There is, however, one idea making big waves at Frontier Labs that should matter to businesses, namely Recursive Self-Improvement (RSI).
The idea is straightforward enough: a system is set up so it can change its own configuration; it then tries something new, measures whether it worked, keeps what helped, discards what failed, and repeats the process again and again, continuously improving itself.
As a physicist by training, the structure feels immediately familiar. It's a closed-loop feedback system: measure the output, compare it to what you wanted, adjust, repeat. Control theory has understood this for decades. What's new is not the principle. It's the speed and autonomy at which AI can now run these loops.
That is why recursive self-improvement is such a hot topic for AI labs. Once a lab can bend AI innovation back onto itself, it might set off a runaway cycle of improvement, leaving all other competitors behind.
But this logic is not limited to AI models. In fact, I believe it should first be applied to businesses.
Coding showed the pattern first
We already saw the first version of this in software development.
Coding is a great domain for AI because it is digital, fast, and testable. An AI system can write code, run it, see whether it compiles, inspect the error, revise the code, and try again. The feedback loop can be closed.
This is why tools like Claude Code and similar systems feel different from earlier productivity tools. They are not just helping someone think about software. They are increasingly participating in the work itself.
The result is the dramatic change we are witnessing right now. Developers who were prepared are moving faster than ever before.
I believe a similar shift is coming for business processes. Not because business processes are as easy as code. To the contrary. Business processes are messier, slower, noisier, political, and constrained by compliance, incentives, and implicit rules. But the pattern still transfers.
The business version of recursive self-improvement
In business processes, RSI does not mean the system becomes superintelligent. Your ERP won’t wake up and decide to autonomously redesign the company.
It means a business can define what it wants a process to achieve. That goal is then translated into a governed improvement loop, supported by AI, agents, process intelligence, benchmarks, and execution tools. The system observes how the process runs, identifies improvement options, changes what can be safely changed, measures the outcome, and uses the result to improve the next cycle. We can already see the first steps in this direction with our Process Consulting Agent (PCA) in SAP Signavio’s Lab Space, where specialized agents help analyze processes, surface bottlenecks, compare performance against company or industry benchmarks, and suggest where to go deeper next.
Today, our PCA turns process questions into actionable insights. Over time, the same pattern can extend from insight to governed action. That could mean adjusting ERP configurations, changing default values, updating workflow steps, improving how certain orders are booked, and so forth.
The output is not a better SaaS app, but an overall improvement in the end-to-end business process.
The important point is that the loop is starting to close, and I am convinced that it will close much faster than people expect. Internally, we already see specialized agents moving from diagnosis to problem resolution in ERP systems. It is still early, and governed change matters enormously, but recursive business improvement no longer feels like a distant dream.
The loop needs a gauge
AI models are not enough, and no matter which version of ChatGPT or Claude we will use, it won’t get us there. The mistake is to treat raw intelligence as the main asset. But intelligence without context, constraints, tools, permissions, and feedback is not an enterprise capability.
Every improvement loop needs a way to know whether something got better. In coding, this is often straightforward. The program compiles, or it does not. Tests pass, or they fail.
In business, the question is harder. A process can be faster but more expensive. Cheaper but worse for customers. More automated but less resilient. More controlled but slower.
That is why RSI for business processes cannot be built on a single metric. It needs process performance indicators (PPIs), business KPIs, process benchmarks, compliance constraints, and human judgment.
A company still needs to decide what it wants to be. Making money is necessary, but it is not a mission. Efficiency matters, but efficiency alone is not a strategy. Leadership still needs to define the trade-offs and, in turn, decide what is worth optimizing.
Process observability becomes strategic infrastructure
This is where SAP Signavio and the innovations we have built over the last few years become central. They now come together into a clearer vision and a more tangible reality: helping companies build the foundation for recursive business improvement.
In many ways, this is the strategy we started pursuing in 2019. We believed that with SAP Signavio, enterprise systems should become so capable and so connected to the business that companies would no longer need to manually translate every business objective into system changes. The ambition was always that the business goal should drive the system, not the other way around.
Process RSI starts with seeing the business clearly. Automating a poorly understood process does not improve the business. It creates faster uncertainty.
You need to know where the bottlenecks are. Where exceptions happen. Where process variants create cost. Where compliance risk appears. Where customer experience suffers. Where the business is already good enough. And where it needs to be genuinely better than everyone else on the market.
That requires process observability.
It also requires coherent definitions of performance across all your systems. You need battle-tested process performance indicators. You need benchmarks. You need the ability to compare yourself against relevant peers. You need to understand the full set of trade-offs before changing how work gets done.
SAP Signavio provides this foundation that RSI can build upon.
The new moat is the proprietary business process loop
AI models will become part of more and more business processes. No business will be differentiated simply by having AI. The same is true for the standardization of agents, agent harnesses, and orchestration patterns. Each tool will commoditize.
What will not commoditize is the company’s proprietary operating reality, which we call its Company Memory. Company Memory matters because recursive improvement needs more than data. It needs memory of what has worked, what is allowed, what should be avoided, and which trade-offs the company is willing to make. Without that memory layer, agents may optimize locally while violating the logic that made the business work in the first place.
This goes beyond the scope of ERPs. ERPs are systems of record, they capture what happened. Orders, invoices, suppliers, payments, materials, postings, approvals, transactions, and so forth. That will remain important.
But the next frontier are systems that close the loop for business processes. This does not happen overnight. Early loops will include manual steps. Humans will approve changes. Governance will matter. Compliance will matter. But the direction is clear, and we have line of sight on the future.
That is why investing in process telemetry, process observability, benchmarks, governed change capabilities, and yes, even good old process models, is not optional preparation. It is the rational move if competitors may close their business feedback loops before you do.
Leaders should care because we have already seen a dramatic step change in engineering. Teams that were ready are pulling ahead. Teams that were not are now trying to understand why their operating rhythm feels slow.
Software development is important, but business processes are where economic reality manifests. If recursive improvement reaches this layer, the impact is not another productivity feature. It is a different tempo of business transformation.
Business process transformation becomes the strategy layer
For SAP Signavio, this extends the familiar mission of business transformation management into the AI era. We have always helped companies understand and improve how work gets done. In the age of AI agents and recursive improvement loops, that mission becomes even more important.
The future will not be won by companies that merely attach agents or AI to existing systems or processes. It will be won by companies that close and own their business feedback loops.
Recursive self-improvement is the Silicon Valley term. Business process RSI is the enterprise opportunity, and SAP Signavio is the partner to get there.
With thanks to Gero Decker and Lukas N.P. Egger for their input, review, and the many discussions that helped sharpen this piece.