Where it can and will have an impact is when you decide you need something in-house, whether from another vendor or from your own team. Building a tool to work seamlessly with your platform is just not the same problem it used to be. As monoliths inevitably age out, you and your vendors now have powerful tools to make that happen faster. It’s still not clear what the final landscape will look like, but it’s likely to be some form of the ‘strangler’ approach — gradually removing your system of record that has, perhaps over-aggressively, become the ‘thing that is supposed to do everything.’
All of this has become ‘doable’ and ‘usable,’ which still means significant human interaction — much of which is not likely to go away entirely. The question is what will get better. We do not know exactly where in the AI toolchain the next improvements will land, but we know they will. This is not the final state, and it’s already a game-changer.
Time To Make A Plan
As a vendor, it’s critical that we stay at the forefront of our space. Tinkering with AI in its early stages — seeing what works, learning what does not — is part of what makes our firm good at what we do. Said another way: we have to. When those tools start to show real promise, we are that much better positioned to serve our clients and to ensure that what we deliver is not just a fun side-project, but something that works in a critical trading environment.
At this point, we are a step ahead. Curious to learn more? Please reach out.
You need a plan too. As we often tell clients — and it always surprises them to hear a vendor say this — autonomy matters. Chat with us. Chat with your vendors: how much of their code is AI-generated? If the answer is very little, ask what you are paying for. Talk to other service providers. Try the AI tools yourself. If you have not started yet, now is the time.
The papers that follow will go deeper: specific capabilities, real implementations, and an honest look at where the limits still are. But we wanted to start here, with the simple case for urgency.